# BYOAIK: Bring Your Own AI Key > BYOAIK is an independent directory of 257 AI tools you run with your own > API key, a local model, or a self-hosted setup. Every listing documents which AI providers > it supports (OpenAI, Anthropic, Google Gemini, OpenRouter, Ollama, and more), how it stores > and transmits your API key, its pricing model, license, and a 0-100 trust score. "No AI > markup" means you pay the model provider directly instead of a reseller. Useful facts for agents: - Every tool has a detail page at https://www.byoaik.com/tools// with full key-handling details. - The full dataset is available as JSON (no auth, CORS-open): GET https://qhfc1deef2.execute-api.us-east-1.amazonaws.com/tools returns all published tools; GET https://qhfc1deef2.execute-api.us-east-1.amazonaws.com/tools/ returns one tool with fields like providers_supported, key_storage_type, key_transmission_type, risk_level, pricing_model, open_source, platforms, and trust_score. - llms-full.txt (https://www.byoaik.com/llms-full.txt) contains the complete directory with long descriptions and key-handling notes for every tool. - Verification is tiered and evidence-derived, not a single badge. Each listing's verification_status is COMPUTED from its evidence[] array (what was checked, by which method, with what confidence, and a source link), so a tool cannot claim a tier its evidence does not support. Tiers weakest to strongest: UNVERIFIED, RESEARCH_ASSISTED (AI-assisted reading of public docs, not independently confirmed), COMMUNITY_REPORTED, VENDOR_CLAIMED, DOCS_VERIFIED, SOURCE_VERIFIED (BYOAIK's scanner found it in public source: static analysis, not a security audit), RUNTIME_VERIFIED, BYOAIK_VERIFIED. Current counts: RESEARCH_ASSISTED 181, SOURCE_VERIFIED 74, DOCS_VERIFIED 2. ## Pages - [All tools](https://www.byoaik.com/tools/): the full directory, filterable by provider, category, pricing, and risk - [Providers](https://www.byoaik.com/providers/): browse by whose API key you bring - [Collections](https://www.byoaik.com/collections/): curated lists - [Methodology](https://www.byoaik.com/methodology/): how trust scores and key-handling reviews work - [Submit](https://www.byoaik.com/submit/): submit a tool for review ## Providers - [OpenAI](https://www.byoaik.com/providers/openai/): 208 tools accept your OpenAI key - [Anthropic](https://www.byoaik.com/providers/anthropic/): 180 tools accept your Anthropic key - [Google Gemini](https://www.byoaik.com/providers/gemini/): 153 tools accept your Google Gemini key - [OpenRouter](https://www.byoaik.com/providers/openrouter/): 110 tools accept your OpenRouter key - [Ollama](https://www.byoaik.com/providers/ollama/): 211 tools accept your Ollama key - [Groq](https://www.byoaik.com/providers/groq/): 77 tools accept your Groq key - [Mistral](https://www.byoaik.com/providers/mistral/): 69 tools accept your Mistral key - [DeepSeek](https://www.byoaik.com/providers/deepseek/): 67 tools accept your DeepSeek key - [Azure OpenAI](https://www.byoaik.com/providers/azure-openai/): 93 tools accept your Azure OpenAI key - [AWS Bedrock](https://www.byoaik.com/providers/bedrock/): 9 tools accept your AWS Bedrock key - [Cohere](https://www.byoaik.com/providers/cohere/): 26 tools accept your Cohere key - [Together AI](https://www.byoaik.com/providers/together/): 28 tools accept your Together AI key - [LM Studio](https://www.byoaik.com/providers/lm-studio/): 5 tools accept your LM Studio key - [xAI Grok](https://www.byoaik.com/providers/xai-grok/): 41 tools accept your xAI Grok key - [Perplexity](https://www.byoaik.com/providers/perplexity/): 25 tools accept your Perplexity key - [Hugging Face](https://www.byoaik.com/providers/hugging-face/): 35 tools accept your Hugging Face key - [Fireworks](https://www.byoaik.com/providers/fireworks/): 18 tools accept your Fireworks key - [Custom / OpenAI-compatible](https://www.byoaik.com/providers/openai-compatible/): 226 tools accept your Custom / OpenAI-compatible key ## Collections - [Best Open-Source ChatGPT Alternatives](https://www.byoaik.com/collections/open-source-chatgpt-alternatives/): Self-hostable chat UIs that run on your own provider keys. - [BYOK Coding Agents](https://www.byoaik.com/collections/byok-coding-agents/): AI coding assistants where you pay your model provider directly. - [Self-Hosted RAG & Research](https://www.byoaik.com/collections/self-hosted-rag-research/): Chat with your documents and the web using models you control. - [Terminal & CLI AI Tools](https://www.byoaik.com/collections/terminal-cli-ai-tools/): Bring-your-own-key AI that lives in your terminal. - [Local-First & Private AI](https://www.byoaik.com/collections/local-first-private-ai/): Keep your keys and data on your own machine. ## Tools ### Khoj - Page: https://www.byoaik.com/tools/khoj/ - Website: https://khoj.dev/ - Source: https://github.com/khoj-ai/khoj - Category: Knowledge Management | Pricing: OPEN_SOURCE | Open source: yes (AGPL-3.0) | Trust score: 94/100 - Platforms: Web, Desktop, Mobile, Self-hosted, Docker - Bring-your-own-key providers: OpenAI, Anthropic, Google Gemini, DeepSeek, OpenRouter, Hugging Face, Ollama, Custom / OpenAI-compatible - Key storage: LOCAL_DEVICE | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Slowing down (last commit about 3 months ago) Khoj is an open-source (AGPL-3.0), self-hostable personal AI application that lets you chat with LLMs grounded in your own documents and online sources, and build custom agents. It is accessible via web, desktop, mobile, an Obsidian plugin, an Emacs package, and WhatsApp. Self-hosting users bring their own provider API keys: the admin panel and environment variables accept OPENAI_API_KEY, ANTHROPIC_API_KEY, and GEMINI_API_KEY, plus a configurable OpenAI-compatible API base URL that points Khoj at Ollama, LMStudio, LiteLLM, vLLM, Hugging Face, OpenRouter, and any other OpenAI-compatible endpoint. A managed cloud version is also offered. Key handling: Your API key is stored locally on your device. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 95/100 (assessable for any licence), verifiability 90/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 23/25: The key is held on your own device. - Request Routing 20/20: Requests go straight from you to the AI provider. The provider endpoint is configurable, so you can point it at the provider or your own gateway. - Transparency 20/20: Source is public under AGPL-3.0, so anyone can check how the key is handled. Key handling was located in the published source. - Privacy 14/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 7/10: Key handling was found in the published source by BYOAIK's scanner. What was checked (verification tier SOURCE_VERIFIED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [STRONG via SOURCE_SCAN] The user supplies their own OpenAI key: the project references its API key variable. (https://github.com/khoj-ai/khoj/blob/HEAD/docker-compose.yml) - [STRONG via SOURCE_SCAN] The user supplies their own Anthropic key: the project references its API key variable. (https://github.com/khoj-ai/khoj/blob/HEAD/docker-compose.yml) - [STRONG via SOURCE_SCAN] The user supplies their own Google Gemini key: the project references its API key variable. (https://github.com/khoj-ai/khoj/blob/HEAD/docker-compose.yml) - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/khoj-ai/khoj/blob/HEAD/docker-compose.yml) - [STRONG via SOURCE_SCAN] The provider endpoint is configurable, so requests can be pointed straight at the provider or at your own gateway. (https://github.com/khoj-ai/khoj/blob/HEAD/docker-compose.yml) - [CONFIRMED via SOURCE_SCAN] Ships a container definition, so it can be self-hosted on your own infrastructure. (https://github.com/khoj-ai/khoj/blob/HEAD/docker-compose.yml) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-08-02, about 2 month(s) ago. (https://github.com/khoj-ai/khoj/commits) ### OpenWhispr - Page: https://www.byoaik.com/tools/openwhispr/ - Website: https://openwhispr.com/ - Source: https://github.com/OpenWhispr/openwhispr - Category: Voice and Transcription Tools | Pricing: OPEN_SOURCE | Open source: yes (MIT) | Trust score: 94/100 - Platforms: Desktop, Self-hosted, Docker - Bring-your-own-key providers: OpenAI, Anthropic, Google Gemini, Groq, Mistral, Ollama, Custom / OpenAI-compatible - Key storage: LOCAL_DEVICE | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit yesterday) OpenWhispr is an open-source, cross-platform (macOS, Windows, Linux) voice-to-text dictation desktop app. You press a hotkey, speak, and the transcribed text appears at your cursor, with optional AI post-processing/agent reasoning, meeting transcription, and speaker diarization. It can run fully offline using local speech-to-text engines (OpenAI Whisper via whisper.cpp, NVIDIA Parakeet via sherpa-onnx) and local LLMs via llama.cpp, so audio never leaves the device. Alternatively, it does cloud processing on a bring-your-own-key basis: users enter their own provider API keys for OpenAI, Anthropic Claude, Google Gemini, Groq, and Mistral (plus AssemblyAI/Deepgram for transcription). Keys are encrypted at rest in the OS keychain (Keychain/DPAPI/libsecret) via Electron safeStorage. It explicitly supports "any OpenAI-compatible API," including Ollama, LocalAI, and self-hosted Whisper servers, allowing custom base URLs/endpoints. Licensed MIT and free forever with local models or your own keys; a paid Pro/Business/Enterprise tier offers zero-setup managed cloud and extra features. Maintained by Gizmo Labs Inc. Key handling: Your API key is stored locally on your device. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 95/100 (assessable for any licence), verifiability 90/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 25/25: The key is held on your own device. Observed using the operating system keychain rather than a file. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under MIT, so anyone can check how the key is handled. Key handling was located in the published source. - Privacy 15/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. The project states it sends no telemetry. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 7/10: Key handling was found in the published source by BYOAIK's scanner. What was checked (verification tier SOURCE_VERIFIED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/OpenWhispr/openwhispr/blob/HEAD/README.md) - [REPORTED via SOURCE_SCAN] The project states it sends no telemetry. (https://github.com/OpenWhispr/openwhispr/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] The user supplies their own OpenAI key: the project references its API key variable. (https://github.com/OpenWhispr/openwhispr/blob/HEAD/.env.example) - [STRONG via SOURCE_SCAN] The user supplies their own Anthropic key: the project references its API key variable. (https://github.com/OpenWhispr/openwhispr/blob/HEAD/.env.example) - [STRONG via SOURCE_SCAN] The user supplies their own Google Gemini key: the project references its API key variable. (https://github.com/OpenWhispr/openwhispr/blob/HEAD/.env.example) - [STRONG via SOURCE_SCAN] The user supplies their own Groq key: the project references its API key variable. (https://github.com/OpenWhispr/openwhispr/blob/HEAD/.env.example) - [STRONG via SOURCE_SCAN] The user supplies their own Mistral key: the project references its API key variable. (https://github.com/OpenWhispr/openwhispr/blob/HEAD/.env.example) - [STRONG via SOURCE_SCAN] Uses the operating system keychain / credential store to hold the key. (https://github.com/OpenWhispr/openwhispr/blob/HEAD/package.json) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-29, about 0 month(s) ago. (https://github.com/OpenWhispr/openwhispr/commits) ### Agentic Ledger - Page: https://www.byoaik.com/tools/agentic-ledger/ - Website: https://agentic-ledger.dev - Source: https://github.com/ShekharBhardwaj/AgenticLedger - Category: Developer Infrastructure | Pricing: OPEN_SOURCE | Open source: yes (MIT) | Trust score: 93/100 - Platforms: CLI, Self-hosted, Docker, Web - Bring-your-own-key providers: Anthropic, OpenAI, OpenRouter, Custom / OpenAI-compatible, LM Studio - Key storage: ENVIRONMENT_VARIABLE | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Editorial badge: Most Innovative: The only tool here that makes an agent's spending legible and enforceable: a transparent proxy that captures every call and cost with no code changes, then puts hard dollar budgets in the request path where the agent itself cannot lift them. - Disclosure: Agentic Ledger is built by the maintainer of BYOAIK. It is listed on the same published criteria as every other tool, and its trust score is unadjusted. - Maintenance: Actively developed (last commit 8 days ago) - Developer claimed: ownership proven 2026-08-10; key handling below is the developer's stated architecture Agentic Ledger is an MIT-licensed, local-first observability proxy for AI agents. You point your agent's base URL at it (one env var: ANTHROPIC_BASE_URL or OPENAI_BASE_URL) and every LLM call is captured with full prompts, tool calls, and cache-aware cost accounting, with no code changes to the agent. It is explicitly bring-your-own-key by construction: the proxy passes your existing provider credentials straight through and never stores them, and keys it does need (for optional replay) can be supplied via file using the Docker-secrets _FILE pattern to keep them out of shell history. Team cards let one real provider key be shared as revocable per-team allowances with hard dollar budgets enforced in the request path. Data lives in a single SQLite file on your machine with no account, no telemetry, and no cloud component. Distributed via pip and a multi-arch Docker image, with a CLI and a local web dashboard; supports Anthropic and OpenAI wire formats, OpenAI-compatible gateways such as LiteLLM and OpenRouter, LM Studio for local replay, plus OTel ingest from Gemini CLI, Codex CLI, AutoGen and Pydantic AI. Notable features: - Transparent proxy: one env var, zero code changes, works with any framework or provider - Cache-aware cost accounting, including the prompt-cache traffic most tools miss - Loop Lens for overnight agent runs: cost per iteration, stuck-loop detection, circuit breaker - Replay a whole run against a different model and get a match report card - Cost what-if: reprice a captured run on another model with no API calls or spend - Hard budgets enforced before the call reaches the provider, per session, agent, user, and day - Kill switch that survives proxy restarts and sits outside the agent's reach - Team cards: one real provider key, revocable per-team allowances with their own budgets - Flow and Trace views built from real parent links, plus OTel ingest - MCP server that returns compact summaries instead of dumping your context window Key handling: Your API key is supplied via an environment variable or local config file. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 93/100 (assessable for any licence), verifiability 90/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 22/25: The key is supplied by an environment variable or local config file you control. - Request Routing 20/20: Requests go straight from you to the AI provider. The provider endpoint is configurable, so you can point it at the provider or your own gateway. - Transparency 20/20: Source is public under MIT, so anyone can check how the key is handled. Key handling was located in the published source. - Privacy 14/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 7/10: Key handling was found in the published source by BYOAIK's scanner. What was checked (verification tier SOURCE_VERIFIED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [STRONG via SOURCE_SCAN] The user supplies their own OpenAI key: the project references its API key variable. (https://github.com/ShekharBhardwaj/AgenticLedger/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] The user supplies their own Anthropic key: the project references its API key variable. (https://github.com/ShekharBhardwaj/AgenticLedger/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] Reads the API key from an environment variable at runtime. (https://github.com/ShekharBhardwaj/AgenticLedger/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/ShekharBhardwaj/AgenticLedger/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] The provider endpoint is configurable, so requests can be pointed straight at the provider or at your own gateway. (https://github.com/ShekharBhardwaj/AgenticLedger/blob/HEAD/README.md) - [CONFIRMED via SOURCE_SCAN] Ships a container definition, so it can be self-hosted on your own infrastructure. (https://github.com/ShekharBhardwaj/AgenticLedger/blob/HEAD/docker-compose.yml) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-29, about 0 month(s) ago. (https://github.com/ShekharBhardwaj/AgenticLedger/commits) ### manga-image-translator - Page: https://www.byoaik.com/tools/manga-image-translator/ - Website: https://github.com/zyddnys/manga-image-translator - Source: https://github.com/zyddnys/manga-image-translator - Category: Productivity Tools | Pricing: OPEN_SOURCE | Open source: yes (GPL-3.0) | Trust score: 93/100 - Platforms: Self-hosted, Web, CLI, Library/SDK, Docker - Bring-your-own-key providers: OpenAI, Google Gemini, Groq, DeepSeek, Together AI, Ollama, Custom / OpenAI-compatible - Key storage: ENVIRONMENT_VARIABLE | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit 28 days ago) A self-hostable Python tool that detects text in manga/comic/images via OCR, removes it through inpainting, machine-translates it, and re-renders the translated text back into the image. It ships a web server (FastAPI backend + Vue front), a CLI, and Docker/docker-compose images, plus optional GPU acceleration. Translation can route through many backends: LLM providers configured with the end user's own API keys (OpenAI/ChatGPT, Google Gemini, Groq, DeepSeek, Together AI, and any OpenAI-compatible endpoint including local Ollama), conventional MT services (DeepL, Baidu, Youdao, Caiyun, Papago), the Sakura LLM, and fully offline models (NLLB, Sugoi, M2M100). Keys are supplied via a .env file or Docker environment variables (OPENAI_API_KEY, DEEPSEEK_API_KEY, GROQ_API_KEY, GEMINI_API_KEY, TOGETHER_API_KEY, CUSTOM_OPENAI_API_BASE/CUSTOM_OPENAI_API_KEY, etc.), confirmed directly in manga_translator/translators/keys.py and examples/Example.env. Note: the candidate's claimed license (MIT) is incorrect, the repo is GPL-3.0, and star count is ~10k, not 6k. Key handling: Your API key is supplied via an environment variable or local config file. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 93/100 (assessable for any licence), verifiability 90/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 22/25: The key is supplied by an environment variable or local config file you control. - Request Routing 20/20: Requests go straight from you to the AI provider. The provider endpoint is configurable, so you can point it at the provider or your own gateway. - Transparency 20/20: Source is public under GPL-3.0, so anyone can check how the key is handled. Key handling was located in the published source. - Privacy 14/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 7/10: Key handling was found in the published source by BYOAIK's scanner. What was checked (verification tier SOURCE_VERIFIED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [STRONG via SOURCE_SCAN] The user supplies their own OpenAI key: the project references its API key variable. (https://github.com/zyddnys/manga-image-translator/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] The user supplies their own Google Gemini key: the project references its API key variable. (https://github.com/zyddnys/manga-image-translator/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] The user supplies their own Groq key: the project references its API key variable. (https://github.com/zyddnys/manga-image-translator/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] The user supplies their own DeepSeek key: the project references its API key variable. (https://github.com/zyddnys/manga-image-translator/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/zyddnys/manga-image-translator/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] The provider endpoint is configurable, so requests can be pointed straight at the provider or at your own gateway. (https://github.com/zyddnys/manga-image-translator/blob/HEAD/README.md) - [CONFIRMED via SOURCE_SCAN] Ships a container definition, so it can be self-hosted on your own infrastructure. (https://github.com/zyddnys/manga-image-translator/blob/HEAD/docker-compose.yml) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-25, about 0 month(s) ago. (https://github.com/zyddnys/manga-image-translator/commits) ### DeerFlow - Page: https://www.byoaik.com/tools/deerflow/ - Website: https://github.com/bytedance/deer-flow - Source: https://github.com/bytedance/deer-flow - Category: Agent Frameworks | Pricing: OPEN_SOURCE | Open source: yes (MIT) | Trust score: 93/100 - Platforms: Web, CLI, Self-hosted, Docker - Bring-your-own-key providers: OpenAI, OpenRouter, Ollama, Custom / OpenAI-compatible - Key storage: ENVIRONMENT_VARIABLE | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit today) DeerFlow is ByteDance's open-source long-horizon SuperAgent framework that orchestrates sub-agents, search, memory, tools, and sandboxes to handle deep-research, code-generation, and content tasks spanning minutes to hours. It started as a deep-research framework and has expanded into a general multi-agent execution engine. Models are defined in conf.yaml/config.yaml using LangChain-style providers (e.g. langchain_openai:ChatOpenAI) with api_key referencing environment variables such as $OPENAI_API_KEY. The end user supplies their own provider credentials: keys are set in a local .env file (recommended) or exported in the shell, and there is no centralized key management. It is model-agnostic and works with any OpenAI-compatible API, including OpenAI, OpenRouter gateways (with custom base URLs), and local backends like vLLM and Ollama. It ships a web UI (localhost:2026) plus console/CLI usage and is self-hosted via local dev or Docker. Licensed MIT, ~72k GitHub stars, actively maintained by ByteDance (last push June 2026, not archived). Key handling: Your API key is supplied via an environment variable or local config file. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 93/100 (assessable for any licence), verifiability 90/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 22/25: The key is supplied by an environment variable or local config file you control. - Request Routing 20/20: Requests go straight from you to the AI provider. The provider endpoint is configurable, so you can point it at the provider or your own gateway. - Transparency 20/20: Source is public under MIT, so anyone can check how the key is handled. Key handling was located in the published source. - Privacy 14/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 7/10: Key handling was found in the published source by BYOAIK's scanner. What was checked (verification tier SOURCE_VERIFIED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [STRONG via SOURCE_SCAN] The user supplies their own OpenAI key: the project references its API key variable. (https://github.com/bytedance/deer-flow/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] The user supplies their own OpenRouter key: the project references its API key variable. (https://github.com/bytedance/deer-flow/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/bytedance/deer-flow/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] The provider endpoint is configurable, so requests can be pointed straight at the provider or at your own gateway. (https://github.com/bytedance/deer-flow/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] The user supplies their own Anthropic key: the project references its API key variable. (https://github.com/bytedance/deer-flow/blob/HEAD/.env.example) - [STRONG via SOURCE_SCAN] The user supplies their own Google Gemini key: the project references its API key variable. (https://github.com/bytedance/deer-flow/blob/HEAD/.env.example) - [STRONG via SOURCE_SCAN] The user supplies their own DeepSeek key: the project references its API key variable. (https://github.com/bytedance/deer-flow/blob/HEAD/.env.example) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-29, about 0 month(s) ago. (https://github.com/bytedance/deer-flow/commits) ### PrivateGPT - Page: https://www.byoaik.com/tools/privategpt/ - Website: https://github.com/zylon-ai/private-gpt - Source: https://github.com/zylon-ai/private-gpt - Category: Document Analysis | Pricing: OPEN_SOURCE | Open source: yes (Apache-2.0) | Trust score: 93/100 - Platforms: Self-hosted, Web, Library/SDK, CLI, Docker - Bring-your-own-key providers: OpenAI, Azure OpenAI, Google Gemini, Ollama, Custom / OpenAI-compatible - Key storage: ENVIRONMENT_VARIABLE | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit 10 days ago) PrivateGPT is an Apache-2.0, source-available project (by Zylon) for ingesting documents and asking questions over them via RAG, with a focus on privacy and 100% offline operation. It exposes both a UI and an OpenAI-compatible API. Beyond fully local inference (Ollama, llama.cpp, vLLM, LocalAI), it supports bring-your-own-key configuration through multiple LLM modes selected in settings.yaml / env vars: `openai` (OpenAI key), `openailike` (any OpenAI-compatible endpoint with configurable api_base + api_key, e.g. LM Studio, vLLM, remote proxies), `azopenai` (Azure OpenAI endpoint + key), `gemini` (Google Gemini API key), and `sagemaker`. Users set their own api_key and api_base, so any remote OpenAI-compatible provider can be plugged in. Very popular (~57k stars) and actively maintained (v1.0.1 released June 2026). Key handling: Your API key is supplied via an environment variable or local config file. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 93/100 (assessable for any licence), verifiability 90/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 22/25: The key is supplied by an environment variable or local config file you control. - Request Routing 20/20: Requests go straight from you to the AI provider. The provider endpoint is configurable, so you can point it at the provider or your own gateway. - Transparency 20/20: Source is public under Apache-2.0, so anyone can check how the key is handled. - Privacy 14/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 7/10: Key handling was found in the published source by BYOAIK's scanner. What was checked (verification tier SOURCE_VERIFIED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/zylon-ai/private-gpt/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] The provider endpoint is configurable, so requests can be pointed straight at the provider or at your own gateway. (https://github.com/zylon-ai/private-gpt/blob/HEAD/README.md) - [CONFIRMED via SOURCE_SCAN] Ships a container definition, so it can be self-hosted on your own infrastructure. (https://github.com/zylon-ai/private-gpt/blob/HEAD/Dockerfile) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-21, about 0 month(s) ago. (https://github.com/zylon-ai/private-gpt/commits) ### GPT Researcher - Page: https://www.byoaik.com/tools/gpt-researcher/ - Website: https://gptr.dev - Source: https://github.com/assafelovic/gpt-researcher - Category: Research Tools | Pricing: OPEN_SOURCE | Open source: yes (Apache-2.0) | Trust score: 93/100 - Platforms: Web, Self-hosted, CLI, Library/SDK, Docker - Bring-your-own-key providers: OpenAI, Anthropic, Google Gemini, Azure OpenAI, Groq, Mistral, Cohere, Ollama, Hugging Face, Together AI, Fireworks, Custom / OpenAI-compatible - Key storage: ENVIRONMENT_VARIABLE | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit today) GPT Researcher is an open-source autonomous research agent that plans queries, scrapes and aggregates many web (and local) sources, and produces detailed, citation-backed reports. It is provider-agnostic: users bring their own LLM keys via environment variables or a .env file and select providers with provider:model strings across FAST_LLM/SMART_LLM/STRATEGIC_LLM. OpenAI-compatible and local endpoints are supported through OPENAI_BASE_URL/OPENAI_API_BASE (used for Ollama and other self-hosted models). It is self-hostable via pip, Docker, or its FastAPI server, with a web UI, CLI, and Python SDK usage. Web search uses a configurable retriever (Tavily by default, with its own BYO key). Apache-2.0 licensed, ~27.8k stars, and actively maintained (v3.5.0, May 2026). Key handling: Your API key is supplied via an environment variable or local config file. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 93/100 (assessable for any licence), verifiability 90/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 22/25: The key is supplied by an environment variable or local config file you control. - Request Routing 20/20: Requests go straight from you to the AI provider. The provider endpoint is configurable, so you can point it at the provider or your own gateway. - Transparency 20/20: Source is public under Apache-2.0, so anyone can check how the key is handled. Key handling was located in the published source. - Privacy 14/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 7/10: Key handling was found in the published source by BYOAIK's scanner. What was checked (verification tier SOURCE_VERIFIED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [STRONG via SOURCE_SCAN] The user supplies their own OpenAI key: the project references its API key variable. (https://github.com/assafelovic/gpt-researcher/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] The provider endpoint is configurable, so requests can be pointed straight at the provider or at your own gateway. (https://github.com/assafelovic/gpt-researcher/blob/HEAD/README.md) - [CONFIRMED via SOURCE_SCAN] Ships a container definition, so it can be self-hosted on your own infrastructure. (https://github.com/assafelovic/gpt-researcher/blob/HEAD/docker-compose.yml) - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/assafelovic/gpt-researcher/blob/HEAD/pyproject.toml) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-26, about 0 month(s) ago. (https://github.com/assafelovic/gpt-researcher/commits) ### ShellGPT (sgpt) - Page: https://www.byoaik.com/tools/shellgpt-sgpt/ - Website: https://github.com/TheR1D/shell_gpt - Source: https://github.com/TheR1D/shell_gpt - Category: Developer Infrastructure | Pricing: OPEN_SOURCE | Open source: yes (MIT) | Trust score: 93/100 - Platforms: CLI, Self-hosted, Docker - Bring-your-own-key providers: OpenAI, Ollama, Custom / OpenAI-compatible, Anthropic, Google Gemini, Azure OpenAI, Groq, Mistral, Cohere - Key storage: ENVIRONMENT_VARIABLE | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Slowing down (last commit about 4 months ago) ShellGPT (sgpt) is an open-source command-line productivity tool that uses large language models to generate shell commands, code snippets, and answers directly in the terminal, with shell integration for Bash, Zsh, and PowerShell on Linux, macOS, and Windows. It is bring-your-own-key by design: on first run it prompts for your own API key and stores it in ~/.config/shell_gpt/.sgptrc, and also reads the OPENAI_API_KEY environment variable. Beyond OpenAI, it supports a configurable API_BASE_URL for any OpenAI-compatible endpoint, and installing the litellm extra (pip install "shell-gpt[litellm]") with USE_LITELLM=true routes requests through LiteLLM to many providers and to local models via Ollama (e.g. DEFAULT_MODEL=ollama/..., API_BASE_URL=http://localhost:11434). No hosted backend or vendor account is required. Key handling: Your API key is supplied via an environment variable or local config file. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 93/100 (assessable for any licence), verifiability 90/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 22/25: The key is supplied by an environment variable or local config file you control. - Request Routing 20/20: Requests go straight from you to the AI provider. The provider endpoint is configurable, so you can point it at the provider or your own gateway. - Transparency 20/20: Source is public under MIT, so anyone can check how the key is handled. Key handling was located in the published source. - Privacy 14/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 7/10: Key handling was found in the published source by BYOAIK's scanner. What was checked (verification tier SOURCE_VERIFIED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [STRONG via SOURCE_SCAN] The user supplies their own OpenAI key: the project references its API key variable. (https://github.com/TheR1D/shell_gpt/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/TheR1D/shell_gpt/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] The provider endpoint is configurable, so requests can be pointed straight at the provider or at your own gateway. (https://github.com/TheR1D/shell_gpt/blob/HEAD/README.md) - [CONFIRMED via SOURCE_SCAN] Ships a container definition, so it can be self-hosted on your own infrastructure. (https://github.com/TheR1D/shell_gpt/blob/HEAD/Dockerfile) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-07-02, about 3 month(s) ago. (https://github.com/TheR1D/shell_gpt/commits) ### bilingual_book_maker - Page: https://www.byoaik.com/tools/bilingual-book-maker/ - Website: https://github.com/yihong0618/bilingual_book_maker - Source: https://github.com/yihong0618/bilingual_book_maker - Category: Productivity Tools | Pricing: OPEN_SOURCE | Open source: yes (MIT) | Trust score: 93/100 - Platforms: CLI, Self-hosted, Docker - Bring-your-own-key providers: OpenAI, Anthropic, Google Gemini, Azure OpenAI, Groq, xAI Grok, DeepSeek, Ollama, Custom / OpenAI-compatible - Key storage: ENVIRONMENT_VARIABLE | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit 10 days ago) An open-source Python command-line tool that converts EPUB, TXT, MD, SRT, and PDF files into bilingual (side-by-side) editions by translating them with an LLM or translation engine of your choice. The end user supplies their own API key per provider via dedicated CLI flags and BBM_* environment variables: --openai_key/BBM_OPENAI_API_KEY for OpenAI, --claude_key/BBM_CLAUDE_API_KEY for Anthropic Claude, --gemini_key/BBM_GOOGLE_GEMINI_KEY for Google Gemini, --groq_key/BBM_GROQ_API_KEY for Groq, --xai_key/BBM_XAI_API_KEY for xAI Grok, --deepl_key/BBM_DEEPL_API_KEY for DeepL, --qwen_key for Qwen, plus --api_key combined with --provider for any custom OpenAI-compatible endpoint. Ollama is supported for fully local/self-hosted models, and free engines (Google Translate, DeepL Free) are also available. Installed via pip (pip install -U bbook_maker) and run as the bbook CLI or python3 make_book.py. MIT-licensed, ~9.3k GitHub stars, actively maintained (latest release v1.1.0, July 2025), Python 3.10+. Key handling: Your API key is supplied via an environment variable or local config file. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 93/100 (assessable for any licence), verifiability 90/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 22/25: The key is supplied by an environment variable or local config file you control. - Request Routing 20/20: Requests go straight from you to the AI provider. The provider endpoint is configurable, so you can point it at the provider or your own gateway. - Transparency 20/20: Source is public under MIT, so anyone can check how the key is handled. Key handling was located in the published source. - Privacy 14/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 7/10: Key handling was found in the published source by BYOAIK's scanner. What was checked (verification tier SOURCE_VERIFIED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [STRONG via SOURCE_SCAN] The user supplies their own OpenAI key: the project references its API key variable. (https://github.com/yihong0618/bilingual_book_maker/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] The user supplies their own Anthropic key: the project references its API key variable. (https://github.com/yihong0618/bilingual_book_maker/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] The user supplies their own Google Gemini key: the project references its API key variable. (https://github.com/yihong0618/bilingual_book_maker/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] The user supplies their own Groq key: the project references its API key variable. (https://github.com/yihong0618/bilingual_book_maker/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] The user supplies their own xAI Grok key: the project references its API key variable. (https://github.com/yihong0618/bilingual_book_maker/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/yihong0618/bilingual_book_maker/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] The provider endpoint is configurable, so requests can be pointed straight at the provider or at your own gateway. (https://github.com/yihong0618/bilingual_book_maker/blob/HEAD/README.md) - [CONFIRMED via SOURCE_SCAN] Ships a container definition, so it can be self-hosted on your own infrastructure. (https://github.com/yihong0618/bilingual_book_maker/blob/HEAD/Dockerfile) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-27, about 0 month(s) ago. (https://github.com/yihong0618/bilingual_book_maker/commits) ### gptme - Page: https://www.byoaik.com/tools/gptme/ - Website: https://gptme.org - Source: https://github.com/gptme/gptme - Category: Developer Infrastructure | Pricing: OPEN_SOURCE | Open source: yes (MIT) | Trust score: 93/100 - Platforms: CLI, Self-hosted, Library/SDK, Docker - Bring-your-own-key providers: OpenAI, Anthropic, Google Gemini, Azure OpenAI, OpenRouter, Groq, DeepSeek, xAI Grok, Ollama, Custom / OpenAI-compatible - Key storage: ENVIRONMENT_VARIABLE | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit today) gptme is a free, open-source (MIT) personal AI agent that runs in any terminal: laptop, SSH session, tmux, headless server, or CI pipeline. It executes shell commands and Python, reads/writes/patches files, browses the web via Playwright, and analyzes images. It is fully provider-agnostic and bring-your-own-key: the end user supplies their own LLM API credentials via environment variables (OPENAI_API_KEY, ANTHROPIC_API_KEY, OPENROUTER_API_KEY, GEMINI_API_KEY, XAI_API_KEY, GROQ_API_KEY, DEEPSEEK_API_KEY), a config file (~/.config/gptme/config.toml), or an interactive /account setup command. No built-in or hosted key is required (an optional gptme.ai gateway is "coming soon"). Supports OpenAI, Anthropic, Google Gemini, xAI Grok, DeepSeek, Groq, OpenRouter, Azure OpenAI, and any OpenAI-compatible server including Ollama and llama.cpp for fully local use. One of the earliest agent CLIs (launched spring 2023) and still actively maintained. Key handling: Your API key is supplied via an environment variable or local config file. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 93/100 (assessable for any licence), verifiability 90/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 22/25: The key is supplied by an environment variable or local config file you control. - Request Routing 20/20: Requests go straight from you to the AI provider. The provider endpoint is configurable, so you can point it at the provider or your own gateway. - Transparency 20/20: Source is public under MIT, so anyone can check how the key is handled. Key handling was located in the published source. - Privacy 14/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 7/10: Key handling was found in the published source by BYOAIK's scanner. What was checked (verification tier SOURCE_VERIFIED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [STRONG via SOURCE_SCAN] The user supplies their own OpenAI key: the project references its API key variable. (https://github.com/gptme/gptme/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] The user supplies their own Anthropic key: the project references its API key variable. (https://github.com/gptme/gptme/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] The user supplies their own Google Gemini key: the project references its API key variable. (https://github.com/gptme/gptme/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] The user supplies their own OpenRouter key: the project references its API key variable. (https://github.com/gptme/gptme/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] The user supplies their own Groq key: the project references its API key variable. (https://github.com/gptme/gptme/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] The user supplies their own DeepSeek key: the project references its API key variable. (https://github.com/gptme/gptme/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] The user supplies their own xAI Grok key: the project references its API key variable. (https://github.com/gptme/gptme/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/gptme/gptme/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] The provider endpoint is configurable, so requests can be pointed straight at the provider or at your own gateway. (https://github.com/gptme/gptme/blob/HEAD/README.md) - [CONFIRMED via SOURCE_SCAN] Ships a container definition, so it can be self-hosted on your own infrastructure. (https://github.com/gptme/gptme/blob/HEAD/docker-compose.yml) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-29, about 0 month(s) ago. (https://github.com/gptme/gptme/commits) ### RamaLama - Page: https://www.byoaik.com/tools/ramalama/ - Website: https://ramalama.ai - Source: https://github.com/containers/ramalama - Category: Local AI Tools | Pricing: OPEN_SOURCE | Open source: yes (MIT) | Trust score: 93/100 - Platforms: CLI, Self-hosted, Docker - Bring-your-own-key providers: Ollama, Hugging Face - Key storage: NO_KEY_LOCAL_MODEL | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit 2 days ago) RamaLama, from the Containers organisation, is a CLI that runs models locally by wrapping inference engines such as llama.cpp inside rootless OCI containers with no network access by default, detecting your GPU and pulling the matching containerised runtime. You run a single command and everything executes locally in an isolated container, so there is no key because inference never leaves the host. Installs through Fedora packages, PyPI or macOS installers, with Windows support through Podman or Docker with WSL2. MIT licensed, about 3,000 stars, pushed the day of review. Key handling: Key handling varies. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 93/100 (assessable for any licence), verifiability 90/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 25/25: No cloud key is needed at all. The model runs locally. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under MIT, so anyone can check how the key is handled. Key handling was located in the published source. - Privacy 14/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 7/10: Key handling was found in the published source by BYOAIK's scanner. What was checked (verification tier SOURCE_VERIFIED, derived from this evidence): - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/containers/ramalama/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] The user supplies their own Hugging Face key: the project references its API key variable. (https://github.com/containers/ramalama/blob/HEAD/pyproject.toml) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-29, about 0 month(s) ago. (https://github.com/containers/ramalama/commits) ### Hugging Face Chat UI - Page: https://www.byoaik.com/tools/hugging-face-chat-ui/ - Website: https://huggingface.co/docs/chat-ui - Source: https://github.com/huggingface/chat-ui - Category: AI Chat Interfaces | Pricing: OPEN_SOURCE | Open source: yes (Apache-2.0) | Trust score: 92/100 - Platforms: Web, Self-hosted, Docker - Bring-your-own-key providers: OpenAI, Ollama, Hugging Face, OpenRouter, Custom / OpenAI-compatible - Key storage: SELF_HOSTED_ONLY | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit 3 days ago) Chat UI is the open-source codebase behind Hugging Face's HuggingChat. It is a SvelteKit web app (MongoDB optional; an embedded DB is used in dev) that you self-host and point at any OpenAI-compatible API. BYOK is the core configuration path: you create a .env.local file and supply OPENAI_BASE_URL plus OPENAI_API_KEY, and Chat UI auto-discovers available models from your endpoint. The docs give ready-made base-URL/key pairs for the Hugging Face Inference Providers router (hf_xxx), Ollama (local), llama.cpp (local), and OpenRouter (sk-or-v1-xxx), and state that any service speaking the OpenAI protocol works by default. It supports MCP tool/function calling, multimodal image input on vision models, an LLM router for model selection, and optional OpenID authentication. Note: recent versions (v0.10.0, May 2026) speak the OpenAI protocol only and removed legacy provider-specific integrations, so non-OpenAI providers like Anthropic or Gemini must be reached through an OpenAI-compatible proxy/router rather than a native key field. Key handling: You self-host the app, so your API key lives in your own deployment. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 92/100 (assessable for any licence), verifiability 90/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 21/25: The key lives in a deployment you run yourself. - Request Routing 20/20: Requests go straight from you to the AI provider. The provider endpoint is configurable, so you can point it at the provider or your own gateway. - Transparency 20/20: Source is public under Apache-2.0, so anyone can check how the key is handled. Key handling was located in the published source. - Privacy 14/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 7/10: Key handling was found in the published source by BYOAIK's scanner. What was checked (verification tier SOURCE_VERIFIED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [STRONG via SOURCE_SCAN] The user supplies their own OpenAI key: the project references its API key variable. (https://github.com/huggingface/chat-ui/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] The user supplies their own Hugging Face key: the project references its API key variable. (https://github.com/huggingface/chat-ui/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/huggingface/chat-ui/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] The provider endpoint is configurable, so requests can be pointed straight at the provider or at your own gateway. (https://github.com/huggingface/chat-ui/blob/HEAD/README.md) - [CONFIRMED via SOURCE_SCAN] Ships a container definition, so it can be self-hosted on your own infrastructure. (https://github.com/huggingface/chat-ui/blob/HEAD/docker-compose.yml) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-29, about 0 month(s) ago. (https://github.com/huggingface/chat-ui/commits) ### LiteLLM - Page: https://www.byoaik.com/tools/litellm/ - Website: https://www.litellm.ai/ - Source: https://github.com/BerriAI/litellm - Category: Self-Hosted AI Platforms | Pricing: OPEN_SOURCE | Open source: yes (MIT (enterprise/ directory separately licensed)) | Trust score: 92/100 - Platforms: Self-hosted, Docker, CLI - Bring-your-own-key providers: OpenAI, Anthropic, Google Gemini, Azure OpenAI, AWS Bedrock, Mistral, DeepSeek, Groq, OpenRouter, Together AI, Cohere, Ollama, Custom / OpenAI-compatible - Key storage: ENVIRONMENT_VARIABLE | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit today) LiteLLM is a gateway that exposes a single OpenAI-compatible interface over more than a hundred model providers, available as a Python SDK and as a self-hosted proxy server with virtual keys, spend tracking, fallbacks and an admin UI. Because you run the proxy, the upstream provider keys stay in your own deployment and never sit with a third party; the virtual keys it issues let you hand scoped access to your own applications without sharing the real secret. Licensing is mixed: everything outside the enterprise/ directory is MIT, and that directory carries its own commercial licence. Key handling: Your API key is supplied via an environment variable or local config file. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 92/100 (assessable for any licence), verifiability 90/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 24/25: The key is supplied by an environment variable or local config file you control. Observed using the operating system keychain rather than a file. - Request Routing 20/20: Requests go straight from you to the AI provider. The provider endpoint is configurable, so you can point it at the provider or your own gateway. - Transparency 20/20: Source is public under MIT (enterprise/ directory separately licensed), so anyone can check how the key is handled. Key handling was located in the published source. - Privacy 11/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. Analytics or error-reporting libraries are present; what they send was not established. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 7/10: Key handling was found in the published source by BYOAIK's scanner. What was checked (verification tier SOURCE_VERIFIED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [STRONG via SOURCE_SCAN] The user supplies their own OpenAI key: the project references its API key variable. (https://github.com/BerriAI/litellm/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] The user supplies their own Anthropic key: the project references its API key variable. (https://github.com/BerriAI/litellm/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/BerriAI/litellm/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] The provider endpoint is configurable, so requests can be pointed straight at the provider or at your own gateway. (https://github.com/BerriAI/litellm/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] The user supplies their own Cohere key: the project references its API key variable. (https://github.com/BerriAI/litellm/blob/HEAD/.env.example) - [CONFIRMED via SOURCE_SCAN] Ships a container definition, so it can be self-hosted on your own infrastructure. (https://github.com/BerriAI/litellm/blob/HEAD/docker-compose.yml) - [STRONG via SOURCE_SCAN] Uses the operating system keychain / credential store to hold the key. (https://github.com/BerriAI/litellm/blob/HEAD/pyproject.toml) - [REPORTED via SOURCE_SCAN] References Sentry, so some analytics or error reporting is present. This scan cannot tell whether it is opt-in or what it sends. (https://github.com/BerriAI/litellm/blob/HEAD/pyproject.toml) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-29, about 0 month(s) ago. (https://github.com/BerriAI/litellm/commits) ### Agent Desk - Page: https://www.byoaik.com/tools/agent-desk/ - Website: https://github.com/huabeitech/agent-desk - Source: https://github.com/huabeitech/agent-desk - Category: Enterprise AI Tools | Pricing: OPEN_SOURCE | Open source: yes (Apache-2.0) | Trust score: 92/100 - Platforms: Self-hosted, Docker, Web - Bring-your-own-key providers: OpenAI, Custom / OpenAI-compatible, Ollama - Key storage: SELF_HOSTED_ONLY | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit 19 days ago) Agent Desk is an open-source AI customer support system combining knowledge-based answers, human handoff and ticket workflows. The runtime supports Skills, MCP and OpenAI-compatible model access, so the operator nominates the endpoint and supplies the credential. It runs on SQLite or MySQL with Qdrant for vectors, and is built for intranet and enterprise self-hosting where support conversations should not leave the network. Apache-2.0. Key handling: You self-host the app, so your API key lives in your own deployment. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 92/100 (assessable for any licence), verifiability 90/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 21/25: The key lives in a deployment you run yourself. - Request Routing 20/20: Requests go straight from you to the AI provider. The provider endpoint is configurable, so you can point it at the provider or your own gateway. - Transparency 20/20: Source is public under Apache-2.0, so anyone can check how the key is handled. - Privacy 14/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 7/10: Key handling was found in the published source by BYOAIK's scanner. What was checked (verification tier SOURCE_VERIFIED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [STRONG via SOURCE_SCAN] The provider endpoint is configurable, so requests can be pointed straight at the provider or at your own gateway. (https://github.com/huabeitech/agent-desk/blob/HEAD/config/config.example.yaml) - [CONFIRMED via SOURCE_SCAN] Ships a container definition, so it can be self-hosted on your own infrastructure. (https://github.com/huabeitech/agent-desk/blob/HEAD/docker-compose.yml) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-27, about 0 month(s) ago. (https://github.com/huabeitech/agent-desk/commits) ### OpenDocuments - Page: https://www.byoaik.com/tools/opendocuments/ - Website: https://github.com/joungminsung/OpenDocuments - Source: https://github.com/joungminsung/OpenDocuments - Category: Enterprise AI Tools | Pricing: OPEN_SOURCE | Open source: yes (MIT) | Trust score: 92/100 - Platforms: Self-hosted, Docker, Web - Bring-your-own-key providers: OpenAI, Ollama, Custom / OpenAI-compatible - Key storage: SELF_HOSTED_ONLY | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Slowing down (last commit about 3 months ago) OpenDocuments is a self-hosted retrieval platform that searches across GitHub, Notion, Google Drive, local files and web sources and answers with citations, aimed at cases like finding a policy across an HR space, a handbook and an email thread. It deploys on your own infrastructure, and the project states that with a local LLM served through Ollama the data never leaves your network. MIT licensed. Key handling: You self-host the app, so your API key lives in your own deployment. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 92/100 (assessable for any licence), verifiability 90/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 21/25: The key lives in a deployment you run yourself. - Request Routing 20/20: Requests go straight from you to the AI provider. The provider endpoint is configurable, so you can point it at the provider or your own gateway. - Transparency 20/20: Source is public under MIT, so anyone can check how the key is handled. Key handling was located in the published source. - Privacy 14/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 7/10: Key handling was found in the published source by BYOAIK's scanner. What was checked (verification tier SOURCE_VERIFIED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [STRONG via SOURCE_SCAN] The user supplies their own OpenAI key: the project references its API key variable. (https://github.com/joungminsung/OpenDocuments/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/joungminsung/OpenDocuments/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] The provider endpoint is configurable, so requests can be pointed straight at the provider or at your own gateway. (https://github.com/joungminsung/OpenDocuments/blob/HEAD/README.md) - [CONFIRMED via SOURCE_SCAN] Ships a container definition, so it can be self-hosted on your own infrastructure. (https://github.com/joungminsung/OpenDocuments/blob/HEAD/docker-compose.yml) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-07-26, about 2 month(s) ago. (https://github.com/joungminsung/OpenDocuments/commits) ### Bifrost - Page: https://www.byoaik.com/tools/bifrost/ - Website: https://github.com/maximhq/bifrost - Source: https://github.com/maximhq/bifrost - Category: Self-Hosted AI Platforms | Pricing: OPEN_SOURCE | Open source: yes (Apache-2.0) | Trust score: 92/100 - Platforms: Self-hosted, Docker - Bring-your-own-key providers: OpenAI, Anthropic, Google Gemini, Azure OpenAI, AWS Bedrock, Mistral, Cohere, Ollama, Custom / OpenAI-compatible - Key storage: SELF_HOSTED_ONLY | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit today) Bifrost is an open-source AI gateway that fronts multiple model providers behind one interface, with MCP support, routing, failover and governance controls. Run on your own infrastructure it keeps provider credentials inside your deployment rather than with a vendor, and gives you a single place to see and cap what is spent against them. Apache-2.0. Key handling: You self-host the app, so your API key lives in your own deployment. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 92/100 (assessable for any licence), verifiability 90/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 21/25: The key lives in a deployment you run yourself. - Request Routing 20/20: Requests go straight from you to the AI provider. The provider endpoint is configurable, so you can point it at the provider or your own gateway. - Transparency 20/20: Source is public under Apache-2.0, so anyone can check how the key is handled. - Privacy 14/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 7/10: Key handling was found in the published source by BYOAIK's scanner. What was checked (verification tier SOURCE_VERIFIED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/maximhq/bifrost/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] The provider endpoint is configurable, so requests can be pointed straight at the provider or at your own gateway. (https://github.com/maximhq/bifrost/blob/HEAD/README.md) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-29, about 0 month(s) ago. (https://github.com/maximhq/bifrost/commits) ### OpenMAIC - Page: https://www.byoaik.com/tools/openmaic/ - Website: https://github.com/THU-MAIC/OpenMAIC - Source: https://github.com/THU-MAIC/OpenMAIC - Category: Education Tools | Pricing: OPEN_SOURCE | Open source: yes (MIT) | Trust score: 92/100 - Platforms: Web, Self-hosted, Docker - Bring-your-own-key providers: OpenAI, Anthropic, Google Gemini, DeepSeek, Groq, OpenRouter, LM Studio, Ollama, Custom / OpenAI-compatible - Key storage: SELF_HOSTED_ONLY | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit today) OpenMAIC is an open multi-agent interactive classroom from Tsinghua University's MAIC group: agents generate slides, quizzes, simulations and project work around a topic rather than serving one fixed lesson. It is self-hosted and provider-agnostic, with OPENAI_API_KEY, Anthropic, Ollama and any OpenAI-compatible endpoint configured by the operator, so a school can run the whole thing against a local model and keep student work on its own machines. MIT, and by a wide margin the most-adopted education tool in this directory. Key handling: You self-host the app, so your API key lives in your own deployment. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 92/100 (assessable for any licence), verifiability 90/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 21/25: The key lives in a deployment you run yourself. - Request Routing 20/20: Requests go straight from you to the AI provider. The provider endpoint is configurable, so you can point it at the provider or your own gateway. - Transparency 20/20: Source is public under MIT, so anyone can check how the key is handled. Key handling was located in the published source. - Privacy 14/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 7/10: Key handling was found in the published source by BYOAIK's scanner. What was checked (verification tier SOURCE_VERIFIED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [STRONG via SOURCE_SCAN] The user supplies their own OpenAI key: the project references its API key variable. (https://github.com/THU-MAIC/OpenMAIC/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] The user supplies their own Anthropic key: the project references its API key variable. (https://github.com/THU-MAIC/OpenMAIC/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] The user supplies their own OpenRouter key: the project references its API key variable. (https://github.com/THU-MAIC/OpenMAIC/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] The user supplies their own Azure OpenAI key: the project references its API key variable. (https://github.com/THU-MAIC/OpenMAIC/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] The user supplies their own xAI Grok key: the project references its API key variable. (https://github.com/THU-MAIC/OpenMAIC/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/THU-MAIC/OpenMAIC/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] The provider endpoint is configurable, so requests can be pointed straight at the provider or at your own gateway. (https://github.com/THU-MAIC/OpenMAIC/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] The user supplies their own DeepSeek key: the project references its API key variable. (https://github.com/THU-MAIC/OpenMAIC/blob/HEAD/.env.example) - [CONFIRMED via SOURCE_SCAN] Ships a container definition, so it can be self-hosted on your own infrastructure. (https://github.com/THU-MAIC/OpenMAIC/blob/HEAD/docker-compose.yml) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-29, about 0 month(s) ago. (https://github.com/THU-MAIC/OpenMAIC/commits) ### NextChat - Page: https://www.byoaik.com/tools/nextchat/ - Website: https://nextchat.club/ - Source: https://github.com/ChatGPTNextWeb/NextChat - Category: AI Chat Interfaces | Pricing: OPEN_SOURCE | Open source: yes (MIT) | Trust score: 91/100 - Platforms: Web, Desktop, Mobile, Self-hosted, Docker - Bring-your-own-key providers: OpenAI, Anthropic, Google Gemini, Azure OpenAI, Groq, DeepSeek, Ollama, Custom / OpenAI-compatible - Key storage: LOCAL_DEVICE | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit 19 days ago) NextChat (formerly ChatGPT-Next-Web) is an open-source, cross-platform AI assistant you deploy yourself and point at your own API keys. It runs as a web app, a Tauri desktop app (Windows/macOS/Linux), and mobile builds (iOS/Android), with all chat data stored locally. Users supply their own keys for OpenAI, Anthropic Claude, Google Gemini, DeepSeek, Azure OpenAI, Groq, Ollama/LocalAI and others, either via in-app settings or environment variables. A BASE_URL env var lets you point it at any OpenAI-compatible endpoint, and one-click Vercel/Docker deploys are supported. Admins can optionally hide the user-key field (HIDE_USER_API_KEY=1) for shared deployments, but the default flow is fully BYOK. The project is enormously popular (88k+ stars) and actively maintained, with commits as recent as May 2026. Key handling: Your API key is stored locally on your device. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 90/100 (assessable for any licence), verifiability 90/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 23/25: The key is held on your own device. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under MIT, so anyone can check how the key is handled. Key handling was located in the published source. - Privacy 14/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 7/10: Key handling was found in the published source by BYOAIK's scanner. What was checked (verification tier SOURCE_VERIFIED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [STRONG via SOURCE_SCAN] The user supplies their own OpenAI key: the project references its API key variable. (https://github.com/ChatGPTNextWeb/NextChat/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] The user supplies their own Anthropic key: the project references its API key variable. (https://github.com/ChatGPTNextWeb/NextChat/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] The user supplies their own DeepSeek key: the project references its API key variable. (https://github.com/ChatGPTNextWeb/NextChat/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] The user supplies their own xAI Grok key: the project references its API key variable. (https://github.com/ChatGPTNextWeb/NextChat/blob/HEAD/.env.template) - [CONFIRMED via SOURCE_SCAN] Ships a container definition, so it can be self-hosted on your own infrastructure. (https://github.com/ChatGPTNextWeb/NextChat/blob/HEAD/docker-compose.yml) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-08-11, about 2 month(s) ago. (https://github.com/ChatGPTNextWeb/NextChat/commits) ### OpenClaw - Page: https://www.byoaik.com/tools/openclaw/ - Website: https://openclaw.ai/ - Source: https://github.com/openclaw/openclaw - Category: Desktop AI Assistants | Pricing: OPEN_SOURCE | Open source: yes (MIT) | Trust score: 91/100 - Platforms: Desktop, Mobile, Self-hosted, Local-only, Docker - Bring-your-own-key providers: OpenAI, Anthropic, Google Gemini, OpenRouter, Ollama, Custom / OpenAI-compatible - Key storage: LOCAL_DEVICE | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit today) OpenClaw runs an AI assistant on your own computer and meets you in Discord, Slack, Telegram, WhatsApp, iMessage, Teams and twenty more, with native apps across macOS, iOS, Android, Windows and Linux. The key-handling posture is stated bluntly in its README: state, memory and credentials live on your hardware, and models are swappable plugins covering Claude, Codex and local models. Telemetry is a daily version check only, anonymous feature statistics are opt-in, and one config line disables both. Stewarded by the OpenClaw Foundation, an independent 501(c)(3), with no paid tier, hosted service or token. OpenAI is named as a donor rather than an owner, which the Foundation says outright. MIT. Key handling: Your API key is stored locally on your device. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 90/100 (assessable for any licence), verifiability 90/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 23/25: The key is held on your own device. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under MIT, so anyone can check how the key is handled. Key handling was located in the published source. - Privacy 14/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 7/10: Key handling was found in the published source by BYOAIK's scanner. What was checked (verification tier SOURCE_VERIFIED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [STRONG via SOURCE_SCAN] The user supplies their own OpenAI key: the project references its API key variable. (https://github.com/openclaw/openclaw/blob/HEAD/.env.example) - [STRONG via SOURCE_SCAN] The user supplies their own Anthropic key: the project references its API key variable. (https://github.com/openclaw/openclaw/blob/HEAD/.env.example) - [STRONG via SOURCE_SCAN] The user supplies their own Google Gemini key: the project references its API key variable. (https://github.com/openclaw/openclaw/blob/HEAD/.env.example) - [STRONG via SOURCE_SCAN] The user supplies their own OpenRouter key: the project references its API key variable. (https://github.com/openclaw/openclaw/blob/HEAD/.env.example) - [STRONG via SOURCE_SCAN] The user supplies their own Perplexity key: the project references its API key variable. (https://github.com/openclaw/openclaw/blob/HEAD/.env.example) - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/openclaw/openclaw/blob/HEAD/docker-compose.yml) - [CONFIRMED via SOURCE_SCAN] Ships a container definition, so it can be self-hosted on your own infrastructure. (https://github.com/openclaw/openclaw/blob/HEAD/docker-compose.yml) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-29, about 0 month(s) ago. (https://github.com/openclaw/openclaw/commits) ### torchtune - Page: https://www.byoaik.com/tools/torchtune/ - Website: https://github.com/meta-pytorch/torchtune - Source: https://github.com/meta-pytorch/torchtune - Category: Fine-Tuning and Training | Pricing: OPEN_SOURCE | Open source: yes (BSD-3-Clause) | Trust score: 91/100 - Platforms: CLI, Self-hosted, Local-only, Docker - Bring-your-own-key providers: Ollama, Custom / OpenAI-compatible - Key storage: NO_KEY_LOCAL_MODEL | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Stale (last commit about 7 months ago) torchtune is Meta's PyTorch-native library of fine-tuning recipes, written to be read and modified rather than configured, which suits people who want to see exactly what the training loop does. It runs locally with no key and no service. Note the pace: the last commit on the default branch was several months before review, so it is steady rather than busy, and the maintenance badge on this listing reflects that. BSD-3-Clause. Key handling: Key handling varies. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 93/100 (assessable for any licence), verifiability 90/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 25/25: No cloud key is needed at all. The model runs locally. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under BSD-3-Clause, so anyone can check how the key is handled. Key handling was located in the published source. - Privacy 14/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 8/10: Commits within the last six months. - Verification Confidence 7/10: Key handling was found in the published source by BYOAIK's scanner. What was checked (verification tier SOURCE_VERIFIED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [STRONG via SOURCE_SCAN] The user supplies their own Hugging Face key: the project references its API key variable. (https://github.com/meta-pytorch/torchtune/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/meta-pytorch/torchtune/blob/HEAD/pyproject.toml) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-04-23, about 5 month(s) ago. (https://github.com/meta-pytorch/torchtune/commits) ### OpenLLM - Page: https://www.byoaik.com/tools/openllm/ - Website: https://github.com/bentoml/OpenLLM - Source: https://github.com/bentoml/OpenLLM - Category: Self-Hosted AI Platforms | Pricing: OPEN_SOURCE | Open source: yes (Apache-2.0) | Trust score: 91/100 - Platforms: Self-hosted, CLI, Docker, Local-only - Bring-your-own-key providers: Ollama, Custom / OpenAI-compatible - Key storage: ENVIRONMENT_VARIABLE | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit 17 days ago) OpenLLM runs open models as OpenAI-compatible API servers with a single command, covering Llama, Qwen, Mistral and others, with a built-in chat UI. It is the shortest path in this category from nothing to an endpoint of your own that existing BYOK clients can be pointed at. From the BentoML team, and it hands off to BentoML for anything production-shaped. Apache-2.0. Key handling: Your API key is supplied via an environment variable or local config file. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 93/100 (assessable for any licence), verifiability 90/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 22/25: The key is supplied by an environment variable or local config file you control. - Request Routing 20/20: Requests go straight from you to the AI provider. The provider endpoint is configurable, so you can point it at the provider or your own gateway. - Transparency 20/20: Source is public under Apache-2.0, so anyone can check how the key is handled. Key handling was located in the published source. - Privacy 14/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 8/10: Commits within the last six months. - Verification Confidence 7/10: Key handling was found in the published source by BYOAIK's scanner. What was checked (verification tier SOURCE_VERIFIED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [STRONG via SOURCE_SCAN] The user supplies their own Hugging Face key: the project references its API key variable. (https://github.com/bentoml/OpenLLM/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/bentoml/OpenLLM/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] The provider endpoint is configurable, so requests can be pointed straight at the provider or at your own gateway. (https://github.com/bentoml/OpenLLM/blob/HEAD/README.md) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-05-29, about 4 month(s) ago. (https://github.com/bentoml/OpenLLM/commits) ### Higgs Audio - Page: https://www.byoaik.com/tools/higgs-audio/ - Website: https://github.com/boson-ai/higgs-audio - Source: https://github.com/boson-ai/higgs-audio - Category: Music and Audio Generation | Pricing: OPEN_SOURCE | Open source: yes (Apache-2.0) | Trust score: 91/100 - Platforms: CLI, Self-hosted, Local-only, Docker - Bring-your-own-key providers: Ollama, Custom / OpenAI-compatible - Key storage: NO_KEY_LOCAL_MODEL | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Slowing down (last commit about 5 months ago) Higgs Audio from Boson AI generates expressive speech including multi-speaker dialogue and background-aware narration, with weights and inference code published for local use. It fits the same pattern as the rest of this category: the model runs where you put it, so scripts and voice references are not uploaded anywhere. Apache-2.0. Note the pace, several months since the last commit, which the maintenance badge reflects. Key handling: Key handling varies. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 93/100 (assessable for any licence), verifiability 90/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 25/25: No cloud key is needed at all. The model runs locally. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under Apache-2.0, so anyone can check how the key is handled. Key handling was located in the published source. - Privacy 14/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 8/10: Commits within the last six months. - Verification Confidence 7/10: Key handling was found in the published source by BYOAIK's scanner. What was checked (verification tier SOURCE_VERIFIED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [STRONG via SOURCE_SCAN] The user supplies their own Hugging Face key: the project references its API key variable. (https://github.com/boson-ai/higgs-audio/blob/HEAD/README.md) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-06-05, about 4 month(s) ago. (https://github.com/boson-ai/higgs-audio/commits) ### Buzz - Page: https://www.byoaik.com/tools/buzz/ - Website: https://chidiwilliams.github.io/buzz/ - Source: https://github.com/chidiwilliams/buzz - Category: Voice and Transcription Tools | Pricing: OPEN_SOURCE | Open source: yes (MIT) | Trust score: 90/100 - Platforms: Desktop, CLI, Self-hosted, Docker - Bring-your-own-key providers: OpenAI, Groq, Anthropic, Google Gemini, Ollama, Hugging Face, Custom / OpenAI-compatible - Key storage: LOCAL_DEVICE | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit 2 days ago) Buzz is an open-source, cross-platform desktop application (macOS, Windows, Linux) built with Python and Qt that transcribes and translates audio and video. By default it runs entirely offline using local Whisper implementations (Whisper, Whisper.cpp, Faster Whisper, and Whisper-compatible Hugging Face models, with CUDA/Apple Silicon/Vulkan acceleration). For remote speech recognition, users can set their own OpenAI API key in Preferences. The Preferences screen also exposes a configurable Base URL (default https://api.openai.com/v1/), so the end user can point Buzz at any OpenAI-compatible endpoint. The maintainer-documented Discussion #827 confirms working setups with the user's own keys for Groq (https://api.groq.com/openai/v1, free transcription), Lemonfox, and self-hosted Whisper.cpp servers (RunPod/Vast.ai/TensorDock), and for the translation feature, OpenAI-compatible endpoints from Anthropic, Google Gemini, Ollama, and Groq via custom base URL plus API key. Exports to TXT/SRT/VTT; supports files, YouTube links, live microphone, watch folders, and a CLI. Key handling: Your API key is stored locally on your device. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 88/100 (assessable for any licence), verifiability 90/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 25/25: The key is held on your own device. Observed using the operating system keychain rather than a file. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under MIT, so anyone can check how the key is handled. - Privacy 11/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. Analytics or error-reporting libraries are present; what they send was not established. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 7/10: Key handling was found in the published source by BYOAIK's scanner. What was checked (verification tier SOURCE_VERIFIED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [STRONG via SOURCE_SCAN] Uses the operating system keychain / credential store to hold the key. (https://github.com/chidiwilliams/buzz/blob/HEAD/pyproject.toml) - [REPORTED via SOURCE_SCAN] References PostHog, so some analytics or error reporting is present. This scan cannot tell whether it is opt-in or what it sends. (https://github.com/chidiwilliams/buzz/blob/HEAD/pyproject.toml) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-23, about 0 month(s) ago. (https://github.com/chidiwilliams/buzz/commits) ### Skyvern - Page: https://www.byoaik.com/tools/skyvern/ - Website: https://github.com/Skyvern-AI/skyvern - Source: https://github.com/Skyvern-AI/skyvern - Category: Agent Frameworks | Pricing: OPEN_SOURCE | Open source: yes (AGPL-3.0) | Trust score: 90/100 - Platforms: Self-hosted, Library/SDK, CLI, Web, Docker - Bring-your-own-key providers: OpenAI, Anthropic, Google Gemini, Azure OpenAI, OpenRouter, Groq, Ollama, Custom / OpenAI-compatible - Key storage: ENVIRONMENT_VARIABLE | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit today) Skyvern is an open-source AI agent that automates browser-based workflows using LLMs and computer vision rather than brittle XPath/CSS selectors, letting it operate on websites it has never seen before. It exposes a Playwright-compatible SDK plus an API and workflow builder, and runs self-hosted (Python, Docker) or via Skyvern Cloud. Users bring their own LLM provider key: the .env.example ships empty OPENAI_API_KEY / ANTHROPIC_API_KEY / GEMINI_API_KEY fields, and the self-hosted LLM Configuration docs document an ENABLE_=true + _API_KEY pattern across OpenAI, Anthropic, Azure OpenAI, Google Gemini/Vertex, AWS Bedrock, Ollama (local), OpenRouter, Groq, plus an OpenAI-compatible/custom-endpoint option and several regional providers (MiniMax, VolcEngine, Novita, Moonshot). Licensed AGPL-3.0; anti-bot detection, proxy network, and CAPTCHA solvers are reserved for the managed cloud offering. ~22k GitHub stars; latest release v1.0.43 (June 18, 2026). Key handling: Your API key is supplied via an environment variable or local config file. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 88/100 (assessable for any licence), verifiability 90/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 22/25: The key is supplied by an environment variable or local config file you control. - Request Routing 20/20: Requests go straight from you to the AI provider. The provider endpoint is configurable, so you can point it at the provider or your own gateway. - Transparency 20/20: Source is public under AGPL-3.0, so anyone can check how the key is handled. Key handling was located in the published source. - Privacy 11/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. Analytics or error-reporting libraries are present; what they send was not established. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 7/10: Key handling was found in the published source by BYOAIK's scanner. What was checked (verification tier SOURCE_VERIFIED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/Skyvern-AI/skyvern/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] The provider endpoint is configurable, so requests can be pointed straight at the provider or at your own gateway. (https://github.com/Skyvern-AI/skyvern/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] The user supplies their own OpenAI key: the project references its API key variable. (https://github.com/Skyvern-AI/skyvern/blob/HEAD/.env.example) - [STRONG via SOURCE_SCAN] The user supplies their own Anthropic key: the project references its API key variable. (https://github.com/Skyvern-AI/skyvern/blob/HEAD/.env.example) - [STRONG via SOURCE_SCAN] The user supplies their own Google Gemini key: the project references its API key variable. (https://github.com/Skyvern-AI/skyvern/blob/HEAD/.env.example) - [STRONG via SOURCE_SCAN] The user supplies their own xAI Grok key: the project references its API key variable. (https://github.com/Skyvern-AI/skyvern/blob/HEAD/.env.example) - [STRONG via SOURCE_SCAN] The user supplies their own OpenRouter key: the project references its API key variable. (https://github.com/Skyvern-AI/skyvern/blob/HEAD/docker-compose.yml) - [STRONG via SOURCE_SCAN] The user supplies their own Groq key: the project references its API key variable. (https://github.com/Skyvern-AI/skyvern/blob/HEAD/docker-compose.yml) - [CONFIRMED via SOURCE_SCAN] Ships a container definition, so it can be self-hosted on your own infrastructure. (https://github.com/Skyvern-AI/skyvern/blob/HEAD/docker-compose.yml) - [REPORTED via SOURCE_SCAN] References PostHog, so some analytics or error reporting is present. This scan cannot tell whether it is opt-in or what it sends. (https://github.com/Skyvern-AI/skyvern/blob/HEAD/pyproject.toml) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-29, about 0 month(s) ago. (https://github.com/Skyvern-AI/skyvern/commits) ### OWL (CAMEL-AI) - Page: https://www.byoaik.com/tools/owl-camel-ai/ - Website: https://github.com/camel-ai/owl - Source: https://github.com/camel-ai/owl - Category: Agent Frameworks | Pricing: OPEN_SOURCE | Open source: yes (Apache-2.0) | Trust score: 90/100 - Platforms: Library/SDK, CLI, Web, Self-hosted, Docker - Bring-your-own-key providers: OpenAI, Anthropic, Google Gemini, Azure OpenAI, OpenRouter, DeepSeek, Ollama, Custom / OpenAI-compatible - Key storage: ENVIRONMENT_VARIABLE | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit 6 days ago) OWL (Optimized Workforce Learning) is an open-source multi-agent collaboration framework from CAMEL-AI for automating real-world tasks via coordinated LLM agents and toolkits (web browsing, code execution, document/image/video processing). It is a Python library/CLI with an optional Gradio web UI (English, Chinese, Japanese) for model selection, API key management, interactive chat, and task history. The end user brings their own provider API keys, set via environment variables, a .env file (copied from .env_template), or directly in the web UI. It ranked #1 among open-source submissions on the GAIA benchmark (~58.18 avg) and was accepted to NeurIPS 2025. Runs locally and via Docker; no data leaves the machine beyond the configured model API calls. Key handling: Your API key is supplied via an environment variable or local config file. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 88/100 (assessable for any licence), verifiability 90/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 22/25: The key is supplied by an environment variable or local config file you control. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under Apache-2.0, so anyone can check how the key is handled. Key handling was located in the published source. - Privacy 14/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 7/10: Key handling was found in the published source by BYOAIK's scanner. What was checked (verification tier SOURCE_VERIFIED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [STRONG via SOURCE_SCAN] The user supplies their own OpenAI key: the project references its API key variable. (https://github.com/camel-ai/owl/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/camel-ai/owl/blob/HEAD/README.md) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-20, about 0 month(s) ago. (https://github.com/camel-ai/owl/commits) ### Kotaemon - Page: https://www.byoaik.com/tools/kotaemon/ - Website: https://cinnamon.github.io/kotaemon/ - Source: https://github.com/Cinnamon/kotaemon - Category: Document Analysis | Pricing: OPEN_SOURCE | Open source: yes (Apache-2.0) | Trust score: 90/100 - Platforms: Self-hosted, Docker, Web - Bring-your-own-key providers: OpenAI, Azure OpenAI, Ollama, Custom / OpenAI-compatible - Key storage: SELF_HOSTED_ONLY | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Slowing down (last commit about 5 months ago) Kotaemon is an open-source RAG web app for chatting with your own documents, built by Cinnamon AI, with citation highlighting aimed at both end users and developers customising the pipeline. Credentials are set in the .env file or through the in-app Resources tab, where the README instructs users to set OPENAI_API_KEY and check that the configured api_key is present for LLMs and embeddings. Supports OpenAI, Azure OpenAI and local models via Ollama or llama.cpp, self-hosted through Docker. Apache-2.0, 25,696 stars. Key handling: You self-host the app, so your API key lives in your own deployment. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 92/100 (assessable for any licence), verifiability 90/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 21/25: The key lives in a deployment you run yourself. - Request Routing 20/20: Requests go straight from you to the AI provider. The provider endpoint is configurable, so you can point it at the provider or your own gateway. - Transparency 20/20: Source is public under Apache-2.0, so anyone can check how the key is handled. Key handling was located in the published source. - Privacy 14/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 8/10: Commits within the last six months. - Verification Confidence 7/10: Key handling was found in the published source by BYOAIK's scanner. What was checked (verification tier SOURCE_VERIFIED, derived from this evidence): - [STRONG via SOURCE_SCAN] The user supplies their own OpenAI key: the project references its API key variable. (https://github.com/Cinnamon/kotaemon/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] The user supplies their own Azure OpenAI key: the project references its API key variable. (https://github.com/Cinnamon/kotaemon/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/Cinnamon/kotaemon/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] The provider endpoint is configurable, so requests can be pointed straight at the provider or at your own gateway. (https://github.com/Cinnamon/kotaemon/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] The user supplies their own Mistral key: the project references its API key variable. (https://github.com/Cinnamon/kotaemon/blob/HEAD/.env.example) - [STRONG via SOURCE_SCAN] The user supplies their own Cohere key: the project references its API key variable. (https://github.com/Cinnamon/kotaemon/blob/HEAD/.env.example) - [CONFIRMED via SOURCE_SCAN] Ships a container definition, so it can be self-hosted on your own infrastructure. (https://github.com/Cinnamon/kotaemon/blob/HEAD/Dockerfile) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-05-30, about 4 month(s) ago. (https://github.com/Cinnamon/kotaemon/commits) ### avante.nvim - Page: https://www.byoaik.com/tools/avante-nvim/ - Website: https://github.com/yetone/avante.nvim - Source: https://github.com/avante-corp/avante.nvim - Category: Coding Assistants | Pricing: OPEN_SOURCE | Open source: yes (Apache-2.0) | Trust score: 90/100 - Platforms: Library/SDK, Self-hosted, CLI, Docker - Bring-your-own-key providers: Anthropic, OpenAI, Azure OpenAI, Google Gemini, Cohere, Ollama, OpenRouter, DeepSeek, Groq, Mistral, Perplexity, Custom / OpenAI-compatible - Key storage: ENVIRONMENT_VARIABLE | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit today) avante.nvim is an open-source Neovim plugin that emulates the Cursor AI IDE inside the editor, providing agentic AI-assisted code suggestions that you can review as diffs and apply directly to your source files. It is fully bring-your-own-key: users supply their own provider API keys via scoped environment variables (e.g. AVANTE_ANTHROPIC_API_KEY, AVANTE_OPENAI_API_KEY) or legacy global vars, or by pointing api_key_name at a shell secret-manager command (cmd: prefix). Each provider's endpoint/base URL is configurable, and the __inherited_from = 'openai' mechanism lets users wire up any OpenAI-compatible vendor. Supported providers include Anthropic, OpenAI, Azure OpenAI, Google Gemini, Cohere, Amazon Bedrock, Ollama/local models, OpenRouter, DeepSeek, Groq, Mistral, Perplexity, and arbitrary custom OpenAI-compatible endpoints. The plugin is free and Apache-2.0 licensed; the user pays their chosen provider directly. Verified June 2026: ~17,981 GitHub stars, Apache-2.0, actively maintained (last push June 20 2026, recent release avante-libs v0.1.2 on June 3 2026). Key handling: Your API key is supplied via an environment variable or local config file. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 88/100 (assessable for any licence), verifiability 90/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 22/25: The key is supplied by an environment variable or local config file you control. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under Apache-2.0, so anyone can check how the key is handled. Key handling was located in the published source. - Privacy 14/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 7/10: Key handling was found in the published source by BYOAIK's scanner. What was checked (verification tier SOURCE_VERIFIED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [STRONG via SOURCE_SCAN] The user supplies their own OpenAI key: the project references its API key variable. (https://github.com/avante-corp/avante.nvim/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] The user supplies their own Anthropic key: the project references its API key variable. (https://github.com/avante-corp/avante.nvim/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] The user supplies their own Azure OpenAI key: the project references its API key variable. (https://github.com/avante-corp/avante.nvim/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/avante-corp/avante.nvim/blob/HEAD/README.md) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-26, about 0 month(s) ago. (https://github.com/avante-corp/avante.nvim/commits) ### PaperQA2 - Page: https://www.byoaik.com/tools/paperqa2/ - Website: https://github.com/Future-House/paper-qa - Source: https://github.com/Future-House/paper-qa - Category: Research Tools | Pricing: OPEN_SOURCE | Open source: yes (Apache-2.0) | Trust score: 90/100 - Platforms: CLI, Library/SDK, Self-hosted, Docker - Bring-your-own-key providers: OpenAI, Anthropic, Google Gemini, Ollama, Custom / OpenAI-compatible - Key storage: ENVIRONMENT_VARIABLE | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit 14 days ago) PaperQA2 is a Python library and CLI that performs retrieval-augmented question answering over a collection of scientific papers you supply, answering with citations into the source documents rather than free-form text. The README instructs users to set an API key environment variable such as OPENAI_API_KEY, and states that any LiteLLM-compatible model can be configured, which covers Anthropic, Gemini and local backends through llama.cpp or Ollama. It is library-first rather than a graphical app. Apache-2.0, 9,015 stars. Key handling: Your API key is supplied via an environment variable or local config file. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 88/100 (assessable for any licence), verifiability 90/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 22/25: The key is supplied by an environment variable or local config file you control. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under Apache-2.0, so anyone can check how the key is handled. Key handling was located in the published source. - Privacy 14/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 7/10: Key handling was found in the published source by BYOAIK's scanner. What was checked (verification tier SOURCE_VERIFIED, derived from this evidence): - [STRONG via SOURCE_SCAN] The user supplies their own OpenAI key: the project references its API key variable. (https://github.com/Future-House/paper-qa/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] The user supplies their own Anthropic key: the project references its API key variable. (https://github.com/Future-House/paper-qa/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] The user supplies their own Google Gemini key: the project references its API key variable. (https://github.com/Future-House/paper-qa/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/Future-House/paper-qa/blob/HEAD/README.md) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-08-12, about 2 month(s) ago. (https://github.com/Future-House/paper-qa/commits) ### Hermes Agent - Page: https://www.byoaik.com/tools/hermes-agent/ - Website: https://hermes-agent.nousresearch.com/ - Source: https://github.com/NousResearch/hermes-agent - Category: Agent Frameworks | Pricing: OPEN_SOURCE | Open source: yes (MIT) | Trust score: 90/100 - Platforms: CLI, Self-hosted, Docker, Local-only - Bring-your-own-key providers: OpenAI, Anthropic, Google Gemini, OpenRouter, Ollama, Custom / OpenAI-compatible - Key storage: ENVIRONMENT_VARIABLE | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit today) Hermes is a coding and general-purpose agent from Nous Research that is explicit about not locking you in: use Nous Portal, OpenRouter, OpenAI or your own endpoint, and switch with one command and no code changes. It runs across seven terminal backends from local and Docker through SSH to serverless sandboxes, so the same agent works on a laptop or a GPU cluster. Nous Portal is the company's own subscription bundling model, search, image, speech and browser keys under one bill, which is a convenience option beside bring-your-own-key rather than a requirement. MIT. Key handling: Your API key is supplied via an environment variable or local config file. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 88/100 (assessable for any licence), verifiability 90/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 22/25: The key is supplied by an environment variable or local config file you control. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under MIT, so anyone can check how the key is handled. Key handling was located in the published source. - Privacy 14/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 7/10: Key handling was found in the published source by BYOAIK's scanner. What was checked (verification tier SOURCE_VERIFIED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [STRONG via SOURCE_SCAN] The user supplies their own Google Gemini key: the project references its API key variable. (https://github.com/NousResearch/hermes-agent/blob/HEAD/.env.example) - [STRONG via SOURCE_SCAN] The user supplies their own OpenRouter key: the project references its API key variable. (https://github.com/NousResearch/hermes-agent/blob/HEAD/.env.example) - [STRONG via SOURCE_SCAN] The user supplies their own Groq key: the project references its API key variable. (https://github.com/NousResearch/hermes-agent/blob/HEAD/.env.example) - [STRONG via SOURCE_SCAN] The user supplies their own Hugging Face key: the project references its API key variable. (https://github.com/NousResearch/hermes-agent/blob/HEAD/.env.example) - [STRONG via SOURCE_SCAN] The user supplies their own Fireworks key: the project references its API key variable. (https://github.com/NousResearch/hermes-agent/blob/HEAD/.env.example) - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/NousResearch/hermes-agent/blob/HEAD/.env.example) - [CONFIRMED via SOURCE_SCAN] Ships a container definition, so it can be self-hosted on your own infrastructure. (https://github.com/NousResearch/hermes-agent/blob/HEAD/docker-compose.yml) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-29, about 0 month(s) ago. (https://github.com/NousResearch/hermes-agent/commits) ### Open TutorAI - Page: https://www.byoaik.com/tools/open-tutorai/ - Website: https://github.com/Open-TutorAi/open-tutor-ai-CE - Source: https://github.com/Open-TutorAi/open-tutor-ai-CE - Category: Education Tools | Pricing: OPEN_SOURCE | Open source: yes (BSD-3-Clause) | Trust score: 90/100 - Platforms: Web, Self-hosted, Docker - Bring-your-own-key providers: OpenAI, Mistral, Groq, OpenRouter, LM Studio, Ollama, Custom / OpenAI-compatible - Key storage: SELF_HOSTED_ONLY | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Slowing down (last commit about 4 months ago) Open TutorAI Community Edition is a self-hosted platform for personalised tutoring built on large language models. It connects to Ollama and any OpenAI-compatible API, with documented support for LM Studio, GroqCloud, Mistral and OpenRouter, and a model builder for creating custom agents. Deployment is Docker-first with dedicated Ollama and CUDA images, so both the platform and the model can run on hardware you control. BSD-3-Clause. Key handling: You self-host the app, so your API key lives in your own deployment. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 92/100 (assessable for any licence), verifiability 90/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 21/25: The key lives in a deployment you run yourself. - Request Routing 20/20: Requests go straight from you to the AI provider. The provider endpoint is configurable, so you can point it at the provider or your own gateway. - Transparency 20/20: Source is public under BSD-3-Clause, so anyone can check how the key is handled. Key handling was located in the published source. - Privacy 14/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 8/10: Commits within the last six months. - Verification Confidence 7/10: Key handling was found in the published source by BYOAIK's scanner. What was checked (verification tier SOURCE_VERIFIED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [STRONG via SOURCE_SCAN] The user supplies their own OpenAI key: the project references its API key variable. (https://github.com/Open-TutorAi/open-tutor-ai-CE/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] The user supplies their own Google Gemini key: the project references its API key variable. (https://github.com/Open-TutorAi/open-tutor-ai-CE/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/Open-TutorAi/open-tutor-ai-CE/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] The provider endpoint is configurable, so requests can be pointed straight at the provider or at your own gateway. (https://github.com/Open-TutorAi/open-tutor-ai-CE/blob/HEAD/README.md) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-06-26, about 3 month(s) ago. (https://github.com/Open-TutorAi/open-tutor-ai-CE/commits) ### Marker - Page: https://www.byoaik.com/tools/marker/ - Website: https://github.com/datalab-to/marker - Source: https://github.com/datalab-to/marker - Category: Document Analysis | Pricing: OPEN_SOURCE | Open source: yes (Apache-2.0) | Trust score: 90/100 - Platforms: CLI, Local-only, Self-hosted, Docker - Bring-your-own-key providers: OpenAI, Anthropic, Google Gemini, Azure OpenAI, OpenRouter, Ollama, Custom / OpenAI-compatible - Key storage: ENVIRONMENT_VARIABLE | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit 10 days ago) Marker converts PDFs, documents and images to markdown, running layout, OCR and table recognition through a local inference server it spawns itself on vLLM or llama.cpp. Accuracy can optionally be boosted by passing --use_llm, which the README documents as working with Gemini, Claude, OpenAI-compatible endpoints, Azure, Vertex, OpenRouter or Ollama, all on credentials you supply. The default path needs no cloud key, which matters for documents you would rather not upload. Apache-2.0. Key handling: Your API key is supplied via an environment variable or local config file. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 88/100 (assessable for any licence), verifiability 90/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 22/25: The key is supplied by an environment variable or local config file you control. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under Apache-2.0, so anyone can check how the key is handled. Key handling was located in the published source. - Privacy 14/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 7/10: Key handling was found in the published source by BYOAIK's scanner. What was checked (verification tier SOURCE_VERIFIED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [STRONG via SOURCE_SCAN] The user supplies their own OpenRouter key: the project references its API key variable. (https://github.com/datalab-to/marker/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/datalab-to/marker/blob/HEAD/README.md) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-13, about 1 month(s) ago. (https://github.com/datalab-to/marker/commits) ### BlogPilot - Page: https://www.byoaik.com/tools/blogpilot/ - Website: https://github.com/IamRamgarhia/BlogPilot-Open-Source-AI-SEO-Content-Studio - Source: https://github.com/IamRamgarhia/BlogPilot-Open-Source-AI-SEO-Content-Studio - Category: Marketing and SEO Tools | Pricing: OPEN_SOURCE | Open source: yes (MIT) | Trust score: 90/100 - Platforms: Web, Self-hosted, Docker - Bring-your-own-key providers: OpenAI, Anthropic, Google Gemini, Mistral, DeepSeek, Groq, OpenRouter, Together AI, Ollama, Custom / OpenAI-compatible - Key storage: SELF_HOSTED_ONLY | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Maintained (last commit about 3 months ago) BlogPilot is a self-hostable SEO content studio for writing, keyword research, technical site audits and topic authority scoring, positioned against Surfer, Ahrefs, Clearscope and MarketMuse. It runs on one AI key that you bring, across twelve or more providers with automatic failover, and the README notes a free Google Gemini key is enough to operate it. Several features need no AI key at all, including keyword research from Google Autocomplete and People-Also-Ask, and CrUX field data via PageSpeed Insights. Client data stays in a single SQLite file you hold, and it runs on a laptop, a free Vercel tier or Docker. MIT licensed. Key handling: You self-host the app, so your API key lives in your own deployment. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 88/100 (assessable for any licence), verifiability 90/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 21/25: The key lives in a deployment you run yourself. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under MIT, so anyone can check how the key is handled. Key handling was located in the published source. - Privacy 15/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. The project states it sends no telemetry. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 7/10: Key handling was found in the published source by BYOAIK's scanner. What was checked (verification tier SOURCE_VERIFIED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [STRONG via SOURCE_SCAN] The user supplies their own Google Gemini key: the project references its API key variable. (https://github.com/IamRamgarhia/BlogPilot-Open-Source-AI-SEO-Content-Studio/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/IamRamgarhia/BlogPilot-Open-Source-AI-SEO-Content-Studio/blob/HEAD/README.md) - [REPORTED via SOURCE_SCAN] The project states it sends no telemetry. (https://github.com/IamRamgarhia/BlogPilot-Open-Source-AI-SEO-Content-Studio/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] The user supplies their own OpenAI key: the project references its API key variable. (https://github.com/IamRamgarhia/BlogPilot-Open-Source-AI-SEO-Content-Studio/blob/HEAD/.env.example) - [STRONG via SOURCE_SCAN] The user supplies their own Anthropic key: the project references its API key variable. (https://github.com/IamRamgarhia/BlogPilot-Open-Source-AI-SEO-Content-Studio/blob/HEAD/.env.example) - [STRONG via SOURCE_SCAN] The user supplies their own OpenRouter key: the project references its API key variable. (https://github.com/IamRamgarhia/BlogPilot-Open-Source-AI-SEO-Content-Studio/blob/HEAD/.env.example) - [STRONG via SOURCE_SCAN] The user supplies their own Groq key: the project references its API key variable. (https://github.com/IamRamgarhia/BlogPilot-Open-Source-AI-SEO-Content-Studio/blob/HEAD/.env.example) - [STRONG via SOURCE_SCAN] The user supplies their own Mistral key: the project references its API key variable. (https://github.com/IamRamgarhia/BlogPilot-Open-Source-AI-SEO-Content-Studio/blob/HEAD/.env.example) - [STRONG via SOURCE_SCAN] The user supplies their own DeepSeek key: the project references its API key variable. (https://github.com/IamRamgarhia/BlogPilot-Open-Source-AI-SEO-Content-Studio/blob/HEAD/.env.example) - [STRONG via SOURCE_SCAN] The user supplies their own Together AI key: the project references its API key variable. (https://github.com/IamRamgarhia/BlogPilot-Open-Source-AI-SEO-Content-Studio/blob/HEAD/.env.example) - [STRONG via SOURCE_SCAN] The user supplies their own Perplexity key: the project references its API key variable. (https://github.com/IamRamgarhia/BlogPilot-Open-Source-AI-SEO-Content-Studio/blob/HEAD/.env.example) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-08-15, about 2 month(s) ago. (https://github.com/IamRamgarhia/BlogPilot-Open-Source-AI-SEO-Content-Studio/commits) ### smolagents - Page: https://www.byoaik.com/tools/smolagents/ - Website: https://huggingface.co/docs/smolagents - Source: https://github.com/huggingface/smolagents - Category: Agent Frameworks | Pricing: OPEN_SOURCE | Open source: yes (Apache-2.0) | Trust score: 90/100 - Platforms: CLI, Self-hosted, Local-only, Docker - Bring-your-own-key providers: OpenAI, Anthropic, Hugging Face, Ollama, Custom / OpenAI-compatible - Key storage: ENVIRONMENT_VARIABLE | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Maintained (last commit about 1 month ago) smolagents is Hugging Face's minimal agent library, built around agents that write and run code rather than emit rigid tool calls, with the whole abstraction kept small enough to read in an afternoon. It is model-agnostic across hosted providers, Hugging Face inference and local models, so the credential and the endpoint are the caller's throughout. The small surface is the point: less framework between you and what actually gets sent. Apache-2.0. Key handling: Your API key is supplied via an environment variable or local config file. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 88/100 (assessable for any licence), verifiability 90/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 22/25: The key is supplied by an environment variable or local config file you control. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under Apache-2.0, so anyone can check how the key is handled. Key handling was located in the published source. - Privacy 14/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 7/10: Key handling was found in the published source by BYOAIK's scanner. What was checked (verification tier SOURCE_VERIFIED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [STRONG via SOURCE_SCAN] The user supplies their own Anthropic key: the project references its API key variable. (https://github.com/huggingface/smolagents/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] The user supplies their own OpenRouter key: the project references its API key variable. (https://github.com/huggingface/smolagents/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] The user supplies their own Azure OpenAI key: the project references its API key variable. (https://github.com/huggingface/smolagents/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] The user supplies their own Together AI key: the project references its API key variable. (https://github.com/huggingface/smolagents/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] Reads the API key from an environment variable at runtime. (https://github.com/huggingface/smolagents/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/huggingface/smolagents/blob/HEAD/README.md) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-23, about 0 month(s) ago. (https://github.com/huggingface/smolagents/commits) ### LightRAG - Page: https://www.byoaik.com/tools/lightrag/ - Website: https://github.com/HKUDS/LightRAG - Source: https://github.com/HKUDS/LightRAG - Category: Knowledge Management | Pricing: OPEN_SOURCE | Open source: yes (MIT) | Trust score: 90/100 - Platforms: CLI, Web, Self-hosted, Docker - Bring-your-own-key providers: OpenAI, Anthropic, Google Gemini, Azure OpenAI, Ollama, Custom / OpenAI-compatible - Key storage: ENVIRONMENT_VARIABLE | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit today) LightRAG builds a knowledge graph over your documents and retrieves across it, which handles questions spanning many sources better than plain vector search. Both the LLM and the embedding model are configurable: OPENAI_API_KEY, Gemini or a local Ollama, meaning the whole pipeline can run with no cloud key at all. That matters more here than in a chat tool, because indexing sends every document you own through a model rather than just the question you asked. MIT. Key handling: Your API key is supplied via an environment variable or local config file. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 88/100 (assessable for any licence), verifiability 90/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 22/25: The key is supplied by an environment variable or local config file you control. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under MIT, so anyone can check how the key is handled. Key handling was located in the published source. - Privacy 14/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 7/10: Key handling was found in the published source by BYOAIK's scanner. What was checked (verification tier SOURCE_VERIFIED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [STRONG via SOURCE_SCAN] The user supplies their own OpenAI key: the project references its API key variable. (https://github.com/HKUDS/LightRAG/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/HKUDS/LightRAG/blob/HEAD/README.md) - [CONFIRMED via SOURCE_SCAN] Ships a container definition, so it can be self-hosted on your own infrastructure. (https://github.com/HKUDS/LightRAG/blob/HEAD/docker-compose.yml) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-26, about 0 month(s) ago. (https://github.com/HKUDS/LightRAG/commits) ### Graphiti - Page: https://www.byoaik.com/tools/graphiti/ - Website: https://help.getzep.com/graphiti - Source: https://github.com/getzep/graphiti - Category: Knowledge Management | Pricing: OPEN_SOURCE | Open source: yes (Apache-2.0) | Trust score: 90/100 - Platforms: CLI, Self-hosted, Docker - Bring-your-own-key providers: OpenAI, Anthropic, Google Gemini, Azure OpenAI, Ollama, Custom / OpenAI-compatible - Key storage: ENVIRONMENT_VARIABLE | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit yesterday) Graphiti builds knowledge graphs that model change over time, so a fact can be superseded rather than silently overwritten and an agent can reason about what was true when. It is the memory layer behind Zep, and runs against OpenAI, Anthropic, Gemini, Ollama or any OpenAI-compatible endpoint on credentials you supply. Zep sells a hosted service built on it, which is a different arrangement from running the library yourself. Apache-2.0. Key handling: Your API key is supplied via an environment variable or local config file. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 88/100 (assessable for any licence), verifiability 90/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 22/25: The key is supplied by an environment variable or local config file you control. - Request Routing 20/20: Requests go straight from you to the AI provider. The provider endpoint is configurable, so you can point it at the provider or your own gateway. - Transparency 20/20: Source is public under Apache-2.0, so anyone can check how the key is handled. Key handling was located in the published source. - Privacy 11/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. Analytics or error-reporting libraries are present; what they send was not established. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 7/10: Key handling was found in the published source by BYOAIK's scanner. What was checked (verification tier SOURCE_VERIFIED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [STRONG via SOURCE_SCAN] The user supplies their own OpenAI key: the project references its API key variable. (https://github.com/getzep/graphiti/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/getzep/graphiti/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] The provider endpoint is configurable, so requests can be pointed straight at the provider or at your own gateway. (https://github.com/getzep/graphiti/blob/HEAD/README.md) - [REPORTED via SOURCE_SCAN] References PostHog, so some analytics or error reporting is present. This scan cannot tell whether it is opt-in or what it sends. (https://github.com/getzep/graphiti/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] The user supplies their own Anthropic key: the project references its API key variable. (https://github.com/getzep/graphiti/blob/HEAD/.env.example) - [CONFIRMED via SOURCE_SCAN] Ships a container definition, so it can be self-hosted on your own infrastructure. (https://github.com/getzep/graphiti/blob/HEAD/docker-compose.yml) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-29, about 0 month(s) ago. (https://github.com/getzep/graphiti/commits) ### DeepEval - Page: https://www.byoaik.com/tools/deepeval/ - Website: https://github.com/confident-ai/deepeval - Source: https://github.com/confident-ai/deepeval - Category: Prompt Engineering Tools | Pricing: OPEN_SOURCE | Open source: yes (Apache-2.0) | Trust score: 90/100 - Platforms: CLI, Local-only, Self-hosted, Docker - Bring-your-own-key providers: OpenAI, Anthropic, Google Gemini, Ollama, Custom / OpenAI-compatible - Key storage: ENVIRONMENT_VARIABLE | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit yesterday) DeepEval is an evaluation framework for LLM applications, structured like Pytest, covering metrics such as G-Eval, answer relevancy, hallucination and task completion. Evaluations are driven by LLM-as-a-judge and NLP models that the README states run locally on your machine, with judge models reached through credentials you supply. Worth noting for anyone evaluating sensitive prompts: an eval harness necessarily sees the inputs and outputs it is scoring. Apache-2.0. Key handling: Your API key is supplied via an environment variable or local config file. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 88/100 (assessable for any licence), verifiability 90/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 22/25: The key is supplied by an environment variable or local config file you control. - Request Routing 20/20: Requests go straight from you to the AI provider. The provider endpoint is configurable, so you can point it at the provider or your own gateway. - Transparency 20/20: Source is public under Apache-2.0, so anyone can check how the key is handled. Key handling was located in the published source. - Privacy 11/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. Analytics or error-reporting libraries are present; what they send was not established. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 7/10: Key handling was found in the published source by BYOAIK's scanner. What was checked (verification tier SOURCE_VERIFIED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [STRONG via SOURCE_SCAN] The user supplies their own OpenAI key: the project references its API key variable. (https://github.com/confident-ai/deepeval/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] The user supplies their own Anthropic key: the project references its API key variable. (https://github.com/confident-ai/deepeval/blob/HEAD/.env.example) - [STRONG via SOURCE_SCAN] The user supplies their own OpenRouter key: the project references its API key variable. (https://github.com/confident-ai/deepeval/blob/HEAD/.env.example) - [STRONG via SOURCE_SCAN] The user supplies their own DeepSeek key: the project references its API key variable. (https://github.com/confident-ai/deepeval/blob/HEAD/.env.example) - [STRONG via SOURCE_SCAN] The user supplies their own Azure OpenAI key: the project references its API key variable. (https://github.com/confident-ai/deepeval/blob/HEAD/.env.example) - [STRONG via SOURCE_SCAN] The user supplies their own xAI Grok key: the project references its API key variable. (https://github.com/confident-ai/deepeval/blob/HEAD/.env.example) - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/confident-ai/deepeval/blob/HEAD/.env.example) - [STRONG via SOURCE_SCAN] The provider endpoint is configurable, so requests can be pointed straight at the provider or at your own gateway. (https://github.com/confident-ai/deepeval/blob/HEAD/.env.example) - [REPORTED via SOURCE_SCAN] References PostHog, so some analytics or error reporting is present. This scan cannot tell whether it is opt-in or what it sends. (https://github.com/confident-ai/deepeval/blob/HEAD/pyproject.toml) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-29, about 0 month(s) ago. (https://github.com/confident-ai/deepeval/commits) ### Crawl4AI - Page: https://www.byoaik.com/tools/crawl4ai/ - Website: https://github.com/unclecode/crawl4ai - Source: https://github.com/unclecode/crawl4ai - Category: Research Tools | Pricing: OPEN_SOURCE | Open source: yes (Apache-2.0) | Trust score: 90/100 - Platforms: CLI, Self-hosted, Local-only, Docker - Bring-your-own-key providers: OpenAI, Anthropic, Google Gemini, Ollama, Custom / OpenAI-compatible - Key storage: ENVIRONMENT_VARIABLE | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit 4 days ago) Crawl4AI turns web pages into clean, LLM-ready markdown, with extraction strategies that can optionally call a model using your own credentials. It runs as a Python library on your own machine, so pages are fetched by you rather than by a scraping service that would see every URL you are interested in and bill you per page. Apache-2.0, and one of the most-starred projects in this directory. Key handling: Your API key is supplied via an environment variable or local config file. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 88/100 (assessable for any licence), verifiability 90/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 22/25: The key is supplied by an environment variable or local config file you control. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under Apache-2.0, so anyone can check how the key is handled. Key handling was located in the published source. - Privacy 14/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 7/10: Key handling was found in the published source by BYOAIK's scanner. What was checked (verification tier SOURCE_VERIFIED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [STRONG via SOURCE_SCAN] The user supplies their own OpenAI key: the project references its API key variable. (https://github.com/unclecode/crawl4ai/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] Reads the API key from an environment variable at runtime. (https://github.com/unclecode/crawl4ai/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/unclecode/crawl4ai/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] The user supplies their own Anthropic key: the project references its API key variable. (https://github.com/unclecode/crawl4ai/blob/HEAD/docker-compose.yml) - [STRONG via SOURCE_SCAN] The user supplies their own Google Gemini key: the project references its API key variable. (https://github.com/unclecode/crawl4ai/blob/HEAD/docker-compose.yml) - [STRONG via SOURCE_SCAN] The user supplies their own Groq key: the project references its API key variable. (https://github.com/unclecode/crawl4ai/blob/HEAD/docker-compose.yml) - [STRONG via SOURCE_SCAN] The user supplies their own Mistral key: the project references its API key variable. (https://github.com/unclecode/crawl4ai/blob/HEAD/docker-compose.yml) - [STRONG via SOURCE_SCAN] The user supplies their own DeepSeek key: the project references its API key variable. (https://github.com/unclecode/crawl4ai/blob/HEAD/docker-compose.yml) - [STRONG via SOURCE_SCAN] The user supplies their own Together AI key: the project references its API key variable. (https://github.com/unclecode/crawl4ai/blob/HEAD/docker-compose.yml) - [CONFIRMED via SOURCE_SCAN] Ships a container definition, so it can be self-hosted on your own infrastructure. (https://github.com/unclecode/crawl4ai/blob/HEAD/docker-compose.yml) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-25, about 0 month(s) ago. (https://github.com/unclecode/crawl4ai/commits) ### OpenAI Agents SDK - Page: https://www.byoaik.com/tools/openai-agents-sdk/ - Website: https://openai.github.io/openai-agents-python/ - Source: https://github.com/openai/openai-agents-python - Category: Agent Frameworks | Pricing: OPEN_SOURCE | Open source: yes (MIT) | Trust score: 90/100 - Platforms: CLI, Self-hosted, Local-only, Docker - Bring-your-own-key providers: OpenAI, Azure OpenAI, Ollama, Custom / OpenAI-compatible - Key storage: ENVIRONMENT_VARIABLE | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit today) The Agents SDK is OpenAI's lightweight framework for agents, handoffs, guardrails and tracing, and the successor to the Swarm experiment. Despite the publisher it is not locked to OpenAI: it drives any OpenAI-compatible endpoint, which in practice means local servers and other providers work too. Tracing is built in and is worth understanding before enabling, since traces carry the content of runs. MIT. Key handling: Your API key is supplied via an environment variable or local config file. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 88/100 (assessable for any licence), verifiability 90/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 22/25: The key is supplied by an environment variable or local config file you control. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under MIT, so anyone can check how the key is handled. Key handling was located in the published source. - Privacy 14/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 7/10: Key handling was found in the published source by BYOAIK's scanner. What was checked (verification tier SOURCE_VERIFIED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [STRONG via SOURCE_SCAN] The user supplies their own OpenAI key: the project references its API key variable. (https://github.com/openai/openai-agents-python/blob/HEAD/README.md) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-29, about 0 month(s) ago. (https://github.com/openai/openai-agents-python/commits) ### OpenTutor - Page: https://www.byoaik.com/tools/opentutor/ - Website: https://github.com/zijinz456/OpenTutor - Source: https://github.com/zijinz456/OpenTutor - Category: Education Tools | Pricing: OPEN_SOURCE | Open source: yes (MIT) | Trust score: 89/100 - Platforms: Web, Self-hosted, Docker - Bring-your-own-key providers: OpenAI, Anthropic, Google Gemini, DeepSeek, Groq, OpenRouter, LM Studio, Ollama, Custom / OpenAI-compatible - Key storage: SELF_HOSTED_ONLY | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit today) OpenTutor is a self-hosted, local-first learning workspace. You upload course material and it generates structured notes, quizzes, flashcards and an adaptive tutor. The LLM provider is chosen with an LLM_PROVIDER environment variable and the matching key (for example DEEPSEEK_API_KEY), across ten or more providers including OpenAI, Anthropic, Gemini, DeepSeek, Groq, OpenRouter, LM Studio and Ollama. Running it against Ollama needs no cloud key at all, in which case nothing leaves the machine. MIT licensed. Key handling: You self-host the app, so your API key lives in your own deployment. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 87/100 (assessable for any licence), verifiability 90/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 21/25: The key lives in a deployment you run yourself. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under MIT, so anyone can check how the key is handled. Key handling was located in the published source. - Privacy 14/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 7/10: Key handling was found in the published source by BYOAIK's scanner. What was checked (verification tier SOURCE_VERIFIED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [STRONG via SOURCE_SCAN] The user supplies their own DeepSeek key: the project references its API key variable. (https://github.com/zijinz456/OpenTutor/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/zijinz456/OpenTutor/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] The user supplies their own OpenAI key: the project references its API key variable. (https://github.com/zijinz456/OpenTutor/blob/HEAD/.env.example) - [STRONG via SOURCE_SCAN] The user supplies their own Anthropic key: the project references its API key variable. (https://github.com/zijinz456/OpenTutor/blob/HEAD/.env.example) - [STRONG via SOURCE_SCAN] The user supplies their own Google Gemini key: the project references its API key variable. (https://github.com/zijinz456/OpenTutor/blob/HEAD/.env.example) - [STRONG via SOURCE_SCAN] The user supplies their own OpenRouter key: the project references its API key variable. (https://github.com/zijinz456/OpenTutor/blob/HEAD/.env.example) - [STRONG via SOURCE_SCAN] The user supplies their own Groq key: the project references its API key variable. (https://github.com/zijinz456/OpenTutor/blob/HEAD/.env.example) - [CONFIRMED via SOURCE_SCAN] Ships a container definition, so it can be self-hosted on your own infrastructure. (https://github.com/zijinz456/OpenTutor/blob/HEAD/docker-compose.yml) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-29, about 0 month(s) ago. (https://github.com/zijinz456/OpenTutor/commits) ### gptel - Page: https://www.byoaik.com/tools/gptel/ - Website: https://github.com/karthink/gptel - Source: https://github.com/karthink/gptel - Category: Coding Assistants | Pricing: OPEN_SOURCE | Open source: yes (GPL-3.0) | Trust score: 89/100 - Platforms: Library/SDK, Self-hosted, Docker - Bring-your-own-key providers: OpenAI, Anthropic, Google Gemini, Azure OpenAI, OpenRouter, Groq, Mistral, Perplexity, DeepSeek, xAI Grok, Ollama, Together AI, Custom / OpenAI-compatible - Key storage: SELF_HOSTED_ONLY | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit 4 days ago) gptel is an open-source LLM client for Emacs that lets you chat with and use models from any buffer, with rewrite/refactor commands, dedicated chat buffers, multimodal context (files/images), tool-use, and MCP support. It is bring-your-own-key: users supply their own provider API key by setting gptel-api-key directly, pointing it at a function, or reading it from ~/.authinfo or environment variables. It talks to most major providers and any OpenAI-compatible endpoint via gptel-make-openai with custom host/endpoint, and supports fully local runners (Ollama, Llama.cpp, GPT4All). Actively maintained, GPL-3.0, ~3.4k GitHub stars. Key handling: You self-host the app, so your API key lives in your own deployment. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 87/100 (assessable for any licence), verifiability 90/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 21/25: The key lives in a deployment you run yourself. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under GPL-3.0, so anyone can check how the key is handled. Key handling was located in the published source. - Privacy 14/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 7/10: Key handling was found in the published source by BYOAIK's scanner. What was checked (verification tier SOURCE_VERIFIED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [STRONG via SOURCE_SCAN] The user supplies their own OpenAI key: the project references its API key variable. (https://github.com/karthink/gptel/blob/HEAD/README.org) - [STRONG via SOURCE_SCAN] The user supplies their own Anthropic key: the project references its API key variable. (https://github.com/karthink/gptel/blob/HEAD/README.org) - [STRONG via SOURCE_SCAN] The user supplies their own Google Gemini key: the project references its API key variable. (https://github.com/karthink/gptel/blob/HEAD/README.org) - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/karthink/gptel/blob/HEAD/README.org) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-25, about 0 month(s) ago. (https://github.com/karthink/gptel/commits) ### Open Notebook - Page: https://www.byoaik.com/tools/open-notebook/ - Website: https://www.open-notebook.ai/ - Source: https://github.com/lfnovo/open-notebook - Category: Research Tools | Pricing: OPEN_SOURCE | Open source: yes (MIT) | Trust score: 89/100 - Platforms: Web, Self-hosted, Docker - Bring-your-own-key providers: OpenAI, Anthropic, Google Gemini, Azure OpenAI, OpenRouter, Groq, Mistral, DeepSeek, xAI Grok, Ollama, Custom / OpenAI-compatible - Key storage: SELF_HOSTED_ONLY | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit today) Open Notebook is a self-hosted, MIT-licensed alternative to Google's NotebookLM. You add your own sources (documents, links, media), then chat over them with context-aware AI, run vector search, and generate multi-speaker audio podcasts with configurable per-speaker voices and personalities. It is fully BYOK: keys are entered in-app via Settings -> API Keys -> Add Credential, where you pick a provider, paste your key, run Test Connection, then Discover/Register Models. Thanks to its "Esperanto" provider-abstraction layer it supports 18+ providers (OpenAI, Anthropic, Google Gemini, Mistral, Groq, xAI, DeepSeek, Ollama/LM Studio local, OpenRouter, Azure OpenAI, Vertex AI, plus TTS/STT services like ElevenLabs and Deepgram) and any OpenAI-compatible endpoint via a configurable base URL. It runs locally through Docker (default at localhost:8502) and exposes a REST API. Key handling: You self-host the app, so your API key lives in your own deployment. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 87/100 (assessable for any licence), verifiability 90/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 21/25: The key lives in a deployment you run yourself. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under MIT, so anyone can check how the key is handled. Key handling was located in the published source. - Privacy 14/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 7/10: Key handling was found in the published source by BYOAIK's scanner. What was checked (verification tier SOURCE_VERIFIED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/lfnovo/open-notebook/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] The user supplies their own OpenAI key: the project references its API key variable. (https://github.com/lfnovo/open-notebook/blob/HEAD/.env.example) - [STRONG via SOURCE_SCAN] The user supplies their own Anthropic key: the project references its API key variable. (https://github.com/lfnovo/open-notebook/blob/HEAD/.env.example) - [STRONG via SOURCE_SCAN] The user supplies their own Groq key: the project references its API key variable. (https://github.com/lfnovo/open-notebook/blob/HEAD/.env.example) - [CONFIRMED via SOURCE_SCAN] Ships a container definition, so it can be self-hosted on your own infrastructure. (https://github.com/lfnovo/open-notebook/blob/HEAD/docker-compose.yml) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-12, about 1 month(s) ago. (https://github.com/lfnovo/open-notebook/commits) ### text-generation-webui - Page: https://www.byoaik.com/tools/text-generation-webui/ - Website: https://github.com/oobabooga/textgen - Source: https://github.com/oobabooga/textgen - Category: Local AI Tools | Pricing: OPEN_SOURCE | Open source: yes (AGPL-3.0) | Trust score: 88/100 - Platforms: Desktop, Self-hosted, Docker - Bring-your-own-key providers: Ollama, Custom / OpenAI-compatible, Hugging Face - Key storage: NO_KEY_LOCAL_MODEL | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Maintained (last commit about 3 months ago) text-generation-webui, by oobabooga, is a desktop application for chatting with local models, supporting llama.cpp, Transformers, ExLlamaV3 and TensorRT-LLM backends along with tool calling, vision and web search. You place model files in a local folder and launch the portable build; the README states it runs fully offline and private with zero telemetry, external resources or remote update requests, so no key is ever entered. Portable builds ship for Windows, macOS and Linux with CUDA, Vulkan, ROCm and CPU-only variants. AGPL-3.0, 47,551 stars. Note the repository was renamed from text-generation-webui to oobabooga/textgen. Key handling: Key handling varies. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 95/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 25/25: No cloud key is needed at all. The model runs locally. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under AGPL-3.0, so anyone can check how the key is handled. - Privacy 15/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. The project states it sends no telemetry. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/oobabooga/textgen/blob/HEAD/README.md) - [REPORTED via SOURCE_SCAN] The project states it sends no telemetry. (https://github.com/oobabooga/textgen/blob/HEAD/README.md) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-08-17, about 1 month(s) ago. (https://github.com/oobabooga/textgen/commits) ### AutoGen - Page: https://www.byoaik.com/tools/autogen/ - Website: https://microsoft.github.io/autogen/ - Source: https://github.com/microsoft/autogen - Category: Agent Frameworks | Pricing: OPEN_SOURCE | Open source: yes (MIT) | Trust score: 88/100 - Platforms: Library/SDK, Self-hosted, Docker - Bring-your-own-key providers: OpenAI, Anthropic, Azure OpenAI, Google Gemini, Custom / OpenAI-compatible, Ollama - Key storage: ENVIRONMENT_VARIABLE | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit 11 days ago) AutoGen is Microsoft's programming framework for building multi-agent AI applications and coding agents that converse, use tools and execute code. Model clients such as OpenAIChatCompletionClient accept an api_key argument directly, or read it automatically from OPENAI_API_KEY, AZURE_OPENAI_API_KEY or GEMINI_API_KEY environment variables, with no Microsoft-hosted key or account required. The code is MIT licensed (LICENSE-CODE); GitHub reports CC-BY-4.0 at the top level because that license covers the accompanying research and docs. Key handling: Your API key is supplied via an environment variable or local config file. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 88/100 (assessable for any licence), verifiability 90/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 22/25: The key is supplied by an environment variable or local config file you control. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under MIT, so anyone can check how the key is handled. Key handling was located in the published source. - Privacy 14/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 8/10: Commits within the last six months. - Verification Confidence 7/10: Key handling was found in the published source by BYOAIK's scanner. What was checked (verification tier SOURCE_VERIFIED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [STRONG via SOURCE_SCAN] The user supplies their own OpenAI key: the project references its API key variable. (https://github.com/microsoft/autogen/blob/HEAD/README.md) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-04-06, about 6 month(s) ago. (https://github.com/microsoft/autogen/commits) ### screenshot-to-code - Page: https://www.byoaik.com/tools/screenshot-to-code/ - Website: https://screenshottocode.com - Source: https://github.com/abi/screenshot-to-code - Category: App Builders | Pricing: OPEN_SOURCE | Open source: yes (MIT) | Trust score: 88/100 - Platforms: Web, Self-hosted, Docker - Bring-your-own-key providers: OpenAI, Anthropic, Google Gemini - Key storage: SELF_HOSTED_ONLY | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit 30 days ago) screenshot-to-code converts screenshots, mockups, Figma exports and screen recordings into working front-end code across HTML with Tailwind, React, Vue, Bootstrap and Ionic. The README explains that users supply their own provider credentials either through the settings dialog in the frontend or environment variables in the backend .env file, covering OpenAI, Anthropic and Gemini, and that supplying more keys lets it pick a stronger mix of models per variant. Self-hostable through Docker or a manual frontend and backend setup. MIT licensed, 74,038 stars, pushed the day before review. Key handling: You self-host the app, so your API key lives in your own deployment. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 85/100 (assessable for any licence), verifiability 90/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 21/25: The key lives in a deployment you run yourself. - Request Routing 20/20: Requests go straight from you to the AI provider. The provider endpoint is configurable, so you can point it at the provider or your own gateway. - Transparency 20/20: Source is public under MIT, so anyone can check how the key is handled. Key handling was located in the published source. - Privacy 10/15: Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 7/10: Key handling was found in the published source by BYOAIK's scanner. What was checked (verification tier SOURCE_VERIFIED, derived from this evidence): - [STRONG via SOURCE_SCAN] The user supplies their own OpenAI key: the project references its API key variable. (https://github.com/abi/screenshot-to-code/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] The user supplies their own Anthropic key: the project references its API key variable. (https://github.com/abi/screenshot-to-code/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] The user supplies their own Google Gemini key: the project references its API key variable. (https://github.com/abi/screenshot-to-code/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/abi/screenshot-to-code/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] The provider endpoint is configurable, so requests can be pointed straight at the provider or at your own gateway. (https://github.com/abi/screenshot-to-code/blob/HEAD/README.md) - [CONFIRMED via SOURCE_SCAN] Ships a container definition, so it can be self-hosted on your own infrastructure. (https://github.com/abi/screenshot-to-code/blob/HEAD/docker-compose.yml) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-07-30, about 2 month(s) ago. (https://github.com/abi/screenshot-to-code/commits) ### MetaGPT - Page: https://www.byoaik.com/tools/metagpt/ - Website: https://github.com/FoundationAgents/MetaGPT - Source: https://github.com/FoundationAgents/MetaGPT - Category: Agent Frameworks | Pricing: OPEN_SOURCE | Open source: yes (MIT) | Trust score: 88/100 - Platforms: CLI, Self-hosted, Library/SDK, Docker - Bring-your-own-key providers: OpenAI, Azure OpenAI, Custom / OpenAI-compatible, Ollama - Key storage: ENVIRONMENT_VARIABLE | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Stale (last commit about 10 months ago) MetaGPT assigns SOP-based roles (product manager, architect, engineer, QA) to LLM agents that collaborate to design and write software from a natural-language specification. Configuration lives in config/config2.yaml, where the user fills in an api_key value for the LLM, embedding and search sections, so the framework never proxies through a MetaGPT-owned key. Supports OpenAI, Azure and OpenAI-compatible or local endpoints. MIT licensed with nearly 70,000 stars. Key handling: Your API key is supplied via an environment variable or local config file. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 93/100 (assessable for any licence), verifiability 90/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 22/25: The key is supplied by an environment variable or local config file you control. - Request Routing 20/20: Requests go straight from you to the AI provider. The provider endpoint is configurable, so you can point it at the provider or your own gateway. - Transparency 20/20: Source is public under MIT, so anyone can check how the key is handled. - Privacy 14/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 5/10: Last commit was within the last year. - Verification Confidence 7/10: Key handling was found in the published source by BYOAIK's scanner. What was checked (verification tier SOURCE_VERIFIED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/FoundationAgents/MetaGPT/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] The provider endpoint is configurable, so requests can be pointed straight at the provider or at your own gateway. (https://github.com/FoundationAgents/MetaGPT/blob/HEAD/README.md) - [CONFIRMED via SOURCE_SCAN] Ships a container definition, so it can be self-hosted on your own infrastructure. (https://github.com/FoundationAgents/MetaGPT/blob/HEAD/Dockerfile) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-01-21, about 8 month(s) ago. (https://github.com/FoundationAgents/MetaGPT/commits) ### ChainForge - Page: https://www.byoaik.com/tools/chainforge/ - Website: https://chainforge.ai - Source: https://github.com/ianarawjo/ChainForge - Category: Prompt Engineering Tools | Pricing: OPEN_SOURCE | Open source: yes (MIT) | Trust score: 88/100 - Platforms: Web, Self-hosted, Docker - Bring-your-own-key providers: OpenAI, Anthropic, Google Gemini, AWS Bedrock, Hugging Face, Custom / OpenAI-compatible - Key storage: SELF_HOSTED_ONLY | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit 10 days ago) ChainForge is a visual programming environment, from an academic HCI project, for battle-testing and comparing prompts side by side across models. You enter your OpenAI, Anthropic, Google or Bedrock credentials in a settings panel, or when running locally the app loads them from your environment, so nothing passes through ChainForge's servers in the self-hosted path. Runs as a local web app. MIT licensed, 3,025 stars, last commit about two months before review. Key handling: You self-host the app, so your API key lives in your own deployment. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 85/100 (assessable for any licence), verifiability 90/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 21/25: The key lives in a deployment you run yourself. - Request Routing 20/20: Requests go straight from you to the AI provider. The provider endpoint is configurable, so you can point it at the provider or your own gateway. - Transparency 20/20: Source is public under MIT, so anyone can check how the key is handled. Key handling was located in the published source. - Privacy 10/15: Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 7/10: Key handling was found in the published source by BYOAIK's scanner. What was checked (verification tier SOURCE_VERIFIED, derived from this evidence): - [STRONG via SOURCE_SCAN] The user supplies their own OpenAI key: the project references its API key variable. (https://github.com/ianarawjo/ChainForge/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/ianarawjo/ChainForge/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] The provider endpoint is configurable, so requests can be pointed straight at the provider or at your own gateway. (https://github.com/ianarawjo/ChainForge/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] The user supplies their own Anthropic key: the project references its API key variable. (https://github.com/ianarawjo/ChainForge/blob/HEAD/docker-compose.yml) - [STRONG via SOURCE_SCAN] The user supplies their own DeepSeek key: the project references its API key variable. (https://github.com/ianarawjo/ChainForge/blob/HEAD/docker-compose.yml) - [STRONG via SOURCE_SCAN] The user supplies their own Cohere key: the project references its API key variable. (https://github.com/ianarawjo/ChainForge/blob/HEAD/docker-compose.yml) - [STRONG via SOURCE_SCAN] The user supplies their own Hugging Face key: the project references its API key variable. (https://github.com/ianarawjo/ChainForge/blob/HEAD/docker-compose.yml) - [CONFIRMED via SOURCE_SCAN] Ships a container definition, so it can be self-hosted on your own infrastructure. (https://github.com/ianarawjo/ChainForge/blob/HEAD/docker-compose.yml) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-27, about 0 month(s) ago. (https://github.com/ianarawjo/ChainForge/commits) ### chatgpt-shell - Page: https://www.byoaik.com/tools/chatgpt-shell/ - Website: https://github.com/xenodium/chatgpt-shell - Source: https://github.com/xenodium/chatgpt-shell - Category: IDE Plugins | Pricing: OPEN_SOURCE | Open source: yes (GPL-3.0) | Trust score: 88/100 - Platforms: Desktop, Local-only - Bring-your-own-key providers: OpenAI, Anthropic, Google Gemini, Perplexity, Ollama, Custom / OpenAI-compatible - Key storage: LOCAL_DEVICE | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Maintained (last commit about 2 months ago) chatgpt-shell gives Emacs a comint-style buffer for talking to models, with Org and Markdown blocks, and covers OpenAI, Anthropic, Gemini, Kagi, Perplexity and Ollama. Keys are configured as values or as a function evaluated at runtime, which is how Emacs users keep secrets in auth-source or a password store rather than in an init file. GPL-3.0. Key handling: Your API key is stored locally on your device. Requests are sent directly to the AI provider. Scores: key handling 85/100 (assessable for any licence), verifiability 90/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 23/25: The key is held on your own device. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under GPL-3.0, so anyone can check how the key is handled. Key handling was located in the published source. - Privacy 11/15: Local-first: it works without sending your data anywhere. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 7/10: Key handling was found in the published source by BYOAIK's scanner. What was checked (verification tier SOURCE_VERIFIED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [STRONG via SOURCE_SCAN] The user supplies their own OpenAI key: the project references its API key variable. (https://github.com/xenodium/chatgpt-shell/blob/HEAD/README.org) - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/xenodium/chatgpt-shell/blob/HEAD/README.org) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-18, about 0 month(s) ago. (https://github.com/xenodium/chatgpt-shell/commits) ### ComfyUI - Page: https://www.byoaik.com/tools/comfyui/ - Website: https://www.comfy.org/ - Source: https://github.com/comfy-org/ComfyUI - Category: Image Generation Tools | Pricing: OPEN_SOURCE | Open source: yes (GPL-3.0) | Trust score: 87/100 - Platforms: Desktop, Self-hosted, Docker - Bring-your-own-key providers: Ollama, Custom / OpenAI-compatible - Key storage: NO_KEY_LOCAL_MODEL | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit today) ComfyUI is a graph-based interface for running diffusion models, Stable Diffusion, SDXL, Flux and video models, with control over every step of the pipeline. It is bring-your-own-compute rather than bring-your-own-key in the usual sense: you download open model weights and run inference locally on your own GPU with no vendor account and no pooled API, so there is no per-image markup to pay. Optional API nodes let you attach your own key for cloud image and video providers if you want them. Runs as a desktop app or self-hosted server on Windows, macOS, Linux and Docker. GPL-3.0, 126,801 stars. Key handling: Key handling varies. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 93/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 25/25: No cloud key is needed at all. The model runs locally. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under GPL-3.0, so anyone can check how the key is handled. - Privacy 14/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-29, about 0 month(s) ago. (https://github.com/comfy-org/ComfyUI/commits) ### Stability Matrix - Page: https://www.byoaik.com/tools/stability-matrix/ - Website: https://github.com/LykosAI/StabilityMatrix - Source: https://github.com/LykosAI/StabilityMatrix - Category: Image Generation Tools | Pricing: OPEN_SOURCE | Open source: yes (AGPL-3.0) | Trust score: 87/100 - Platforms: Desktop, Local-only, Self-hosted, Docker - Bring-your-own-key providers: Ollama, Custom / OpenAI-compatible - Key storage: NO_KEY_LOCAL_MODEL | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit 19 days ago) Stability Matrix installs, updates and manages the local generative UIs from a single dashboard, covering ComfyUI, SD.Next, InvokeAI, Fooocus and others, along with shared model storage so weights are not duplicated per app. It is infrastructure for a local setup rather than a generator itself: nothing is sent anywhere, and it exists to make running these tools on your own hardware less tedious. AGPL-3.0. Key handling: Key handling varies. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 93/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 25/25: No cloud key is needed at all. The model runs locally. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under AGPL-3.0, so anyone can check how the key is handled. - Privacy 14/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-16, about 0 month(s) ago. (https://github.com/LykosAI/StabilityMatrix/commits) ### LocalAI - Page: https://www.byoaik.com/tools/localai/ - Website: https://localai.io - Source: https://github.com/mudler/LocalAI - Category: Self-Hosted AI Platforms | Pricing: OPEN_SOURCE | Open source: yes (MIT) | Trust score: 87/100 - Platforms: Self-hosted, Docker, Library/SDK - Bring-your-own-key providers: Ollama, Custom / OpenAI-compatible, Hugging Face - Key storage: NO_KEY_LOCAL_MODEL | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit today) LocalAI is an open-source, self-hosted server that speaks the OpenAI REST API (and parts of others), backed by dozens of swappable engines including llama.cpp, vLLM and whisper.cpp for text, vision, audio and image generation. You run it through Docker or a native binary on your own GPU or CPU, so there is no cloud key to enter. The inference happens on infrastructure you control. Supports NVIDIA, AMD, Intel and Apple Silicon acceleration and pulls models straight from Hugging Face or the Ollama registry. MIT licensed, 48,492 stars, pushed the day of review. Key handling: Key handling varies. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 93/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 25/25: No cloud key is needed at all. The model runs locally. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under MIT, so anyone can check how the key is handled. - Privacy 14/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/mudler/LocalAI/blob/HEAD/README.md) - [CONFIRMED via SOURCE_SCAN] Ships a container definition, so it can be self-hosted on your own infrastructure. (https://github.com/mudler/LocalAI/blob/HEAD/docker-compose.yaml) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-29, about 0 month(s) ago. (https://github.com/mudler/LocalAI/commits) ### Browser Use - Page: https://www.byoaik.com/tools/browser-use/ - Website: https://browser-use.com/ - Source: https://github.com/browser-use/browser-use - Category: Automation Tools | Pricing: OPEN_SOURCE | Open source: yes (MIT) | Trust score: 87/100 - Platforms: CLI, Self-hosted, Local-only, Docker - Bring-your-own-key providers: OpenAI, Anthropic, Google Gemini, OpenRouter, Ollama, Custom / OpenAI-compatible - Key storage: ENVIRONMENT_VARIABLE | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit 10 days ago) Browser Use lets an AI agent drive a real browser, and is deliberately model-agnostic: you connect whichever provider you hold a key for, or a local model, rather than being tied to a vendor's hosted agent. It runs as a Python library on your own machine or server, so the browser session, the credentials it touches and the model choice are all yours. Worth thinking about before use: a browser agent acts with whatever access your browser profile has, which is a far broader blast radius than a chat tool. MIT, and the most-starred project in this directory. Key handling: Your API key is supplied via an environment variable or local config file. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 83/100 (assessable for any licence), verifiability 90/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 22/25: The key is supplied by an environment variable or local config file you control. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under MIT, so anyone can check how the key is handled. Key handling was located in the published source. - Privacy 11/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. Analytics or error-reporting libraries are present; what they send was not established. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 7/10: Key handling was found in the published source by BYOAIK's scanner. What was checked (verification tier SOURCE_VERIFIED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [STRONG via SOURCE_SCAN] The user supplies their own OpenAI key: the project references its API key variable. (https://github.com/browser-use/browser-use/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/browser-use/browser-use/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] The user supplies their own Anthropic key: the project references its API key variable. (https://github.com/browser-use/browser-use/blob/HEAD/.env.example) - [STRONG via SOURCE_SCAN] The user supplies their own DeepSeek key: the project references its API key variable. (https://github.com/browser-use/browser-use/blob/HEAD/.env.example) - [STRONG via SOURCE_SCAN] The user supplies their own Azure OpenAI key: the project references its API key variable. (https://github.com/browser-use/browser-use/blob/HEAD/.env.example) - [STRONG via SOURCE_SCAN] The user supplies their own xAI Grok key: the project references its API key variable. (https://github.com/browser-use/browser-use/blob/HEAD/.env.example) - [CONFIRMED via SOURCE_SCAN] Ships a container definition, so it can be self-hosted on your own infrastructure. (https://github.com/browser-use/browser-use/blob/HEAD/Dockerfile) - [REPORTED via SOURCE_SCAN] References PostHog, so some analytics or error reporting is present. This scan cannot tell whether it is opt-in or what it sends. (https://github.com/browser-use/browser-use/blob/HEAD/pyproject.toml) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-26, about 0 month(s) ago. (https://github.com/browser-use/browser-use/commits) ### KoboldCpp - Page: https://www.byoaik.com/tools/koboldcpp/ - Website: https://github.com/LostRuins/koboldcpp - Source: https://github.com/LostRuins/koboldcpp - Category: Local AI Tools | Pricing: OPEN_SOURCE | Open source: yes (AGPL-3.0) | Trust score: 87/100 - Platforms: Desktop, CLI, Self-hosted, Docker - Bring-your-own-key providers: Ollama, Custom / OpenAI-compatible - Key storage: NO_KEY_LOCAL_MODEL | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit 7 days ago) KoboldCpp is a self-contained distributable built on llama.cpp that bundles a GGUF model runner with the KoboldAI Lite web interface for chat, story writing and roleplay, running entirely on your own machine. You download one binary for Windows, macOS or Linux, or run it via Docker, Colab, Termux or a Raspberry Pi, and load a local model file: nothing leaves the device so there is no key to set. It also exposes an OpenAI-compatible local endpoint other clients can use. AGPL-3.0 for the interface layer with llama.cpp's MIT core, 11,415 stars, pushed the day of review. Key handling: Key handling varies. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 93/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 25/25: No cloud key is needed at all. The model runs locally. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under AGPL-3.0, so anyone can check how the key is handled. - Privacy 14/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/LostRuins/koboldcpp/blob/HEAD/README.md) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-26, about 0 month(s) ago. (https://github.com/LostRuins/koboldcpp/commits) ### llama.cpp - Page: https://www.byoaik.com/tools/llama-cpp/ - Website: https://github.com/ggml-org/llama.cpp - Source: https://github.com/ggml-org/llama.cpp - Category: Local AI Tools | Pricing: OPEN_SOURCE | Open source: yes (MIT) | Trust score: 87/100 - Platforms: CLI, Self-hosted, Library/SDK, Docker - Bring-your-own-key providers: Ollama, Custom / OpenAI-compatible - Key storage: NO_KEY_LOCAL_MODEL | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit today) llama.cpp is a dependency-light C/C++ inference engine that runs GGUF-quantised text and vision models on CPUs, Apple Silicon and NVIDIA or AMD GPUs, either as a one-shot CLI binary or a long-running server exposing an OpenAI-compatible API. There is no account, subscription or key: you build or download the binary and point it at a model file on disk. It underpins most of the other local tools listed here, including Ollama and KoboldCpp. MIT licensed, 124,009 stars, pushed the day of review with near-daily commits. Key handling: Key handling varies. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 93/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 25/25: No cloud key is needed at all. The model runs locally. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under MIT, so anyone can check how the key is handled. - Privacy 14/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/ggml-org/llama.cpp/blob/HEAD/README.md) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-29, about 0 month(s) ago. (https://github.com/ggml-org/llama.cpp/commits) ### Retrieval-based Voice Conversion WebUI - Page: https://www.byoaik.com/tools/retrieval-based-voice-conversion-webui/ - Website: https://github.com/RVC-Project/Retrieval-based-Voice-Conversion-WebUI - Source: https://github.com/RVC-Project/Retrieval-based-Voice-Conversion-WebUI - Category: Music and Audio Generation | Pricing: OPEN_SOURCE | Open source: yes (MIT) | Trust score: 87/100 - Platforms: Self-hosted, Web, Docker - Bring-your-own-key providers: Ollama - Key storage: NO_KEY_LOCAL_MODEL | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Slowing down (last commit about 3 months ago) RVC is an open-source voice conversion interface that lets you train a voice model from a short sample and run inference entirely on your own GPU. It is self-hosted and there is no external key, because both training and inference execute on local hardware. Ships a web interface, supports real-time voice changing and is widely used in cover and dubbing communities. MIT licensed, 37,360 stars, last commit within a month of review. Key handling: Key handling varies. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 93/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 25/25: No cloud key is needed at all. The model runs locally. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under MIT, so anyone can check how the key is handled. - Privacy 14/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-08-04, about 2 month(s) ago. (https://github.com/RVC-Project/Retrieval-based-Voice-Conversion-WebUI/commits) ### LlamaFactory - Page: https://www.byoaik.com/tools/llamafactory/ - Website: https://github.com/hiyouga/LlamaFactory - Source: https://github.com/hiyouga/LlamaFactory - Category: Fine-Tuning and Training | Pricing: OPEN_SOURCE | Open source: yes (Apache-2.0) | Trust score: 87/100 - Platforms: Web, CLI, Self-hosted, Local-only, Docker - Bring-your-own-key providers: Ollama, Custom / OpenAI-compatible - Key storage: NO_KEY_LOCAL_MODEL | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit today) LlamaFactory unifies fine-tuning across more than a hundred language and vision-language models, covering LoRA, QLoRA, full fine-tuning and preference methods, and ships LlamaBoard so runs can be configured and launched from a browser rather than a YAML file. Everything runs on your own hardware: no API key is involved at any point, and your training data stays where you put it. Apache-2.0. Key handling: Key handling varies. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 93/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 25/25: No cloud key is needed at all. The model runs locally. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under Apache-2.0, so anyone can check how the key is handled. - Privacy 14/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/hiyouga/LlamaFactory/blob/HEAD/README.md) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-28, about 0 month(s) ago. (https://github.com/hiyouga/LlamaFactory/commits) ### Ollama - Page: https://www.byoaik.com/tools/ollama/ - Website: https://ollama.com - Source: https://github.com/ollama/ollama - Category: Local AI Tools | Pricing: OPEN_SOURCE | Open source: yes (MIT) | Trust score: 87/100 - Platforms: CLI, Desktop, Docker, Self-hosted, Library/SDK - Bring-your-own-key providers: Ollama - Key storage: NO_KEY_LOCAL_MODEL | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit yesterday) Ollama is a CLI and local server that pulls quantised GGUF models and runs them on your own CPU or GPU, exposing an OpenAI-style REST API on localhost:11434 that most other tools in this directory can point at. There is no key to configure and no account: inference happens on your hardware, which is the strongest form of the no-markup premise. Installs via a script or native installer on macOS, Windows and Linux, with an official Docker image. MIT licensed, 178,589 stars, pushed the day of review. Key handling: Key handling varies. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 93/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 25/25: No cloud key is needed at all. The model runs locally. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under MIT, so anyone can check how the key is handled. - Privacy 14/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/ollama/ollama/blob/HEAD/README.md) - [CONFIRMED via SOURCE_SCAN] Ships a container definition, so it can be self-hosted on your own infrastructure. (https://github.com/ollama/ollama/blob/HEAD/Dockerfile) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-29, about 0 month(s) ago. (https://github.com/ollama/ollama/commits) ### Voicebox - Page: https://www.byoaik.com/tools/voicebox/ - Website: https://github.com/jamiepine/voicebox - Source: https://github.com/jamiepine/voicebox - Category: Music and Audio Generation | Pricing: OPEN_SOURCE | Open source: yes (MIT) | Trust score: 87/100 - Platforms: Desktop, Local-only, Self-hosted, Docker - Bring-your-own-key providers: Ollama, Custom / OpenAI-compatible - Key storage: NO_KEY_LOCAL_MODEL | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Maintained (last commit about 1 month ago) Voicebox is a local-first voice studio that clones a voice from a few seconds of audio and generates speech across many languages, spanning seven TTS engines behind one interface, plus dictation. It is presented as a free alternative to ElevenLabs and WisprFlow in a single app. Because synthesis runs locally, the recordings and the cloned voice stay on the machine rather than being uploaded to a service, which is the whole question with voice cloning. MIT licensed. Key handling: Key handling varies. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 93/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 25/25: No cloud key is needed at all. The model runs locally. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under MIT, so anyone can check how the key is handled. - Privacy 14/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [CONFIRMED via SOURCE_SCAN] Ships a container definition, so it can be self-hosted on your own infrastructure. (https://github.com/jamiepine/voicebox/blob/HEAD/docker-compose.yml) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-07-27, about 2 month(s) ago. (https://github.com/jamiepine/voicebox/commits) ### TRL - Page: https://www.byoaik.com/tools/trl/ - Website: https://github.com/huggingface/trl - Source: https://github.com/huggingface/trl - Category: Fine-Tuning and Training | Pricing: OPEN_SOURCE | Open source: yes (Apache-2.0) | Trust score: 87/100 - Platforms: CLI, Self-hosted, Local-only, Docker - Bring-your-own-key providers: Ollama, Custom / OpenAI-compatible - Key storage: NO_KEY_LOCAL_MODEL | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit yesterday) TRL is Hugging Face's library for post-training language models, covering supervised fine-tuning, reward modelling, DPO and GRPO, and it is the implementation most of the other fine-tuning frameworks build on or benchmark against. It is a library rather than a service: it runs in your own process on your own hardware, with no account and no key, and is the closest thing this space has to a reference implementation. Apache-2.0. Key handling: Key handling varies. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 93/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 25/25: No cloud key is needed at all. The model runs locally. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under Apache-2.0, so anyone can check how the key is handled. - Privacy 14/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/huggingface/trl/blob/HEAD/pyproject.toml) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-29, about 0 month(s) ago. (https://github.com/huggingface/trl/commits) ### MLX LM - Page: https://www.byoaik.com/tools/mlx-lm/ - Website: https://github.com/ml-explore/mlx-lm - Source: https://github.com/ml-explore/mlx-lm - Category: Fine-Tuning and Training | Pricing: OPEN_SOURCE | Open source: yes (MIT) | Trust score: 87/100 - Platforms: CLI, Desktop, Local-only, Self-hosted, Docker - Bring-your-own-key providers: Ollama, Custom / OpenAI-compatible - Key storage: NO_KEY_LOCAL_MODEL | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit 3 days ago) MLX LM runs, quantises and fine-tunes language models on Apple silicon through Apple's own MLX framework, with LoRA training that fits on a MacBook. It is the most accessible entry point in this category: no cluster, no cloud account, no key, and the model and data never leave the machine. From Apple's machine learning research group. MIT. Key handling: Key handling varies. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 93/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 25/25: No cloud key is needed at all. The model runs locally. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under MIT, so anyone can check how the key is handled. - Privacy 14/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-29, about 0 month(s) ago. (https://github.com/ml-explore/mlx-lm/commits) ### SD.Next - Page: https://www.byoaik.com/tools/sd-next/ - Website: https://vladmandic.github.io/sdnext/ - Source: https://github.com/vladmandic/sdnext - Category: Image Generation Tools | Pricing: OPEN_SOURCE | Open source: yes (Apache-2.0) | Trust score: 87/100 - Platforms: Web, Desktop, Self-hosted, Local-only, Docker - Bring-your-own-key providers: Ollama, Custom / OpenAI-compatible - Key storage: NO_KEY_LOCAL_MODEL | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit 10 days ago) SD.Next is a cross-platform WebUI for generating, refining, captioning, upscaling and processing images and video locally, supporting dozens of model architectures rather than a single family. It is one of the more actively developed successors in the Stable Diffusion WebUI lineage, where several once-popular alternatives have gone quiet. Models and outputs stay on your own hardware, so there is no key and no upload. Apache-2.0. Key handling: Key handling varies. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 93/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 25/25: No cloud key is needed at all. The model runs locally. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under Apache-2.0, so anyone can check how the key is handled. - Privacy 14/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-21, about 0 month(s) ago. (https://github.com/vladmandic/sdnext/commits) ### Unsloth - Page: https://www.byoaik.com/tools/unsloth/ - Website: https://unsloth.ai/ - Source: https://github.com/unslothai/unsloth - Category: Fine-Tuning and Training | Pricing: OPEN_SOURCE | Open source: yes (Apache-2.0) | Trust score: 87/100 - Platforms: CLI, Self-hosted, Local-only, Docker - Bring-your-own-key providers: Ollama, Custom / OpenAI-compatible - Key storage: NO_KEY_LOCAL_MODEL | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit today) Unsloth makes LoRA, QLoRA and full fine-tuning of open models substantially faster and less VRAM-hungry than a stock Hugging Face loop, which is what puts fine-tuning within reach of one consumer GPU. There is no API key anywhere in the picture: the base weights, your training data and the resulting adapter all stay on hardware you control, which matters because training data is usually the most sensitive material anyone puts near a model. Apache-2.0, and the most-starred project of its kind. Key handling: Key handling varies. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 93/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 25/25: No cloud key is needed at all. The model runs locally. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under Apache-2.0, so anyone can check how the key is handled. - Privacy 14/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/unslothai/unsloth/blob/HEAD/README.md) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-29, about 0 month(s) ago. (https://github.com/unslothai/unsloth/commits) ### promptfoo - Page: https://www.byoaik.com/tools/promptfoo/ - Website: https://www.promptfoo.dev - Source: https://github.com/promptfoo/promptfoo - Category: Prompt Engineering Tools | Pricing: OPEN_SOURCE | Open source: yes (MIT) | Trust score: 87/100 - Platforms: CLI, Self-hosted, Web, Docker - Bring-your-own-key providers: OpenAI, Anthropic, Google Gemini, Azure OpenAI, Mistral, DeepSeek, Ollama, Custom / OpenAI-compatible - Key storage: ENVIRONMENT_VARIABLE | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit today) promptfoo is an open-source CLI and library for testing prompts, evaluating outputs, comparing models and red-teaming AI applications. You supply your own provider keys as environment variables or in YAML provider configs, and every call to OpenAI, Anthropic, Gemini, DeepSeek, Azure, a local Ollama model or any OpenAI-compatible endpoint goes directly from your machine with nothing proxied. Ships as a Node CLI with a local web viewer for results. MIT licensed, 24,256 stars, last commit the day of review. An optional paid enterprise tier exists alongside the free core. Key handling: Your API key is supplied via an environment variable or local config file. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 83/100 (assessable for any licence), verifiability 90/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 22/25: The key is supplied by an environment variable or local config file you control. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under MIT, so anyone can check how the key is handled. Key handling was located in the published source. - Privacy 11/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. Analytics or error-reporting libraries are present; what they send was not established. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 7/10: Key handling was found in the published source by BYOAIK's scanner. What was checked (verification tier SOURCE_VERIFIED, derived from this evidence): - [STRONG via SOURCE_SCAN] The user supplies their own OpenAI key: the project references its API key variable. (https://github.com/promptfoo/promptfoo/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/promptfoo/promptfoo/blob/HEAD/README.md) - [CONFIRMED via SOURCE_SCAN] Ships a container definition, so it can be self-hosted on your own infrastructure. (https://github.com/promptfoo/promptfoo/blob/HEAD/Dockerfile) - [REPORTED via SOURCE_SCAN] References PostHog, so some analytics or error reporting is present. This scan cannot tell whether it is opt-in or what it sends. (https://github.com/promptfoo/promptfoo/blob/HEAD/package.json) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-29, about 0 month(s) ago. (https://github.com/promptfoo/promptfoo/commits) ### ACE-Step - Page: https://www.byoaik.com/tools/ace-step/ - Website: https://github.com/ace-step/ACE-Step-1.5 - Source: https://github.com/ace-step/ACE-Step-1.5 - Category: Music and Audio Generation | Pricing: OPEN_SOURCE | Open source: yes (MIT) | Trust score: 87/100 - Platforms: Desktop, Self-hosted, Local-only, Docker - Bring-your-own-key providers: Ollama, Custom / OpenAI-compatible - Key storage: NO_KEY_LOCAL_MODEL | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Maintained (last commit about 2 months ago) ACE-Step 1.5 is a music foundation model that generates complete songs locally, with vocals and instrumentals, on Mac, AMD, Intel or CUDA hardware. Both the code and the weights are MIT licensed and commercial use is allowed, so the model runs on your machine rather than behind someone's API. There is no account and no key: the privacy question for music generation is whether your lyrics and prompts leave the machine, and here they do not. Key handling: Key handling varies. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 93/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 25/25: No cloud key is needed at all. The model runs locally. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under MIT, so anyone can check how the key is handled. - Privacy 14/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/ace-step/ACE-Step-1.5/blob/HEAD/README.md) - [CONFIRMED via SOURCE_SCAN] Ships a container definition, so it can be self-hosted on your own infrastructure. (https://github.com/ace-step/ACE-Step-1.5/blob/HEAD/docker-compose.yml) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-08-29, about 1 month(s) ago. (https://github.com/ace-step/ACE-Step-1.5/commits) ### Applio - Page: https://www.byoaik.com/tools/applio/ - Website: https://github.com/IAHispano/Applio - Source: https://github.com/IAHispano/Applio - Category: Music and Audio Generation | Pricing: OPEN_SOURCE | Open source: yes (MIT) | Trust score: 87/100 - Platforms: Desktop, Self-hosted, Local-only, Docker - Bring-your-own-key providers: Ollama, Custom / OpenAI-compatible - Key storage: NO_KEY_LOCAL_MODEL | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit 4 days ago) Applio is a voice conversion tool built on the RVC lineage, packaging model training, inference and management behind a simpler interface than the original web UI. Everything runs locally: the source audio, the trained voice models and the output all stay on the machine, with no service holding the voice. MIT licensed and considerably more active than the original RVC WebUI. Key handling: Key handling varies. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 93/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 25/25: No cloud key is needed at all. The model runs locally. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under MIT, so anyone can check how the key is handled. - Privacy 14/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [CONFIRMED via SOURCE_SCAN] Ships a container definition, so it can be self-hosted on your own infrastructure. (https://github.com/IAHispano/Applio/blob/HEAD/docker-compose.yaml) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-29, about 0 month(s) ago. (https://github.com/IAHispano/Applio/commits) ### Piper - Page: https://www.byoaik.com/tools/piper/ - Website: https://github.com/OHF-Voice/piper1-gpl - Source: https://github.com/OHF-Voice/piper1-gpl - Category: Music and Audio Generation | Pricing: OPEN_SOURCE | Open source: yes (GPL-3.0) | Trust score: 87/100 - Platforms: CLI, Local-only, Self-hosted, Docker - Bring-your-own-key providers: Ollama, Custom / OpenAI-compatible - Key storage: NO_KEY_LOCAL_MODEL | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit yesterday) Piper is a fast, local neural text-to-speech system built for small hardware, widely used as the speech engine in offline voice assistants and accessibility setups. It runs on devices as modest as a Raspberry Pi with no network call and no key, which is the reason it appears in self-hosted home-automation stacks. Now maintained under the Open Home Foundation as piper1-gpl, GPL-3.0. Key handling: Key handling varies. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 93/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 25/25: No cloud key is needed at all. The model runs locally. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under GPL-3.0, so anyone can check how the key is handled. - Privacy 14/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [CONFIRMED via SOURCE_SCAN] Ships a container definition, so it can be self-hosted on your own infrastructure. (https://github.com/OHF-Voice/piper1-gpl/blob/HEAD/Dockerfile) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-28, about 0 month(s) ago. (https://github.com/OHF-Voice/piper1-gpl/commits) ### SwarmUI - Page: https://www.byoaik.com/tools/swarmui/ - Website: https://github.com/mcmonkeyprojects/SwarmUI - Source: https://github.com/mcmonkeyprojects/SwarmUI - Category: Image Generation Tools | Pricing: OPEN_SOURCE | Open source: yes (MIT) | Trust score: 87/100 - Platforms: Self-hosted, Web, Docker - Bring-your-own-key providers: Ollama - Key storage: NO_KEY_LOCAL_MODEL | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit today) SwarmUI, formerly StableSwarmUI, is a modular web interface that sits on top of local diffusion backends such as ComfyUI, giving a friendly front end while all image generation happens on compute you installed. It is self-hosted and drives locally run models rather than any cloud image API, so this is bring-your-own-compute rather than a key. Maintained by a former Stability AI developer and the community. MIT licensed, 4,460 stars, last commit the day of review. Key handling: Key handling varies. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 93/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 25/25: No cloud key is needed at all. The model runs locally. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under MIT, so anyone can check how the key is handled. - Privacy 14/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-29, about 0 month(s) ago. (https://github.com/mcmonkeyprojects/SwarmUI/commits) ### vLLM - Page: https://www.byoaik.com/tools/vllm/ - Website: https://vllm.ai - Source: https://github.com/vllm-project/vllm - Category: Self-Hosted AI Platforms | Pricing: OPEN_SOURCE | Open source: yes (Apache-2.0) | Trust score: 87/100 - Platforms: Self-hosted, Docker, Library/SDK - Bring-your-own-key providers: Ollama, Hugging Face, Custom / OpenAI-compatible - Key storage: NO_KEY_LOCAL_MODEL | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit today) vLLM is an inference and serving library for running open-weight models on your own NVIDIA, AMD or Intel GPUs at production throughput, using PagedAttention and continuous batching. It exposes an OpenAI-compatible API server deployed on hardware you provision, so no vendor key is involved. It is widely used as the serving layer beneath self-hosted deployments and by inference providers that let you bring your own weights. Apache-2.0, 89,119 stars, pushed the day of review with very active development. Key handling: Key handling varies. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 93/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 25/25: No cloud key is needed at all. The model runs locally. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under Apache-2.0, so anyone can check how the key is handled. - Privacy 14/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/vllm-project/vllm/blob/HEAD/README.md) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-29, about 0 month(s) ago. (https://github.com/vllm-project/vllm/commits) ### cognee - Page: https://www.byoaik.com/tools/cognee/ - Website: https://www.cognee.ai/ - Source: https://github.com/topoteretes/cognee - Category: Knowledge Management | Pricing: OPEN_SOURCE | Open source: yes (Apache-2.0) | Trust score: 87/100 - Platforms: CLI, Self-hosted, Docker - Bring-your-own-key providers: OpenAI, Anthropic, Google Gemini, Azure OpenAI, Ollama, Custom / OpenAI-compatible - Key storage: ENVIRONMENT_VARIABLE | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit 2 days ago) cognee replaces ad-hoc RAG plumbing with an extract-cognify-load pipeline that turns documents and conversations into a queryable graph plus vector store. OPENAI_API_KEY or a local Ollama both work, so the pipeline can be run without a cloud credential. Memory systems accumulate: what goes in tends to stay, across sessions, which makes where they run a longer-lived decision than most. Apache-2.0. Key handling: Your API key is supplied via an environment variable or local config file. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 83/100 (assessable for any licence), verifiability 90/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 22/25: The key is supplied by an environment variable or local config file you control. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under Apache-2.0, so anyone can check how the key is handled. Key handling was located in the published source. - Privacy 11/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. Analytics or error-reporting libraries are present; what they send was not established. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 7/10: Key handling was found in the published source by BYOAIK's scanner. What was checked (verification tier SOURCE_VERIFIED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [STRONG via SOURCE_SCAN] The user supplies their own OpenAI key: the project references its API key variable. (https://github.com/topoteretes/cognee/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/topoteretes/cognee/blob/HEAD/README.md) - [CONFIRMED via SOURCE_SCAN] Ships a container definition, so it can be self-hosted on your own infrastructure. (https://github.com/topoteretes/cognee/blob/HEAD/docker-compose.yml) - [REPORTED via SOURCE_SCAN] References PostHog, so some analytics or error reporting is present. This scan cannot tell whether it is opt-in or what it sends. (https://github.com/topoteretes/cognee/blob/HEAD/pyproject.toml) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-29, about 0 month(s) ago. (https://github.com/topoteretes/cognee/commits) ### Chatterbox - Page: https://www.byoaik.com/tools/chatterbox/ - Website: https://github.com/resemble-ai/chatterbox - Source: https://github.com/resemble-ai/chatterbox - Category: Music and Audio Generation | Pricing: OPEN_SOURCE | Open source: yes (MIT) | Trust score: 87/100 - Platforms: Desktop, Self-hosted, Local-only, Docker - Bring-your-own-key providers: Ollama, Custom / OpenAI-compatible - Key storage: NO_KEY_LOCAL_MODEL | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit 4 days ago) Chatterbox is Resemble AI's open-source text-to-speech model, offering zero-shot voice cloning from a few seconds of audio, emotion control and real-time generation across a couple of dozen languages. Unusually for a model from a company whose product is a hosted voice API, it is MIT licensed and runs on your own hardware, so the audio never has to reach the vendor. Paralinguistic tags such as laugh and whisper are supported. Key handling: Key handling varies. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 93/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 25/25: No cloud key is needed at all. The model runs locally. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under MIT, so anyone can check how the key is handled. - Privacy 14/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-07-21, about 2 month(s) ago. (https://github.com/resemble-ai/chatterbox/commits) ### PEFT - Page: https://www.byoaik.com/tools/peft/ - Website: https://huggingface.co/docs/peft - Source: https://github.com/huggingface/peft - Category: Fine-Tuning and Training | Pricing: OPEN_SOURCE | Open source: yes (Apache-2.0) | Trust score: 87/100 - Platforms: CLI, Self-hosted, Local-only, Docker - Bring-your-own-key providers: Ollama, Custom / OpenAI-compatible - Key storage: NO_KEY_LOCAL_MODEL | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit today) PEFT implements parameter-efficient fine-tuning, LoRA, QLoRA, prompt tuning and the rest, which is what makes adapting a large model possible on hardware a person can own rather than a cluster they rent. It is the layer Unsloth, Axolotl and LlamaFactory sit on top of. No key and no service: the base weights, the training data and the resulting adapter all stay where you put them. Apache-2.0. Key handling: Key handling varies. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 93/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 25/25: No cloud key is needed at all. The model runs locally. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under Apache-2.0, so anyone can check how the key is handled. - Privacy 14/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-29, about 0 month(s) ago. (https://github.com/huggingface/peft/commits) ### ms-swift - Page: https://www.byoaik.com/tools/ms-swift/ - Website: https://github.com/modelscope/ms-swift - Source: https://github.com/modelscope/ms-swift - Category: Fine-Tuning and Training | Pricing: OPEN_SOURCE | Open source: yes (Apache-2.0) | Trust score: 87/100 - Platforms: CLI, Web, Self-hosted, Local-only, Docker - Bring-your-own-key providers: Ollama, Custom / OpenAI-compatible - Key storage: NO_KEY_LOCAL_MODEL | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit 2 days ago) ms-swift is Alibaba ModelScope's framework for fine-tuning and deploying language and multimodal models, covering an unusually wide model catalogue including the Qwen family, with LoRA, QLoRA, full tuning and preference methods plus a GUI. Everything runs on infrastructure you provide, so there is no key and the training data stays local. Worth knowing where it comes from: this is the reference toolchain for models published on ModelScope, which is Alibaba's hub. Apache-2.0. Key handling: Key handling varies. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 93/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 25/25: No cloud key is needed at all. The model runs locally. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under Apache-2.0, so anyone can check how the key is handled. - Privacy 14/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/modelscope/ms-swift/blob/HEAD/README.md) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-28, about 0 month(s) ago. (https://github.com/modelscope/ms-swift/commits) ### YuE - Page: https://www.byoaik.com/tools/yue/ - Website: https://github.com/multimodal-art-projection/YuE - Source: https://github.com/multimodal-art-projection/YuE - Category: Music and Audio Generation | Pricing: OPEN_SOURCE | Open source: yes (Apache-2.0) | Trust score: 87/100 - Platforms: CLI, Self-hosted, Local-only, Docker - Bring-your-own-key providers: Ollama, Custom / OpenAI-compatible - Key storage: NO_KEY_LOCAL_MODEL | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit 8 days ago) YuE generates complete songs from lyrics and a style prompt, producing vocals and instrumental together across several minutes rather than short clips, and it runs locally on your own hardware. Apache-2.0 over the code, which puts long-form music generation in the same place as the rest of this category: on a machine you control, with your unreleased lyrics staying there. Key handling: Key handling varies. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 93/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 25/25: No cloud key is needed at all. The model runs locally. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under Apache-2.0, so anyone can check how the key is handled. - Privacy 14/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/multimodal-art-projection/YuE/blob/HEAD/pyproject.toml) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-29, about 0 month(s) ago. (https://github.com/multimodal-art-projection/YuE/commits) ### verl - Page: https://www.byoaik.com/tools/verl/ - Website: https://github.com/verl-project/verl - Source: https://github.com/verl-project/verl - Category: Fine-Tuning and Training | Pricing: OPEN_SOURCE | Open source: yes (Apache-2.0) | Trust score: 87/100 - Platforms: CLI, Self-hosted, Docker - Bring-your-own-key providers: Ollama, Custom / OpenAI-compatible - Key storage: NO_KEY_LOCAL_MODEL | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit today) verl is a reinforcement learning library for post-training language models, covering PPO, GRPO and related algorithms, designed to scale across many GPUs. Originally from ByteDance and now under its own project organisation. This is the part of the stack where a model is shaped by your own preference data, which is usually the most proprietary material a team has, and none of it leaves the hardware you run. Apache-2.0. Key handling: Key handling varies. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 93/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 25/25: No cloud key is needed at all. The model runs locally. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under Apache-2.0, so anyone can check how the key is handled. - Privacy 14/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/verl-project/verl/blob/HEAD/README.md) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-29, about 0 month(s) ago. (https://github.com/verl-project/verl/commits) ### OpenRLHF - Page: https://www.byoaik.com/tools/openrlhf/ - Website: https://github.com/OpenRLHF/OpenRLHF - Source: https://github.com/OpenRLHF/OpenRLHF - Category: Fine-Tuning and Training | Pricing: OPEN_SOURCE | Open source: yes (Apache-2.0) | Trust score: 87/100 - Platforms: CLI, Self-hosted, Docker - Bring-your-own-key providers: Ollama, Custom / OpenAI-compatible - Key storage: NO_KEY_LOCAL_MODEL | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit 6 days ago) OpenRLHF implements RLHF and related alignment methods on Ray, vLLM and DeepSpeed, aiming to stay readable rather than maximally abstracted. That readability is the point for anyone who needs to know exactly what happens to their preference data, which never leaves the cluster you run it on. Apache-2.0. Key handling: Key handling varies. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 93/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 25/25: No cloud key is needed at all. The model runs locally. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under Apache-2.0, so anyone can check how the key is handled. - Privacy 14/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/OpenRLHF/OpenRLHF/blob/HEAD/README.md) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-17, about 0 month(s) ago. (https://github.com/OpenRLHF/OpenRLHF/commits) ### F5-TTS - Page: https://www.byoaik.com/tools/f5-tts/ - Website: https://github.com/SWivid/F5-TTS - Source: https://github.com/SWivid/F5-TTS - Category: Music and Audio Generation | Pricing: OPEN_SOURCE | Open source: yes (MIT) | Trust score: 87/100 - Platforms: CLI, Self-hosted, Local-only, Docker - Bring-your-own-key providers: Ollama, Custom / OpenAI-compatible - Key storage: NO_KEY_LOCAL_MODEL | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit 5 days ago) F5-TTS is a flow-matching text-to-speech model that clones a voice from a few seconds of reference audio and synthesises fluently across languages, with training and inference code published under MIT. Running it locally is the whole point for voice work: a reference sample is biometric material, and the difference between processing it on your own machine and uploading it to a service is not a small one. Key handling: Key handling varies. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 93/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 25/25: No cloud key is needed at all. The model runs locally. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under MIT, so anyone can check how the key is handled. - Privacy 14/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [CONFIRMED via SOURCE_SCAN] Ships a container definition, so it can be self-hosted on your own infrastructure. (https://github.com/SWivid/F5-TTS/blob/HEAD/Dockerfile) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-21, about 0 month(s) ago. (https://github.com/SWivid/F5-TTS/commits) ### big-AGI - Page: https://www.byoaik.com/tools/big-agi/ - Website: https://big-agi.com - Source: https://github.com/enricoros/big-AGI - Category: AI Chat Interfaces | Pricing: OPEN_SOURCE | Open source: yes (MIT) | Trust score: 86/100 - Platforms: Web, Self-hosted, Docker - Bring-your-own-key providers: OpenAI, Anthropic, Google Gemini, Azure OpenAI, OpenRouter, Groq, Mistral, Perplexity, DeepSeek, xAI Grok, Together AI, Ollama, Custom / OpenAI-compatible - Key storage: SELF_HOSTED_ONLY | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit today) big-AGI is an open-source, multi-model AI workspace aimed at "experts." It runs on a bring-your-own-key model: users connect their own provider API keys (OpenAI, Anthropic, Google Gemini, and 20+ others) and big-AGI charges nothing for model usage, with keys stored locally in the browser by default. Standout features include Beam (run one prompt across multiple models simultaneously to compare/validate responses), customizable AI personas, text-to-image generation and editing, voice/TTS, web search with citations, code highlighting and execution, PDF import, and split-window multi-chat. It can be used via the free hosted app at app.big-agi.com (BYOK) or self-hosted from the MIT-licensed codebase via Docker/Vercel; a paid Pro tier adds cloud sync, backup, and priority support. A native desktop app is advertised as "coming soon." The repo is actively maintained (v2.0.5, May 2026, ~7k stars) and explicitly markets itself as non-VC-funded with zero lock-in. Key handling: You self-host the app, so your API key lives in your own deployment. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 82/100 (assessable for any licence), verifiability 90/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 21/25: The key lives in a deployment you run yourself. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under MIT, so anyone can check how the key is handled. Key handling was located in the published source. - Privacy 11/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. Analytics or error-reporting libraries are present; what they send was not established. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 7/10: Key handling was found in the published source by BYOAIK's scanner. What was checked (verification tier SOURCE_VERIFIED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [STRONG via SOURCE_SCAN] The user supplies their own OpenAI key: the project references its API key variable. (https://github.com/enricoros/big-AGI/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/enricoros/big-AGI/blob/HEAD/README.md) - [CONFIRMED via SOURCE_SCAN] Ships a container definition, so it can be self-hosted on your own infrastructure. (https://github.com/enricoros/big-AGI/blob/HEAD/docker-compose.yaml) - [REPORTED via SOURCE_SCAN] References PostHog, so some analytics or error reporting is present. This scan cannot tell whether it is opt-in or what it sends. (https://github.com/enricoros/big-AGI/blob/HEAD/Dockerfile) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-29, about 0 month(s) ago. (https://github.com/enricoros/big-AGI/commits) ### Forge - Page: https://www.byoaik.com/tools/forge/ - Website: https://forgecode.dev - Source: https://github.com/tailcallhq/forgecode - Category: Developer Infrastructure | Pricing: OPEN_SOURCE | Open source: yes (Apache-2.0) | Trust score: 86/100 - Platforms: CLI, Self-hosted, Docker - Bring-your-own-key providers: OpenAI, Anthropic, Google Gemini, OpenRouter, Groq, DeepSeek, xAI Grok, Custom / OpenAI-compatible - Key storage: ENVIRONMENT_VARIABLE | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit 8 days ago) Forge (forgecode) is an open-source, Apache-2.0 licensed AI pair programmer that runs entirely in the terminal/shell (written in Rust). It connects directly to your AI provider of choice using your own API keys: set up via the interactive `forge provider login` / `:login` flow, or via environment variables (OPENAI_API_KEY, ANTHROPIC_API_KEY, OPENROUTER_API_KEY, now deprecated but auto-migrated to file-based storage on first run). It supports 300+ models across OpenAI, Anthropic, Google Vertex/Gemini, xAI Grok, DeepSeek, Groq, Amazon Bedrock, OpenRouter, Cerebras, and any OpenAI-compatible custom endpoint, with the ability to mix and match models mid-session. Users can also leverage existing ChatGPT Plus / Claude subscriptions instead of separate keys. An optional, supplementary 'ForgeCode Services' layer (enhanced codebase understanding, tool-call guardrails, semantic search) exists and requires no key, but model access itself is fully bring-your-own-key. The canonical repo is github.com/tailcallhq/forgecode (the antinomyhq/forgecode URL 301-redirects there; npm package @antinomyhq/forge), with 7421 stars, 1453 forks, actively maintained (pushed 2026-06-19). Key handling: Your API key is supplied via an environment variable or local config file. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 82/100 (assessable for any licence), verifiability 90/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 22/25: The key is supplied by an environment variable or local config file you control. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under Apache-2.0, so anyone can check how the key is handled. Key handling was located in the published source. - Privacy 10/15: Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 7/10: Key handling was found in the published source by BYOAIK's scanner. What was checked (verification tier SOURCE_VERIFIED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [STRONG via SOURCE_SCAN] The user supplies their own OpenAI key: the project references its API key variable. (https://github.com/tailcallhq/forgecode/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] The user supplies their own Anthropic key: the project references its API key variable. (https://github.com/tailcallhq/forgecode/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] The user supplies their own OpenRouter key: the project references its API key variable. (https://github.com/tailcallhq/forgecode/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] The user supplies their own xAI Grok key: the project references its API key variable. (https://github.com/tailcallhq/forgecode/blob/HEAD/README.md) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-29, about 0 month(s) ago. (https://github.com/tailcallhq/forgecode/commits) ### Morphic - Page: https://www.byoaik.com/tools/morphic/ - Website: https://github.com/miurla/morphic - Source: https://github.com/miurla/morphic - Category: Research Tools | Pricing: OPEN_SOURCE | Open source: yes (Apache-2.0) | Trust score: 86/100 - Platforms: Web, Self-hosted, Docker - Bring-your-own-key providers: OpenAI, Anthropic, Google Gemini, Ollama, Custom / OpenAI-compatible - Key storage: SELF_HOSTED_ONLY | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit today) Morphic is an open-source, self-hostable AI-powered answer/search engine with a generative UI built on Next.js and the Vercel AI SDK. It understands a question, runs web search (via Tavily, Exa, Brave, SearXNG, Firecrawl, or Jina), and streams a cited answer rendered with interactive generative-UI components, plus chat history (PostgreSQL/Supabase), file upload, and Quick/Adaptive research modes. It is bring-your-own-key: the end user edits .env.local and supplies their own LLM provider credentials (OPENAI_API_KEY, ANTHROPIC_API_KEY, GOOGLE_GENERATIVE_AI_API_KEY, AI_GATEWAY_API_KEY, OLLAMA_BASE_URL, plus OPENAI_COMPATIBLE_* variables for any OpenAI-compatible endpoint). The model selector uses dynamic provider detection, so configured providers appear automatically. Deployable via Vercel one-click, Docker Compose, Cloudflare Pages, or local dev. At least one AI provider key is required to run. Key handling: You self-host the app, so your API key lives in your own deployment. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 82/100 (assessable for any licence), verifiability 90/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 21/25: The key lives in a deployment you run yourself. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under Apache-2.0, so anyone can check how the key is handled. Key handling was located in the published source. - Privacy 11/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. Analytics or error-reporting libraries are present; what they send was not established. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 7/10: Key handling was found in the published source by BYOAIK's scanner. What was checked (verification tier SOURCE_VERIFIED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [STRONG via SOURCE_SCAN] The user supplies their own OpenAI key: the project references its API key variable. (https://github.com/miurla/morphic/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/miurla/morphic/blob/HEAD/README.md) - [CONFIRMED via SOURCE_SCAN] Ships a container definition, so it can be self-hosted on your own infrastructure. (https://github.com/miurla/morphic/blob/HEAD/docker-compose.yaml) - [REPORTED via SOURCE_SCAN] References PostHog, so some analytics or error reporting is present. This scan cannot tell whether it is opt-in or what it sends. (https://github.com/miurla/morphic/blob/HEAD/package.json) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-26, about 0 month(s) ago. (https://github.com/miurla/morphic/commits) ### Talon - Page: https://www.byoaik.com/tools/talon/ - Website: https://talon-agent.github.io/ - Source: https://github.com/thefalconry/talon - Category: Agent Frameworks | Pricing: OPEN_SOURCE | Open source: yes (MIT) | Trust score: 86/100 - Platforms: CLI, Desktop, Self-hosted, Docker - Bring-your-own-key providers: OpenAI, Anthropic, Custom / OpenAI-compatible - Key storage: ENVIRONMENT_VARIABLE | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit 18 days ago) Talon is a self-hosted, multi-platform runtime for long-running AI agents. Model access is yours to supply: the Claude backend drives an authenticated Claude Code CLI, the Codex backend uses the codex CLI with CODEX_API_KEY, TALON_CODEX_KEY or OPENAI_API_KEY, and any OpenAI-compatible endpoint can be pointed wherever you like. It connects local or remote MCP tools and keeps goals, cron jobs, triggers, memory and background heartbeats on infrastructure you run. Memory can stay entirely local through the mempalace backend rather than a hosted service. MIT licensed and actively developed. Key handling: Your API key is supplied via an environment variable or local config file. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 82/100 (assessable for any licence), verifiability 90/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 22/25: The key is supplied by an environment variable or local config file you control. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under MIT, so anyone can check how the key is handled. Key handling was located in the published source. - Privacy 10/15: Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 7/10: Key handling was found in the published source by BYOAIK's scanner. What was checked (verification tier SOURCE_VERIFIED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [STRONG via SOURCE_SCAN] The user supplies their own OpenAI key: the project references its API key variable. (https://github.com/thefalconry/talon/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] The user supplies their own Anthropic key: the project references its API key variable. (https://github.com/thefalconry/talon/blob/HEAD/docker-compose.yml) - [CONFIRMED via SOURCE_SCAN] Ships a container definition, so it can be self-hosted on your own infrastructure. (https://github.com/thefalconry/talon/blob/HEAD/docker-compose.yml) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-29, about 0 month(s) ago. (https://github.com/thefalconry/talon/commits) ### Stagehand - Page: https://www.byoaik.com/tools/stagehand/ - Website: https://www.stagehand.dev/ - Source: https://github.com/browserbase/stagehand - Category: Automation Tools | Pricing: OPEN_SOURCE | Open source: yes (MIT) | Trust score: 86/100 - Platforms: CLI, Self-hosted, Local-only, Docker - Bring-your-own-key providers: OpenAI, Anthropic, Custom / OpenAI-compatible - Key storage: ENVIRONMENT_VARIABLE | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit today) Stagehand lets you drive a browser with ordinary Playwright code and drop into natural language only where a selector would be brittle, which keeps automation deterministic where it can be. The model credential is passed explicitly in configuration, read from your own environment, and the library runs wherever you run it. Browserbase sells hosted browser infrastructure alongside it, but the library itself runs locally against a browser you control. MIT. Key handling: Your API key is supplied via an environment variable or local config file. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 82/100 (assessable for any licence), verifiability 90/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 22/25: The key is supplied by an environment variable or local config file you control. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under MIT, so anyone can check how the key is handled. Key handling was located in the published source. - Privacy 10/15: Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 7/10: Key handling was found in the published source by BYOAIK's scanner. What was checked (verification tier SOURCE_VERIFIED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [STRONG via SOURCE_SCAN] The user supplies their own OpenAI key: the project references its API key variable. (https://github.com/browserbase/stagehand/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] Reads the API key from an environment variable at runtime. (https://github.com/browserbase/stagehand/blob/HEAD/README.md) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-28, about 0 month(s) ago. (https://github.com/browserbase/stagehand/commits) ### Goose - Page: https://www.byoaik.com/tools/goose/ - Website: https://block.github.io/goose/ - Source: https://github.com/aaif-goose/goose - Category: Agent Frameworks | Pricing: OPEN_SOURCE | Open source: yes (Apache-2.0) | Trust score: 85/100 - Platforms: CLI, Desktop, Self-hosted, Library/SDK, Docker - Bring-your-own-key providers: Anthropic, OpenAI, Google Gemini, Azure OpenAI, OpenRouter, Groq, Mistral, Perplexity, DeepSeek, xAI Grok, Ollama, Custom / OpenAI-compatible - Key storage: LOCAL_DEVICE | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit today) Goose is an open-source AI agent from Block, built in Rust and available as a native desktop app (macOS, Linux, Windows), a full CLI, and an embeddable API. It autonomously installs dependencies, executes commands, edits files, and runs/tests code, extensible via 70+ Model Context Protocol (MCP) extensions. It is fully bring-your-own-key: running `goose configure` (CLI) or the Models tab (Desktop) prompts you to select a provider and enter your own API key. It supports 25+/50+ providers including Anthropic, OpenAI, Google Gemini, xAI Grok, Mistral, Perplexity, DeepSeek, Groq, OpenRouter, and Azure OpenAI; cloud platforms Amazon Bedrock, GCP Vertex AI, Databricks, and Snowflake; local runners like Ollama, LM Studio, and Docker Model Runner; and any custom OpenAI-/Anthropic-/Ollama-compatible endpoint with configurable base URL and headers. Key handling: Your API key is stored locally on your device. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 90/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 23/25: The key is held on your own device. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under Apache-2.0, so anyone can check how the key is handled. - Privacy 14/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/aaif-goose/goose/blob/HEAD/README.md) - [CONFIRMED via SOURCE_SCAN] Ships a container definition, so it can be self-hosted on your own infrastructure. (https://github.com/aaif-goose/goose/blob/HEAD/Dockerfile) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-29, about 0 month(s) ago. (https://github.com/aaif-goose/goose/commits) ### Jan - Page: https://www.byoaik.com/tools/jan/ - Website: https://jan.ai/ - Source: https://github.com/janhq/jan - Category: Desktop AI Assistants | Pricing: OPEN_SOURCE | Open source: yes (Apache-2.0) | Trust score: 85/100 - Platforms: Desktop, CLI, Self-hosted, Docker - Bring-your-own-key providers: OpenAI, Anthropic, Google Gemini, Azure OpenAI, OpenRouter, Groq, Mistral, Hugging Face, Ollama, Custom / OpenAI-compatible - Key storage: LOCAL_DEVICE | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit today) Jan is an open-source (Apache 2.0) desktop AI assistant from Menlo Research positioned as a private, local-first alternative to ChatGPT. It bundles local inference (HuggingFace models like Llama, Gemma, Qwen, GPT-oss) and runs fully offline when desired, exposing a local OpenAI-compatible API server at localhost:1337. Crucially for a BYOK directory, its documented Settings -> Model Providers flow lets a user paste their own API key for cloud providers and pick a model from that provider. Official docs (jan.ai/docs/desktop/remote-models/*) walk through the identical "create & copy an API key -> Settings -> Model Providers -> insert your API Key" flow for OpenAI, Anthropic, Google Gemini, Groq, Mistral, Azure OpenAI, OpenRouter, and Hugging Face, plus Custom Endpoints for any OpenAI- or Anthropic-compatible base URL. It also supports MCP for agentic features and ships a CLI. Available for Windows 10+, macOS 13.6+, and Linux (deb/AppImage/Arm64), with Microsoft Store and Flathub distribution. ~43k GitHub stars and active (latest release v0.8.2, 100+ releases). Note: the candidate's claimed MIT license is incorrect; the actual license is Apache 2.0. Key handling: Your API key is stored locally on your device. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 90/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 23/25: The key is held on your own device. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under Apache-2.0, so anyone can check how the key is handled. - Privacy 14/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/janhq/jan/blob/HEAD/README.md) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-29, about 0 month(s) ago. (https://github.com/janhq/jan/commits) ### BrowserOS - Page: https://www.byoaik.com/tools/browseros/ - Website: https://www.browseros.com/ - Source: https://github.com/browseros-ai/BrowserOS - Category: Agent Frameworks | Pricing: OPEN_SOURCE | Open source: yes (AGPL-3.0) | Trust score: 85/100 - Platforms: Desktop, CLI, Self-hosted, Docker - Bring-your-own-key providers: Anthropic, OpenAI, Google Gemini, OpenRouter, Azure OpenAI, Ollama, DeepSeek, Custom / OpenAI-compatible - Key storage: LOCAL_DEVICE | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit today) BrowserOS is an open-source, privacy-first Chromium fork (a standalone desktop browser, not an extension) with built-in AI agents that automate web tasks from natural-language prompts: clicking, typing, navigating, reading local files, and running multi-step workflows. It ships 53 built-in browser automation tools and 40+ app integrations (Gmail, Slack, Notion, Calendar) via MCP, plus scheduled tasks and a CLI to drive the browser from coding agents like Claude Code. Users bring their own provider API keys and switch between providers anytime; a default model with limited daily usage is included for testing, and local models run free via Ollama / LM Studio. Positioned as a privacy-first alternative to ChatGPT Atlas, Perplexity Comet, and Dia. Built by a YC-backed team (founded 2024, San Francisco). Licensed AGPL-3.0; available on macOS, Windows, and Linux. Key handling: Your API key is stored locally on your device. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 90/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 23/25: The key is held on your own device. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under AGPL-3.0, so anyone can check how the key is handled. - Privacy 14/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/browseros-ai/BrowserOS/blob/HEAD/README.md) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-29, about 0 month(s) ago. (https://github.com/browseros-ai/BrowserOS/commits) ### Krita AI Diffusion - Page: https://www.byoaik.com/tools/krita-ai-diffusion/ - Website: https://github.com/Acly/krita-ai-diffusion - Source: https://github.com/Acly/krita-ai-diffusion - Category: Image Generation Tools | Pricing: OPEN_SOURCE | Open source: yes (GPL-3.0) | Trust score: 85/100 - Platforms: Desktop, Self-hosted, Docker - Bring-your-own-key providers: Ollama, Custom / OpenAI-compatible - Key storage: LOCAL_DEVICE | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Maintained (last commit about 1 month ago) Krita AI Diffusion is a GPL-3.0 open-source plugin for the Krita painting app that adds Stable Diffusion / Flux-based image generation: text-to-image, inpaint, outpaint, refine, live painting, ControlNet/IP-Adapter guidance, regions, and upscaling to 4K/8K. It runs on a ComfyUI backend. Users can let the plugin auto-install a managed local ComfyUI, or in the Configure > Connection tab choose "Custom Server" and paste any ComfyUI API URL to connect to their own local or remote/cloud-hosted backend (e.g. a self-hosted box, RunPod, or RunDiffusion). The author also offers a first-party managed cloud, Interstice Cloud, as a pay-as-you-go token option for users who don't want to run GPUs themselves. Key handling: Your API key is stored locally on your device. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 90/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 23/25: The key is held on your own device. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under GPL-3.0, so anyone can check how the key is handled. - Privacy 14/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-27, about 0 month(s) ago. (https://github.com/Acly/krita-ai-diffusion/commits) ### Airi - Page: https://www.byoaik.com/tools/airi/ - Website: https://airi.moeru.ai/ - Source: https://github.com/moeru-ai/airi - Category: AI Chat Interfaces | Pricing: OPEN_SOURCE | Open source: yes (MIT) | Trust score: 85/100 - Platforms: Web, Desktop, Mobile, Self-hosted, Docker - Bring-your-own-key providers: OpenAI, Anthropic, Google Gemini, Azure OpenAI, OpenRouter, Groq, Mistral, DeepSeek, xAI Grok, Ollama, Together AI, Fireworks, Custom / OpenAI-compatible - Key storage: LOCAL_DEVICE | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit today) Airi (moeru-ai/airi) is an open-source, self-hosted AI virtual companion ("cyber waifu," inspired by Neuro-sama) with VRM and Live2D avatars, real-time voice chat, and even game-playing (Minecraft, Factorio). It runs in the browser (Stage Web at airi.moeru.ai), as a desktop app for Windows/macOS/Linux (Stage Tamagotchi), and as a mobile PWA, plus self-hosted/containerized deployment. Users bring their own keys: the Settings panel (Providers → LLM Providers) exposes apiKey and baseURL fields, and the xsAI adapter supports 30-40+ LLM providers including OpenAI, Anthropic Claude, Google Gemini, DeepSeek, xAI Grok, Groq, Mistral, OpenRouter, Azure OpenAI, Together.ai, Fireworks.ai, Ollama, and any OpenAI-compatible endpoint. Voice/TTS is also multi-provider (ElevenLabs, Azure/Microsoft Speech, OpenAI-compatible TTS, Alibaba Cloud Model Studio, local Kokoro TTS). MIT licensed and free. Key handling: Your API key is stored locally on your device. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 90/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 23/25: The key is held on your own device. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under MIT, so anyone can check how the key is handled. - Privacy 14/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/moeru-ai/airi/blob/HEAD/README.md) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-29, about 0 month(s) ago. (https://github.com/moeru-ai/airi/commits) ### Readest - Page: https://www.byoaik.com/tools/readest/ - Website: https://readest.com/ - Source: https://github.com/readest/readest - Category: Productivity Tools | Pricing: OPEN_SOURCE | Open source: yes (AGPL-3.0) | Trust score: 85/100 - Platforms: Web, Desktop, Mobile, Self-hosted, Docker - Bring-your-own-key providers: OpenAI, OpenRouter, Groq, Together AI, DeepSeek, Ollama, Custom / OpenAI-compatible - Key storage: LOCAL_DEVICE | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit today) Readest is a modern, feature-rich open-source ebook reader (EPUB/PDF/Mobi) built on Next.js and Tauri v2, available on macOS, Windows, Linux, Android, iOS, and the web. It offers immersive reading with annotations, full-text search, TTS, dictionary/Wikipedia lookup, OPDS/Calibre and KOReader sync, and instant in-book translation via DeepL/Yandex. It ships a multi-provider AI subsystem (summaries, embeddings, assistant) where users can bring their own key: the OpenRouterProvider uses the AI SDK's createOpenAICompatible with a user-supplied API key and a configurable base URL that works with any OpenAI-compatible endpoint (OpenAI, OpenRouter, Together, Groq, vLLM/LiteLLM, etc.), plus a local Ollama provider with a configurable base URL. A settings AI panel exposes the API key, base URL, and model picker. A hosted AI Gateway and cloud sync are offered via optional Plus/Pro paid plans. Key handling: Your API key is stored locally on your device. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 90/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 23/25: The key is held on your own device. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under AGPL-3.0, so anyone can check how the key is handled. - Privacy 14/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [CONFIRMED via SOURCE_SCAN] Ships a container definition, so it can be self-hosted on your own infrastructure. (https://github.com/readest/readest/blob/HEAD/Dockerfile) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-29, about 0 month(s) ago. (https://github.com/readest/readest/commits) ### Local Deep Research - Page: https://www.byoaik.com/tools/local-deep-research/ - Website: https://github.com/LearningCircuit/local-deep-research - Source: https://github.com/LearningCircuit/local-deep-research - Category: Research Tools | Pricing: OPEN_SOURCE | Open source: yes (MIT) | Trust score: 85/100 - Platforms: Web, Self-hosted, CLI, Library/SDK, Docker - Bring-your-own-key providers: OpenAI, Anthropic, Google Gemini, OpenRouter, Ollama, Custom / OpenAI-compatible - Key storage: ENVIRONMENT_VARIABLE | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit today) Local Deep Research (LDR) is an open-source, self-hosted AI research assistant that runs autonomous multi-step "deep research" over many sources (web via SearXNG/Tavily/SerpAPI/Brave, academic via arXiv/PubMed/Semantic Scholar, Wikipedia, and your own private documents through LangChain retrievers), then synthesizes findings into reports with inline citations. It is designed to run entirely locally and encrypted, but supports a wide range of LLM backends. Users bring their own API keys for cloud LLMs (OpenAI, Anthropic Claude, Google Gemini, OpenRouter's 100+ models, plus any OpenAI- or Anthropic-compatible endpoint) and for premium search providers; local models via Ollama, LM Studio, and llama.cpp incur no cost. Keys are configured either through environment variables (e.g. LDR_LLM_OPENAI_API_KEY, LDR_LLM_ANTHROPIC_API_KEY) or directly in the web Settings UI (Settings -> LLM Provider -> select provider and enter credentials). It ships a web UI (localhost:5000), Docker images with GPU support, a CLI, a Python client library plus HTTP REST API, and an MCP server for Claude Desktop/Code integration. v1.7.0 (June 2026) added an interactive multi-turn Chat Mode with live citation streaming. Key handling: Your API key is supplied via an environment variable or local config file. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 90/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 22/25: The key is supplied by an environment variable or local config file you control. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under MIT, so anyone can check how the key is handled. - Privacy 15/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. The project states it sends no telemetry. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/LearningCircuit/local-deep-research/blob/HEAD/README.md) - [REPORTED via SOURCE_SCAN] The project states it sends no telemetry. (https://github.com/LearningCircuit/local-deep-research/blob/HEAD/README.md) - [CONFIRMED via SOURCE_SCAN] Ships a container definition, so it can be self-hosted on your own infrastructure. (https://github.com/LearningCircuit/local-deep-research/blob/HEAD/docker-compose.yml) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-29, about 0 month(s) ago. (https://github.com/LearningCircuit/local-deep-research/commits) ### Koharu - Page: https://www.byoaik.com/tools/koharu/ - Website: https://github.com/mayocream/koharu - Source: https://github.com/koharu-rs/koharu - Category: Productivity Tools | Pricing: OPEN_SOURCE | Open source: yes (GPL-3.0) | Trust score: 85/100 - Platforms: Desktop, Self-hosted, CLI, Docker - Bring-your-own-key providers: OpenAI, Anthropic, Google Gemini, DeepSeek, OpenRouter, Ollama, Custom / OpenAI-compatible - Key storage: LOCAL_DEVICE | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit 2 days ago) Koharu is a local-first, Rust + Tauri desktop application that automates manga translation by chaining object detection, OCR, inpainting (source-text removal), and an LLM translation step, then rendering vertical CJK / RTL text and exporting layered PSDs. It runs GGUF models locally via llama.cpp (CUDA/Vulkan/Metal) by default, but the koharu-llm crate ships explicit provider modules (openai.rs, claude.rs, gemini.rs, deepseek.rs, openai_compatible.rs) so a user can instead plug in their own hosted API key for OpenAI, Anthropic Claude, Google Gemini, DeepSeek, or any OpenAI-compatible endpoint (OpenRouter, LM Studio), plus machine-translation services (DeepL, Google Cloud Translation, Caiyun). API keys are stored in the OS keychain via the Rust `keyring` crate (set_password/get_password) rather than plaintext config, with a filesystem credential fallback on Linux. Ships an HTTP API and MCP server for automation, runs on Windows/macOS/Linux, and offers a Docker headless mode. Key handling: Your API key is stored locally on your device. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 90/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 23/25: The key is held on your own device. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under GPL-3.0, so anyone can check how the key is handled. - Privacy 14/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/koharu-rs/koharu/blob/HEAD/README.md) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-29, about 0 month(s) ago. (https://github.com/koharu-rs/koharu/commits) ### Meetily - Page: https://www.byoaik.com/tools/meetily/ - Website: https://meetily.ai/ - Source: https://github.com/Zackriya-Solutions/meetily - Category: Productivity Tools | Pricing: FREEMIUM | Open source: yes (MIT) | Trust score: 85/100 - Platforms: Desktop, Self-hosted, Docker - Bring-your-own-key providers: Anthropic, OpenAI, Groq, OpenRouter, Ollama, Custom / OpenAI-compatible - Key storage: LOCAL_DEVICE | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit 6 days ago) Meetily is an open-source, self-hosted AI meeting assistant (Tauri/Rust + Next.js) that records, transcribes, and summarizes meetings entirely on-device. Transcription runs locally via Whisper/Parakeet models with speaker diarization, and no audio leaves the machine. For summaries, users choose a local LLM through Ollama or bring their own provider API key for Anthropic Claude, Groq, OpenRouter, or OpenAI, including a Custom OpenAI-compatible Endpoint option. MIT licensed and free as a Community Edition, with a paid Pro tier (from $10/user/month) and an optional Hosted AI summary service. Distributed as native macOS and Windows installers; Linux is build-from-source. Repo has ~12.8k stars and is actively maintained (v0.4.0, June 2026). Key handling: Your API key is stored locally on your device. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 90/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 23/25: The key is held on your own device. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under MIT, so anyone can check how the key is handled. - Privacy 14/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/Zackriya-Solutions/meetily/blob/HEAD/README.md) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-10, about 1 month(s) ago. (https://github.com/Zackriya-Solutions/meetily/commits) ### Relaticle - Page: https://www.byoaik.com/tools/relaticle/ - Website: https://relaticle.com - Source: https://github.com/relaticle/relaticle - Category: Enterprise AI Tools | Pricing: OPEN_SOURCE | Open source: yes (AGPL-3.0) | Trust score: 85/100 - Platforms: Web, Self-hosted, Docker - Bring-your-own-key providers: Anthropic, Google Gemini, Custom / OpenAI-compatible - Key storage: SELF_HOSTED_ONLY | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit today) Relaticle is an open-source CRM covering contacts, deals and pipelines, built around an AI chat assistant and roughly thirty MCP tools so external agents can query and act on CRM data. The project states that inference runs through whichever provider the team configures, Anthropic, Google Gemini or any OpenAI-compatible endpoint, and that self-hosted teams supply their own provider keys, with conversation history kept in your own database. Self-hosted through Docker on a Laravel and Filament stack. AGPL-3.0, 1,502 stars, daily commits. Key handling: You self-host the app, so your API key lives in your own deployment. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 80/100 (assessable for any licence), verifiability 90/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 21/25: The key lives in a deployment you run yourself. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under AGPL-3.0, so anyone can check how the key is handled. Key handling was located in the published source. - Privacy 10/15: Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 7/10: Key handling was found in the published source by BYOAIK's scanner. What was checked (verification tier SOURCE_VERIFIED, derived from this evidence): - [STRONG via SOURCE_SCAN] The user supplies their own OpenAI key: the project references its API key variable. (https://github.com/relaticle/relaticle/blob/HEAD/.env.example) - [STRONG via SOURCE_SCAN] The user supplies their own Anthropic key: the project references its API key variable. (https://github.com/relaticle/relaticle/blob/HEAD/.env.example) - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/relaticle/relaticle/blob/HEAD/.env.example) - [CONFIRMED via SOURCE_SCAN] Ships a container definition, so it can be self-hosted on your own infrastructure. (https://github.com/relaticle/relaticle/blob/HEAD/Dockerfile) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-29, about 0 month(s) ago. (https://github.com/relaticle/relaticle/commits) ### OpenAI Completion (Sublime Text) - Page: https://www.byoaik.com/tools/openai-completion-sublime-text/ - Website: https://github.com/yaroslavyaroslav/OpenAI-sublime-text - Source: https://github.com/yaroslavyaroslav/OpenAI-sublime-text - Category: IDE Plugins | Pricing: OPEN_SOURCE | Open source: yes (MIT) | Trust score: 85/100 - Platforms: Desktop, Self-hosted, Docker - Bring-your-own-key providers: OpenAI, Anthropic, Google Gemini, Ollama - Key storage: LOCAL_DEVICE | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Maintained (last commit about 1 month ago) This Sublime Text package adds inline AI chat, completion and refactoring commands and is entirely bring-your-own-key: you add your own key to the plugin's settings file or point it at a local Ollama server, rather than using any bundled credential. The project advertises support for current OpenAI, Anthropic and Gemini models alongside Ollama. Distributed free through Package Control. MIT licensed, 466 stars, last commit within a month of review. Key handling: Your API key is stored locally on your device. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 90/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 23/25: The key is held on your own device. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under MIT, so anyone can check how the key is handled. - Privacy 14/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/yaroslavyaroslav/OpenAI-sublime-text/blob/HEAD/README.md) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-23, about 0 month(s) ago. (https://github.com/yaroslavyaroslav/OpenAI-sublime-text/commits) ### Moshi - Page: https://www.byoaik.com/tools/moshi/ - Website: https://kyutai.org/moshi - Source: https://github.com/kyutai-labs/moshi - Category: Music and Audio Generation | Pricing: OPEN_SOURCE | Open source: yes (Apache-2.0) | Trust score: 85/100 - Platforms: CLI, Desktop, Self-hosted, Local-only, Docker - Bring-your-own-key providers: Ollama, Custom / OpenAI-compatible - Key storage: NO_KEY_LOCAL_MODEL | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Slowing down (last commit about 6 months ago) Moshi from Kyutai is a real-time speech-to-speech model that listens and talks at the same time rather than taking turns, and it runs locally in PyTorch, MLX or Rust implementations. Voice assistants are the case where the hosted default is most uncomfortable, because the microphone is always the input; running the model on your own machine changes what that means. Apache-2.0, from a non-profit research lab. Key handling: Key handling varies. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 93/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 25/25: No cloud key is needed at all. The model runs locally. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under Apache-2.0, so anyone can check how the key is handled. - Privacy 14/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 8/10: Commits within the last six months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [CONFIRMED via SOURCE_SCAN] Ships a container definition, so it can be self-hosted on your own infrastructure. (https://github.com/kyutai-labs/moshi/blob/HEAD/docker-compose.yml) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-05-16, about 4 month(s) ago. (https://github.com/kyutai-labs/moshi/commits) ### LlamaCoder - Page: https://www.byoaik.com/tools/llamacoder/ - Website: https://llamacoder.together.ai/ - Source: https://github.com/Nutlope/llamacoder - Category: App Builders | Pricing: OPEN_SOURCE | Open source: yes (MIT) | Trust score: 85/100 - Platforms: Web, Self-hosted, Local-only, Docker - Bring-your-own-key providers: Together AI, Custom / OpenAI-compatible - Key storage: SELF_HOSTED_ONLY | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit 17 days ago) LlamaCoder turns a prompt into a working React application with live preview, an open alternative to v0 and Claude Artifacts built on open models. The key is the operator's: TOGETHER_API_KEY in the environment, with the app running wherever you run it, so neither the prompts nor the generated code need pass through a product you do not control. MIT. Key handling: You self-host the app, so your API key lives in your own deployment. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 80/100 (assessable for any licence), verifiability 90/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 21/25: The key lives in a deployment you run yourself. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under MIT, so anyone can check how the key is handled. Key handling was located in the published source. - Privacy 10/15: Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 7/10: Key handling was found in the published source by BYOAIK's scanner. What was checked (verification tier SOURCE_VERIFIED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [STRONG via SOURCE_SCAN] The user supplies their own Together AI key: the project references its API key variable. (https://github.com/Nutlope/llamacoder/blob/HEAD/README.md) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-15, about 0 month(s) ago. (https://github.com/Nutlope/llamacoder/commits) ### Stable Audio Tools - Page: https://www.byoaik.com/tools/stable-audio-tools/ - Website: https://github.com/Stability-AI/stable-audio-tools - Source: https://github.com/Stability-AI/stable-audio-tools - Category: Music and Audio Generation | Pricing: OPEN_SOURCE | Open source: yes (MIT (model weights licensed separately)) | Trust score: 85/100 - Platforms: CLI, Self-hosted, Local-only, Docker - Bring-your-own-key providers: Ollama, Custom / OpenAI-compatible - Key storage: NO_KEY_LOCAL_MODEL | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Slowing down (last commit about 5 months ago) Stable Audio Tools is Stability AI's open code for generative audio models, covering both inference and training, so it can generate sound and music and be fine-tuned on your own material. The code is MIT; model weights carry their own separate terms, which is the usual arrangement in this space and worth checking before commercial use. Everything runs on your hardware. Key handling: Key handling varies. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 93/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 25/25: No cloud key is needed at all. The model runs locally. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under MIT (model weights licensed separately), so anyone can check how the key is handled. - Privacy 14/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 8/10: Commits within the last six months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-05-26, about 4 month(s) ago. (https://github.com/Stability-AI/stable-audio-tools/commits) ### Maid - Page: https://www.byoaik.com/tools/maid/ - Website: https://github.com/Mobile-Artificial-Intelligence/maid - Source: https://github.com/Mobile-Artificial-Intelligence/maid - Category: AI Chat Interfaces | Pricing: OPEN_SOURCE | Open source: yes (MIT) | Trust score: 84/100 - Platforms: Mobile - Bring-your-own-key providers: OpenAI, Anthropic, DeepSeek, Mistral, Ollama - Key storage: LOCAL_DEVICE | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit 11 days ago) Maid is a free, open-source Android app for running llama.cpp GGUF models fully on-device, or connecting to Anthropic, DeepSeek, Mistral, Ollama and OpenAI models remotely. Keys are entered per remote provider inside the app, with no account and no Maid-hosted proxy involved; local inference needs no key and no internet connection at all. Distributed through Google Play and GitHub releases. MIT licensed, 2,629 stars, pushed within a month of review. Key handling: Your API key is stored locally on your device. Requests are sent directly to the AI provider. Scores: key handling 88/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 23/25: The key is held on your own device. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under MIT, so anyone can check how the key is handled. - Privacy 13/15: Local-first: it works without sending your data anywhere. The project states it sends no telemetry. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/Mobile-Artificial-Intelligence/maid/blob/HEAD/README.md) - [REPORTED via SOURCE_SCAN] The project states it sends no telemetry. (https://github.com/Mobile-Artificial-Intelligence/maid/blob/HEAD/README.md) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-14, about 1 month(s) ago. (https://github.com/Mobile-Artificial-Intelligence/maid/commits) ### Open Interpreter - Page: https://www.byoaik.com/tools/open-interpreter/ - Website: https://www.openinterpreter.com - Source: https://github.com/openinterpreter/openinterpreter - Category: Coding Assistants | Pricing: OPEN_SOURCE | Open source: yes (Apache-2.0) | Trust score: 84/100 - Platforms: CLI, Self-hosted, Library/SDK, Docker - Bring-your-own-key providers: OpenAI, Anthropic, OpenRouter, DeepSeek, Groq, Ollama, Custom / OpenAI-compatible - Key storage: ENVIRONMENT_VARIABLE | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit today) Open Interpreter is an open-source (Apache-2.0) coding agent that lets you drive an LLM in natural language from your terminal to write and execute code locally inside native sandboxing on macOS, Linux, and Windows. It is fully bring-your-own-key: you authenticate each provider with your own credentials via environment variables (OPENAI_API_KEY, ANTHROPIC_API_KEY, OPENROUTER_API_KEY, DEEPSEEK_API_KEY, GROQ_API_KEY, MOONSHOT_API_KEY, DASHSCOPE_API_KEY) or the --api_key / --api_base / --api_version CLI arguments, with config kept locally under ~/.openinterpreter. It ships built-in providers (OpenAI, Amazon Bedrock, Ollama, LM Studio) and generated providers (Anthropic, OpenRouter, DeepSeek, Moonshot/Kimi, Qwen/Alibaba, Groq, and more), and supports any custom OpenAI-compatible or Anthropic-compatible endpoint by setting a base_url and wire_api. The project has recently pivoted to a Rust-based agent emulating the Codex harness aimed at low-cost models, switchable from the TUI via /model, and is also usable as an Agent Client Protocol agent inside editors. Requests with your own key go directly from your device to the chosen provider. Verified against the README, the providers docs, and the all-settings docs. Key handling: Your API key is supplied via an environment variable or local config file. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 88/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 22/25: The key is supplied by an environment variable or local config file you control. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under Apache-2.0, so anyone can check how the key is handled. - Privacy 14/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-27, about 0 month(s) ago. (https://github.com/openinterpreter/openinterpreter/commits) ### Qwen Code - Page: https://www.byoaik.com/tools/qwen-code/ - Website: https://qwenlm.github.io/qwen-code-docs/ - Source: https://github.com/QwenLM/qwen-code - Category: Developer Infrastructure | Pricing: OPEN_SOURCE | Open source: yes (Apache-2.0) | Trust score: 84/100 - Platforms: CLI, Self-hosted, Docker - Bring-your-own-key providers: OpenAI, Anthropic, Google Gemini, Azure OpenAI, OpenRouter, Ollama, Custom / OpenAI-compatible - Key storage: ENVIRONMENT_VARIABLE | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit today) Qwen Code is an open-source, Apache-2.0 AI coding agent that lives in your terminal, forked from Google's Gemini CLI and tuned for Qwen3-Coder models. It supports an "API Key" authentication method that lets users bring their own provider API key: the docs explicitly cover connecting to OpenAI, Anthropic, Google Gemini, Azure OpenAI, OpenRouter, ModelScope, Alibaba Cloud Model Studio, or self-hosted endpoints (Ollama, vLLM, LM Studio) via OpenAI-compatible protocols. Credentials are supplied through environment variables referenced in ~/.qwen/settings.json (envKey) or a .env file, and are never persisted in settings (read at runtime from process.env). Multiple providers can be configured and switched at runtime with the /model command. The previously free Qwen OAuth tier was discontinued on 2026-04-15, explicitly steering users toward bring-your-own-key or paid plans. The CLI itself is free and open source; only provider/inference costs apply. Installable via npm (@qwen-code/qwen-code), currently at v0.18.4, actively maintained with ~25.4k GitHub stars. Key handling: Your API key is supplied via an environment variable or local config file. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 88/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 22/25: The key is supplied by an environment variable or local config file you control. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under Apache-2.0, so anyone can check how the key is handled. - Privacy 14/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/QwenLM/qwen-code/blob/HEAD/README.md) - [CONFIRMED via SOURCE_SCAN] Ships a container definition, so it can be self-hosted on your own infrastructure. (https://github.com/QwenLM/qwen-code/blob/HEAD/Dockerfile) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-29, about 0 month(s) ago. (https://github.com/QwenLM/qwen-code/commits) ### SWE-agent - Page: https://www.byoaik.com/tools/swe-agent/ - Website: https://swe-agent.com - Source: https://github.com/SWE-agent/SWE-agent - Category: Coding Assistants | Pricing: OPEN_SOURCE | Open source: yes (MIT) | Trust score: 84/100 - Platforms: CLI, Self-hosted, Docker - Bring-your-own-key providers: OpenAI, Anthropic, Together AI, Azure OpenAI, Ollama, Custom / OpenAI-compatible - Key storage: ENVIRONMENT_VARIABLE | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Slowing down (last commit about 4 months ago) SWE-agent (NeurIPS 2024) takes a GitHub issue or task description and autonomously edits, tests, and opens a pull request against a real repository through its Agent-Computer Interface. Run from the CLI, it authenticates from a keys.cfg file at the repo root containing lines such as OPENAI_API_KEY or ANTHROPIC_API_KEY, so requests go straight to the provider you choose, including local and open models. MIT licensed with 20,000+ stars and active maintenance by Princeton NLP. Key handling: Your API key is supplied via an environment variable or local config file. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 88/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 22/25: The key is supplied by an environment variable or local config file you control. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under MIT, so anyone can check how the key is handled. - Privacy 14/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-07-16, about 2 month(s) ago. (https://github.com/SWE-agent/SWE-agent/commits) ### Draw Things - Page: https://www.byoaik.com/tools/draw-things/ - Website: https://github.com/drawthingsai/draw-things-community - Source: https://github.com/drawthingsai/draw-things-community - Category: Image Generation Tools | Pricing: OPEN_SOURCE | Open source: yes (GPL-3.0) | Trust score: 84/100 - Platforms: Mobile, Desktop, Local-only - Bring-your-own-key providers: Ollama, Custom / OpenAI-compatible - Key storage: NO_KEY_LOCAL_MODEL | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit today) Draw Things runs image generation locally on Apple hardware, including iPhone and iPad, with the community edition published under GPL-3.0. It is one of the few options that puts the whole pipeline on a phone, so prompts and generated images never leave the device and no account or key is required. Notable for covering a platform where the alternatives are almost all hosted services. Key handling: Key handling varies. Requests are sent directly to the AI provider. Scores: key handling 88/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 25/25: No cloud key is needed at all. The model runs locally. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under GPL-3.0, so anyone can check how the key is handled. - Privacy 11/15: Local-first: it works without sending your data anywhere. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-29, about 0 month(s) ago. (https://github.com/drawthingsai/draw-things-community/commits) ### Letta - Page: https://www.byoaik.com/tools/letta/ - Website: https://www.letta.com/ - Source: https://github.com/letta-ai/letta - Category: Agent Frameworks | Pricing: OPEN_SOURCE | Open source: yes (Apache-2.0) | Trust score: 84/100 - Platforms: Self-hosted, CLI, Library/SDK, Web, Docker - Bring-your-own-key providers: OpenAI, Anthropic, Google Gemini, Azure OpenAI, OpenRouter, Groq, Mistral, DeepSeek, xAI Grok, Ollama, Together AI, Custom / OpenAI-compatible - Key storage: ENVIRONMENT_VARIABLE | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit 21 days ago) Letta (formerly MemGPT, from UC Berkeley's Sky Computing Lab) is an open-source platform for building stateful AI agents with advanced, self-editing tiered memory that persists and self-improves across sessions. It exposes a server/API plus Python and TypeScript SDKs and a CLI (Letta Code), and is fully model-agnostic. Users bring their own provider API keys via environment variables (e.g. OPENAI_API_KEY, ANTHROPIC_API_KEY, GEMINI_API_KEY) when self-hosting the Docker server, or via /connect in Letta Code. BYOK is documented as working on all cloud tiers with OpenAI, Anthropic, and OpenRouter keys. It also supports configurable OpenAI-compatible base URLs/proxy endpoints for local and alternative backends (Ollama, vLLM, LM Studio), plus enterprise backends like Azure OpenAI, AWS Bedrock, and Google Vertex. Key handling: Your API key is supplied via an environment variable or local config file. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 88/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 22/25: The key is supplied by an environment variable or local config file you control. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under Apache-2.0, so anyone can check how the key is handled. - Privacy 14/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-10, about 1 month(s) ago. (https://github.com/letta-ai/letta/commits) ### bolt.diy - Page: https://www.byoaik.com/tools/bolt-diy/ - Website: https://stackblitz-labs.github.io/bolt.diy/ - Source: https://github.com/stackblitz-labs/bolt.diy - Category: Coding Assistants | Pricing: OPEN_SOURCE | Open source: yes (MIT) | Trust score: 84/100 - Platforms: Web, Self-hosted, Docker - Bring-your-own-key providers: OpenAI, Anthropic, Google Gemini, OpenRouter, Groq, Mistral, Cohere, Perplexity, DeepSeek, xAI Grok, Ollama, Hugging Face, Together AI, Custom / OpenAI-compatible - Key storage: SELF_HOSTED_ONLY | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit 25 days ago) bolt.diy is the community-driven open-source fork of Bolt.new (originally started by Cole Medin, now maintained under stackblitz-labs). It lets you prompt, run, edit, and deploy full-stack web apps entirely in the browser, choosing the LLM per prompt. Users bring their own provider API keys, configured either through the in-app Providers settings tab (with real-time validation and secure browser-cookie storage) or via a .env.local file (e.g. ANTHROPIC_API_KEY=...). It supports 19+ providers including OpenAI, Anthropic, Google Gemini, Groq, xAI Grok, DeepSeek, Mistral, Cohere, Together, Perplexity, HuggingFace, OpenRouter, Amazon Bedrock, plus local runtimes (Ollama, LM Studio) and any OpenAI-compatible endpoint. It is MIT-licensed, self-hostable (run locally or deploy to Vercel/Netlify/Cloudflare), and bolt.diy itself does not resell model access. Latest milestone release v1.0.0 (May 2025) added Supabase, Vercel deployment, and Expo support; the repo shows active maintenance with 1,600+ commits. Key handling: You self-host the app, so your API key lives in your own deployment. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 87/100 (assessable for any licence), verifiability 90/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 21/25: The key lives in a deployment you run yourself. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under MIT, so anyone can check how the key is handled. Key handling was located in the published source. - Privacy 14/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 5/10: Last commit was within the last year. - Verification Confidence 7/10: Key handling was found in the published source by BYOAIK's scanner. What was checked (verification tier SOURCE_VERIFIED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [STRONG via SOURCE_SCAN] The user supplies their own OpenAI key: the project references its API key variable. (https://github.com/stackblitz-labs/bolt.diy/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] The user supplies their own Anthropic key: the project references its API key variable. (https://github.com/stackblitz-labs/bolt.diy/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/stackblitz-labs/bolt.diy/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] The user supplies their own Groq key: the project references its API key variable. (https://github.com/stackblitz-labs/bolt.diy/blob/HEAD/.env.example) - [STRONG via SOURCE_SCAN] The user supplies their own Mistral key: the project references its API key variable. (https://github.com/stackblitz-labs/bolt.diy/blob/HEAD/.env.example) - [STRONG via SOURCE_SCAN] The user supplies their own DeepSeek key: the project references its API key variable. (https://github.com/stackblitz-labs/bolt.diy/blob/HEAD/.env.example) - [STRONG via SOURCE_SCAN] The user supplies their own Together AI key: the project references its API key variable. (https://github.com/stackblitz-labs/bolt.diy/blob/HEAD/.env.example) - [STRONG via SOURCE_SCAN] The user supplies their own Cohere key: the project references its API key variable. (https://github.com/stackblitz-labs/bolt.diy/blob/HEAD/.env.example) - [STRONG via SOURCE_SCAN] The user supplies their own xAI Grok key: the project references its API key variable. (https://github.com/stackblitz-labs/bolt.diy/blob/HEAD/.env.example) - [STRONG via SOURCE_SCAN] The user supplies their own Fireworks key: the project references its API key variable. (https://github.com/stackblitz-labs/bolt.diy/blob/HEAD/.env.example) - [STRONG via SOURCE_SCAN] The user supplies their own Perplexity key: the project references its API key variable. (https://github.com/stackblitz-labs/bolt.diy/blob/HEAD/.env.example) - [CONFIRMED via SOURCE_SCAN] Ships a container definition, so it can be self-hosted on your own infrastructure. (https://github.com/stackblitz-labs/bolt.diy/blob/HEAD/docker-compose.yaml) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-02-07, about 8 month(s) ago. (https://github.com/stackblitz-labs/bolt.diy/commits) ### BeeAI Framework - Page: https://www.byoaik.com/tools/beeai-framework/ - Website: https://framework.beeai.dev - Source: https://github.com/i-am-bee/beeai-framework - Category: Agent Frameworks | Pricing: OPEN_SOURCE | Open source: yes (Apache-2.0) | Trust score: 84/100 - Platforms: Library/SDK, Self-hosted, CLI, Docker - Bring-your-own-key providers: OpenAI, Anthropic, Google Gemini, Azure OpenAI, Groq, Mistral, xAI Grok, Ollama, Hugging Face, Custom / OpenAI-compatible - Key storage: ENVIRONMENT_VARIABLE | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit 5 days ago) BeeAI Framework is an open-source (Apache-2.0) toolkit from IBM Research's i-am-bee project for building autonomous agents and multi-agent systems in both Python and TypeScript. Its unified "Backend" module abstracts over 13+ LLM providers (Ollama, OpenAI, Anthropic, watsonx.ai, Groq, Amazon Bedrock, Google Vertex, Azure OpenAI, xAI, Google Gemini, MistralAI, MiniMax, plus a LangChain adapter for anything else) behind a single ChatModel interface. It provides agents, tools, RAG, memory, workflows, and "requirement agents" for predictable behavior across models. Developers supply their own provider credentials via environment variables (OPENAI_API_KEY, ANTHROPIC_API_KEY, GROQ_API_KEY, GEMINI_API_KEY, etc.); the framework has no built-in key management or hosted gateway, so it is strictly bring-your-own-key. As a developer library/SDK it is consumed via pip/npm rather than as an end-user app. Actively maintained: latest Python release python_v0.1.81 (May 2026), commits through June 18 2026 including security dependency remediation, ~3.3k stars, 43 open issues, not archived. Key handling: Your API key is supplied via an environment variable or local config file. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 88/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 22/25: The key is supplied by an environment variable or local config file you control. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under Apache-2.0, so anyone can check how the key is handled. - Privacy 14/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/i-am-bee/beeai-framework/blob/HEAD/README.md) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-28, about 0 month(s) ago. (https://github.com/i-am-bee/beeai-framework/commits) ### PocketPal AI - Page: https://www.byoaik.com/tools/pocketpal-ai/ - Website: https://github.com/a-ghorbani/pocketpal-ai - Source: https://github.com/a-ghorbani/pocketpal-ai - Category: Local AI Tools | Pricing: OPEN_SOURCE | Open source: yes (MIT) | Trust score: 84/100 - Platforms: Mobile - Bring-your-own-key providers: Ollama, Hugging Face - Key storage: NO_KEY_LOCAL_MODEL | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit yesterday) PocketPal AI is a free, open-source mobile app that downloads and runs quantised GGUF models such as Gemma, Qwen, Phi and Llama directly on the device using llama.cpp, with on-device text to speech. The README states the model lives on your phone and conversations never leave it, so there is no key field: download and inference both happen locally with CPU, GPU or NPU acceleration and it works in airplane mode. Available on the App Store and Google Play as well as buildable from source, with no subscription or paid tier to unlock the AI. MIT licensed, 7,932 stars, pushed the day before review. Key handling: Key handling varies. Requests are sent directly to the AI provider. Scores: key handling 88/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 25/25: No cloud key is needed at all. The model runs locally. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under MIT, so anyone can check how the key is handled. - Privacy 11/15: Local-first: it works without sending your data anywhere. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/a-ghorbani/pocketpal-ai/blob/HEAD/README.md) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-28, about 0 month(s) ago. (https://github.com/a-ghorbani/pocketpal-ai/commits) ### Agent Zero - Page: https://www.byoaik.com/tools/agent-zero/ - Website: https://agent-zero.ai - Source: https://github.com/agent0ai/agent-zero - Category: Agent Frameworks | Pricing: OPEN_SOURCE | Open source: yes (MIT) | Trust score: 84/100 - Platforms: Self-hosted, Docker, Web - Bring-your-own-key providers: OpenAI, Anthropic, OpenRouter, Ollama, Custom / OpenAI-compatible - Key storage: ENVIRONMENT_VARIABLE | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit 7 days ago) Agent Zero is a general-purpose autonomous agent that spawns sub-agents, writes and executes code in a Dockerised Linux sandbox, and persists memory across sessions, usable for coding, research and automation tasks. Setup documents two BYOK paths: entering keys directly in the web UI settings page, or setting them in a .env file, for OpenAI, Anthropic, OpenRouter and other providers, plus native Ollama support for fully local models. Around 18,800 stars and active development. Key handling: Your API key is supplied via an environment variable or local config file. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 88/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 22/25: The key is supplied by an environment variable or local config file you control. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under MIT, so anyone can check how the key is handled. - Privacy 14/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/agent0ai/agent-zero/blob/HEAD/docs/README.md) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-23, about 0 month(s) ago. (https://github.com/agent0ai/agent-zero/commits) ### Windmill - Page: https://www.byoaik.com/tools/windmill/ - Website: https://www.windmill.dev - Source: https://github.com/windmill-labs/windmill - Category: Agent Frameworks | Pricing: OPEN_SOURCE | Open source: yes (AGPL-3.0 (community edition; mixed with Apache-2.0 for specs/APIs and a proprietary license for enterprise features)) | Trust score: 84/100 - Platforms: Web, Self-hosted, CLI, Docker - Bring-your-own-key providers: OpenAI, Anthropic, Google Gemini, Azure OpenAI, OpenRouter, Groq, Mistral, DeepSeek, Together AI, Ollama, Custom / OpenAI-compatible - Key storage: ENVIRONMENT_VARIABLE | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit today) Windmill is an open-source developer platform and workflow engine that turns scripts (Python, TypeScript, Go, Bash, SQL, and more) into webhooks, scheduled jobs, multi-step workflows, and auto-generated internal UIs/apps. It positions itself as an open-source alternative to Retool and Temporal. Its AI features (Windmill AI) provide code generation, flow generation, and AI agent steps inside workflows. AI is BYOK: an admin enables it in the workspace settings 'Windmill AI' tab by adding a model resource containing their own provider API key/endpoint (also configurable instance-wide by superadmins). Supported providers include OpenAI, Azure OpenAI, Anthropic, Mistral (Codestral), DeepSeek, Google Gemini, Groq, OpenRouter, Together AI, AWS Bedrock, and 'Custom AI' for any OpenAI-compatible base URL + API key (including local Ollama). Self-hosted Community Edition is free under AGPLv3; cloud and enterprise tiers add paid features. Key handling: Your API key is supplied via an environment variable or local config file. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 88/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 22/25: The key is supplied by an environment variable or local config file you control. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under AGPL-3.0 (community edition; mixed with Apache-2.0 for specs/APIs and a proprietary license for enterprise features), so anyone can check how the key is handled. - Privacy 14/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [CONFIRMED via SOURCE_SCAN] Ships a container definition, so it can be self-hosted on your own infrastructure. (https://github.com/windmill-labs/windmill/blob/HEAD/docker-compose.yml) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-29, about 0 month(s) ago. (https://github.com/windmill-labs/windmill/commits) ### SurfSense - Page: https://www.byoaik.com/tools/surfsense/ - Website: https://www.surfsense.com - Source: https://github.com/MODSetter/SurfSense - Category: Research Tools | Pricing: OPEN_SOURCE | Open source: yes (Apache-2.0) | Trust score: 84/100 - Platforms: Web, Self-hosted, Docker - Bring-your-own-key providers: OpenAI, Anthropic, Google Gemini, Groq, DeepSeek, Ollama, Custom / OpenAI-compatible - Key storage: SELF_HOSTED_ONLY | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit 5 days ago) SurfSense is a self-hostable, open-source research and knowledge assistant positioned as a private alternative to NotebookLM, Perplexity, and Glean. It connects your own LLM to internal documents and external sources (search engines like Tavily/LinkUp/SearxNG, plus Slack, Notion, Linear, Jira, Confluence, GitHub, YouTube, Google Drive and more) and answers with cited hybrid (semantic + full-text) search, generating reports, podcasts, and chats over your knowledge base. It is BYOK by design: the dashboard lets users create custom LLM configurations by entering their own API credentials, selecting models, and defining custom API base URLs, with all calls routed through LiteLLM for 100+ providers. A CUSTOM/custom_provider option supports any OpenAI-compatible endpoint, and it can run fully local with no OpenAI key via Ollama or vLLM. It is deployed via Docker/Docker Compose or manual installation on your own infrastructure. Verified via GitHub API (15,033 stars, Apache-2.0, not archived, pushed 2026-06-20, v0.0.29) and corroborated by README, the official site, and codebase analysis of the LLM config/API-key management system. Key handling: You self-host the app, so your API key lives in your own deployment. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 88/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 21/25: The key lives in a deployment you run yourself. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under Apache-2.0, so anyone can check how the key is handled. - Privacy 15/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. The project states it sends no telemetry. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/MODSetter/SurfSense/blob/HEAD/README.md) - [REPORTED via SOURCE_SCAN] The project states it sends no telemetry. (https://github.com/MODSetter/SurfSense/blob/HEAD/README.md) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-29, about 0 month(s) ago. (https://github.com/MODSetter/SurfSense/commits) ### Pi - Page: https://www.byoaik.com/tools/pi/ - Website: https://github.com/earendil-works/pi - Source: https://github.com/earendil-works/pi - Category: Developer Infrastructure | Pricing: OPEN_SOURCE | Open source: yes (MIT) | Trust score: 84/100 - Platforms: CLI, Library/SDK, Self-hosted, Docker - Bring-your-own-key providers: OpenAI, Anthropic, Google Gemini, Azure OpenAI, OpenRouter, Groq, Mistral, DeepSeek, xAI Grok, Ollama, Hugging Face, Together AI, Fireworks, Custom / OpenAI-compatible - Key storage: ENVIRONMENT_VARIABLE | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit today) Pi (by earendil-works) is an MIT-licensed AI agent toolkit published as a TypeScript monorepo of npm packages: pi-ai (unified multi-provider LLM API), pi-agent-core (agent runtime/loop with tool calling), pi-tui (terminal UI library with differential rendering), and pi-coding-agent (an interactive coding-agent CLI with read/bash/edit/write tools and session management). The coding agent is installable via npm as @earendil-works/pi-coding-agent and connects to LLM providers through the unified pi-ai layer. BYOK is a first-class, free path: users supply their own provider credentials via environment variables (ANTHROPIC_API_KEY, OPENAI_API_KEY, GEMINI_API_KEY, GROQ_API_KEY, MISTRAL_API_KEY, XAI_API_KEY, OPENROUTER_API_KEY, DEEPSEEK_API_KEY, CEREBRAS_API_KEY, NVIDIA_API_KEY, AZURE_OPENAI_API_KEY, HF_TOKEN, TOGETHER_API_KEY, FIREWORKS_API_KEY, CLOUDFLARE_API_KEY, etc.) or via a CLI --api-key flag, and pay providers directly with no payment to pi.dev. Local and custom OpenAI-compatible models (Ollama, LM Studio, vLLM) are added via a models.json file; Amazon Bedrock and Google Vertex AI are supported through AWS/GCP auth. An optional 'unified billing' hosted offering exists via Cloudflare AI Gateway, but it is fully optional and BYOK runs the tool for free. The project is very actively maintained (release v0.79.9 on 2026-06-20) and highly popular (64k+ stars, 7.8k+ forks, 100+ contributors). Key handling: Your API key is supplied via an environment variable or local config file. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 88/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 22/25: The key is supplied by an environment variable or local config file you control. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under MIT, so anyone can check how the key is handled. - Privacy 14/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-29, about 0 month(s) ago. (https://github.com/earendil-works/pi/commits) ### CrewAI - Page: https://www.byoaik.com/tools/crewai/ - Website: https://www.crewai.com/ - Source: https://github.com/crewAIInc/crewAI - Category: Agent Frameworks | Pricing: OPEN_SOURCE | Open source: yes (MIT) | Trust score: 84/100 - Platforms: Library/SDK, CLI, Self-hosted, Docker - Bring-your-own-key providers: OpenAI, Anthropic, Google Gemini, Azure OpenAI, OpenRouter, Groq, Mistral, Perplexity, Ollama, Hugging Face, Fireworks, Custom / OpenAI-compatible - Key storage: ENVIRONMENT_VARIABLE | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit today) CrewAI is a lean, standalone Python framework (independent of LangChain) for building multi-agent systems where role-playing autonomous agents collaborate to complete tasks via "crews" and "flows." It is fully model-agnostic: the user brings their own LLM provider API key, supplied through environment variables (OPENAI_API_KEY, ANTHROPIC_API_KEY, GOOGLE_API_KEY, etc.), a YAML config, or the Python LLM class. Through native integrations plus LiteLLM it supports a very broad set of providers and lets users point at any OpenAI-compatible or self-hosted endpoint by setting a custom base_url, including local models via Ollama. It is a developer library/SDK installed via pip and run self-hosted; a separate paid hosted "CrewAI Enterprise/AMP" platform also exists, but the open-source framework itself requires no CrewAI account and uses only the user's own provider keys. Key handling: Your API key is supplied via an environment variable or local config file. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 88/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 22/25: The key is supplied by an environment variable or local config file you control. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under MIT, so anyone can check how the key is handled. - Privacy 14/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/crewAIInc/crewAI/blob/HEAD/README.md) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-29, about 0 month(s) ago. (https://github.com/crewAIInc/crewAI/commits) ### LLM - Page: https://www.byoaik.com/tools/llm/ - Website: https://llm.datasette.io/ - Source: https://github.com/simonw/llm - Category: Developer Infrastructure | Pricing: OPEN_SOURCE | Open source: yes (Apache-2.0) | Trust score: 84/100 - Platforms: CLI, Library/SDK, Self-hosted, Docker - Bring-your-own-key providers: OpenAI, Anthropic, Google Gemini, OpenRouter, Mistral, Groq, Ollama, Custom / OpenAI-compatible - Key storage: ENVIRONMENT_VARIABLE | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit 3 days ago) LLM is an open-source command-line tool and Python library by Simon Willison for interacting with OpenAI, Anthropic Claude, Google Gemini, Llama and dozens of other large language models. It is BYOK by design: users store their own provider API keys with commands like `llm keys set openai` or `llm keys set anthropic` (saved locally in keys.json), and an extensive plugin system (llm-anthropic, llm-gemini, llm-mistral, llm-openrouter, llm-groq, llm-ollama, and more) lets users add their own keys for additional providers or point at local/self-hosted models. OpenAI-compatible endpoints with a custom base URL are supported via an extra-openai-models.yaml config, enabling any OpenAI-API-compatible service (e.g. LocalAI, OpenRouter). It works as both a CLI and a Python SDK and supports prompts, chat, embeddings, and logging to SQLite. Key handling: Your API key is supplied via an environment variable or local config file. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 88/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 22/25: The key is supplied by an environment variable or local config file you control. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under Apache-2.0, so anyone can check how the key is handled. - Privacy 14/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/simonw/llm/blob/HEAD/README.md) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-22, about 0 month(s) ago. (https://github.com/simonw/llm/commits) ### Fabric - Page: https://www.byoaik.com/tools/fabric/ - Website: https://github.com/danielmiessler/fabric - Source: https://github.com/danielmiessler/fabric - Category: Developer Infrastructure | Pricing: OPEN_SOURCE | Open source: yes (MIT) | Trust score: 84/100 - Platforms: CLI, Web, Self-hosted, Library/SDK, Docker - Bring-your-own-key providers: OpenAI, Anthropic, Google Gemini, Azure OpenAI, OpenRouter, Groq, Mistral, DeepSeek, Perplexity, Together AI, Ollama, Custom / OpenAI-compatible - Key storage: ENVIRONMENT_VARIABLE | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit today) Fabric is an open-source (MIT) command-line framework by Daniel Miessler that organizes AI prompts into reusable, version-controlled "patterns" (e.g. summarize, analyze code, extract wisdom) runnable from the CLI, a built-in web UI, or REST API. The `fabric --setup` interactive wizard initializes config and lets users enter their OWN API keys for each vendor; if no vendor is specified, it auto-detects based on model name and the keys you've provided. It supports 20+ providers through a three-tier architecture: native SDK integrations (OpenAI, Anthropic/Claude, Google Gemini, Azure OpenAI, Amazon Bedrock, Vertex AI), an OpenAI-compatible layer for vendors like Groq, Mistral, DeepSeek, Together, OpenRouter, and Perplexity (150+ providers via openai_compatible plugins with configurable base URLs), plus local models via Ollama. Runs on macOS, Linux, and Windows. Actively maintained with frequent releases (v1.4.455 on Jun 9, 2026; 408 total releases) and continuous provider updates including Claude Opus 4.7 support. Key handling: Your API key is supplied via an environment variable or local config file. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 88/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 22/25: The key is supplied by an environment variable or local config file you control. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under MIT, so anyone can check how the key is handled. - Privacy 14/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/danielmiessler/fabric/blob/HEAD/README.md) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-29, about 0 month(s) ago. (https://github.com/danielmiessler/fabric/commits) ### assistant-ui - Page: https://www.byoaik.com/tools/assistant-ui/ - Website: https://www.assistant-ui.com - Source: https://github.com/assistant-ui/assistant-ui - Category: Agent Frameworks | Pricing: OPEN_SOURCE | Open source: yes (MIT) | Trust score: 84/100 - Platforms: Web, Mobile, CLI, Self-hosted, Library/SDK, Docker - Bring-your-own-key providers: OpenAI, Anthropic, Google Gemini, Azure OpenAI, OpenRouter, Mistral, Perplexity, Ollama, Fireworks, Custom / OpenAI-compatible - Key storage: ENVIRONMENT_VARIABLE | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit yesterday) assistant-ui is an open-source (MIT) TypeScript/React library of composable primitives and prebuilt components for building ChatGPT-style chat interfaces, copilots, and agent UIs. It is backend-agnostic: developers wire it to their own server and supply their own provider API keys. Broad model support comes via the Vercel AI SDK (OpenAI, Anthropic, Google Gemini, Mistral, Perplexity, AWS Bedrock, Azure, Fireworks, Ollama), plus LangChain/LangGraph, custom HTTP transports, and OpenAI-compatible gateways (OpenRouter, LiteLLM, Portkey). Features include streaming, message editing/branching, attachments, markdown/LaTeX rendering, auto-scroll, and speech-to-text, built on shadcn/ui and Tailwind. It spans Web (React 18+), React Native (Expo), and React Ink (CLI). The library is free; an optional managed "assistant-cloud" backend adds thread persistence and chat history (Free / Pro $50/mo / Enterprise, billed by MAU) and offers encrypted per-provider credential storage. Y Combinator-backed; used by Mastra, LangChain, and Helicone. Note: it is a developer library, not a turnkey end-user app, so the party bringing the key is the developer integrating it. Key handling: Your API key is supplied via an environment variable or local config file. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 88/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 22/25: The key is supplied by an environment variable or local config file you control. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under MIT, so anyone can check how the key is handled. - Privacy 14/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/assistant-ui/assistant-ui/blob/HEAD/README.md) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-29, about 0 month(s) ago. (https://github.com/assistant-ui/assistant-ui/commits) ### Axolotl - Page: https://www.byoaik.com/tools/axolotl/ - Website: https://github.com/axolotl-ai-cloud/axolotl - Source: https://github.com/axolotl-ai-cloud/axolotl - Category: Fine-Tuning and Training | Pricing: OPEN_SOURCE | Open source: yes (Apache-2.0) | Trust score: 84/100 - Platforms: CLI, Self-hosted, Docker - Bring-your-own-key providers: Ollama, Custom / OpenAI-compatible - Key storage: NO_KEY_LOCAL_MODEL | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit today) Axolotl configures fine-tuning through declarative YAML rather than code, which makes runs reproducible and reviewable, and it is the option teams reach for when training has to span multiple GPUs via FSDP or DeepSpeed. Training happens on infrastructure you provide, so no key is involved and the dataset never leaves it. Apache-2.0. Key handling: Key handling varies. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 88/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 25/25: No cloud key is needed at all. The model runs locally. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under Apache-2.0, so anyone can check how the key is handled. - Privacy 11/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. Analytics or error-reporting libraries are present; what they send was not established. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [CONFIRMED via SOURCE_SCAN] Ships a container definition, so it can be self-hosted on your own infrastructure. (https://github.com/axolotl-ai-cloud/axolotl/blob/HEAD/docker-compose.yaml) - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/axolotl-ai-cloud/axolotl/blob/HEAD/pyproject.toml) - [REPORTED via SOURCE_SCAN] References PostHog, so some analytics or error reporting is present. This scan cannot tell whether it is opt-in or what it sends. (https://github.com/axolotl-ai-cloud/axolotl/blob/HEAD/pyproject.toml) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-28, about 0 month(s) ago. (https://github.com/axolotl-ai-cloud/axolotl/commits) ### LangGraph - Page: https://www.byoaik.com/tools/langgraph/ - Website: https://langchain-ai.github.io/langgraph/ - Source: https://github.com/langchain-ai/langgraph - Category: Agent Frameworks | Pricing: OPEN_SOURCE | Open source: yes (MIT) | Trust score: 84/100 - Platforms: CLI, Self-hosted, Local-only, Docker - Bring-your-own-key providers: OpenAI, Anthropic, Google Gemini, OpenRouter, Ollama, Custom / OpenAI-compatible - Key storage: ENVIRONMENT_VARIABLE | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit 2 days ago) LangGraph models agents as explicit graphs with persistent state, so long-running workflows can be checkpointed, resumed and inspected rather than living inside one opaque loop. Model access is the caller's: you configure whichever provider you hold credentials for, including local endpoints, and the library runs in your own process. Now the piece of the LangChain ecosystem most production deployments actually build on. MIT. Key handling: Your API key is supplied via an environment variable or local config file. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 88/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 22/25: The key is supplied by an environment variable or local config file you control. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under MIT, so anyone can check how the key is handled. - Privacy 14/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-27, about 0 month(s) ago. (https://github.com/langchain-ai/langgraph/commits) ### DSPy - Page: https://www.byoaik.com/tools/dspy/ - Website: https://dspy.ai - Source: https://github.com/stanfordnlp/dspy - Category: Prompt Engineering Tools | Pricing: OPEN_SOURCE | Open source: yes (MIT) | Trust score: 84/100 - Platforms: Library/SDK, CLI, Self-hosted, Docker - Bring-your-own-key providers: OpenAI, Anthropic, Google Gemini, Groq, Together AI, Mistral, DeepSeek, Ollama, Custom / OpenAI-compatible - Key storage: ENVIRONMENT_VARIABLE | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit today) DSPy is a framework from Stanford NLP for building and optimising LLM pipelines programmatically, compiling declarative modules into effective prompts instead of manual tweaking. You configure a model with your own key directly in code or through provider environment variables, and all calls run from your environment through LiteLLM with no DSPy-hosted proxy. Distributed as a Python library. MIT licensed, 37,229 stars, last commit the day of review. Key handling: Your API key is supplied via an environment variable or local config file. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 88/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 22/25: The key is supplied by an environment variable or local config file you control. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under MIT, so anyone can check how the key is handled. - Privacy 14/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-27, about 0 month(s) ago. (https://github.com/stanfordnlp/dspy/commits) ### marimo - Page: https://www.byoaik.com/tools/marimo/ - Website: https://marimo.io/ - Source: https://github.com/marimo-team/marimo - Category: Developer Infrastructure | Pricing: OPEN_SOURCE | Open source: yes (Apache-2.0) | Trust score: 84/100 - Platforms: Web, CLI, Local-only, Self-hosted, Docker - Bring-your-own-key providers: OpenAI, Anthropic, Google Gemini, Ollama, Custom / OpenAI-compatible - Key storage: ENVIRONMENT_VARIABLE | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit today) marimo is a reactive Python notebook that stores notebooks as ordinary .py files, which makes them diffable, reviewable and runnable as scripts rather than opaque JSON. Its AI assistance is configured with an API key you provide, and it can be pointed at a local model. The file format is the quiet advantage here: what a notebook sends and does is readable in a diff rather than buried in cell metadata. Apache-2.0. Key handling: Your API key is supplied via an environment variable or local config file. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 88/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 22/25: The key is supplied by an environment variable or local config file you control. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under Apache-2.0, so anyone can check how the key is handled. - Privacy 14/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-29, about 0 month(s) ago. (https://github.com/marimo-team/marimo/commits) ### PydanticAI - Page: https://www.byoaik.com/tools/pydanticai/ - Website: https://ai.pydantic.dev/ - Source: https://github.com/pydantic/pydantic-ai - Category: Agent Frameworks | Pricing: OPEN_SOURCE | Open source: yes (MIT) | Trust score: 84/100 - Platforms: CLI, Self-hosted, Local-only, Docker - Bring-your-own-key providers: OpenAI, Anthropic, Google Gemini, OpenRouter, Ollama, Custom / OpenAI-compatible - Key storage: ENVIRONMENT_VARIABLE | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit today) PydanticAI brings the validation discipline of Pydantic to agents: structured, type-checked outputs rather than hoping a model returns parseable text. Models are addressed as plain strings such as anthropic:claude or openai:gpt, so swapping provider is a one-line change and local endpoints are first-class. Validation matters here beyond ergonomics, because a typed boundary is one more place a malformed or hostile model response gets caught before it reaches your code. MIT. Key handling: Your API key is supplied via an environment variable or local config file. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 88/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 22/25: The key is supplied by an environment variable or local config file you control. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under MIT, so anyone can check how the key is handled. - Privacy 14/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/pydantic/pydantic-ai/blob/HEAD/README.md) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-29, about 0 month(s) ago. (https://github.com/pydantic/pydantic-ai/commits) ### DocsGPT - Page: https://www.byoaik.com/tools/docsgpt/ - Website: https://www.docsgpt.cloud - Source: https://github.com/arc53/DocsGPT - Category: Document Analysis | Pricing: OPEN_SOURCE | Open source: yes (MIT) | Trust score: 84/100 - Platforms: Self-hosted, Docker, Web, CLI - Bring-your-own-key providers: OpenAI, Anthropic, Google Gemini, Azure OpenAI, Ollama, Custom / OpenAI-compatible - Key storage: ENVIRONMENT_VARIABLE | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit today) DocsGPT is an open-source platform for chatting with, searching and building agents over your own documents, with an agent builder and a deep-research mode. Its quickstart has a 'Connect Cloud API Provider' step which states you will need an API key from your chosen provider, entered through the .env file or in-app settings, and the README confirms it works with OpenAI, Google and Anthropic as well as local models via Ollama and llama.cpp. Self-hosted through Docker. MIT licensed, 18,206 stars, pushed the same day as review. Key handling: Your API key is supplied via an environment variable or local config file. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 88/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 22/25: The key is supplied by an environment variable or local config file you control. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under MIT, so anyone can check how the key is handled. - Privacy 14/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/arc53/DocsGPT/blob/HEAD/README.md) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-29, about 0 month(s) ago. (https://github.com/arc53/DocsGPT/commits) ### txtai - Page: https://www.byoaik.com/tools/txtai/ - Website: https://neuml.github.io/txtai/ - Source: https://github.com/neuml/txtai - Category: Developer Infrastructure | Pricing: OPEN_SOURCE | Open source: yes (Apache-2.0) | Trust score: 84/100 - Platforms: CLI, Self-hosted, Local-only, Docker - Bring-your-own-key providers: OpenAI, Anthropic, Hugging Face, Ollama, Custom / OpenAI-compatible - Key storage: ENVIRONMENT_VARIABLE | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit 2 days ago) txtai is an embeddings database for semantic search, RAG and language model workflows. Its default posture is local: embeddings and many pipelines run on your own machine with models pulled to disk, and cloud providers are something you opt into with your own key rather than the path of least resistance. That ordering is unusual and is the reason it suits work where documents should not leave the building. Apache-2.0. Key handling: Your API key is supplied via an environment variable or local config file. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 88/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 22/25: The key is supplied by an environment variable or local config file you control. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under Apache-2.0, so anyone can check how the key is handled. - Privacy 14/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/neuml/txtai/blob/HEAD/README.md) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-29, about 0 month(s) ago. (https://github.com/neuml/txtai/commits) ### Agno - Page: https://www.byoaik.com/tools/agno/ - Website: https://agno.com/ - Source: https://github.com/agno-agi/agno - Category: Agent Frameworks | Pricing: OPEN_SOURCE | Open source: yes (Apache-2.0) | Trust score: 84/100 - Platforms: CLI, Self-hosted, Docker - Bring-your-own-key providers: OpenAI, Anthropic, Google Gemini, OpenRouter, Ollama, Custom / OpenAI-compatible - Key storage: ENVIRONMENT_VARIABLE | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit today) Agno builds multi-agent systems with memory, knowledge and tool use, aiming at low overhead per agent. It is model-agnostic, running on whichever provider you configure. Worth knowing before you adopt it: the README states Agno sends a telemetry event per agent run so the maintainers can see which providers to prioritise, that prompts, messages and outputs are never included, and that AGNO_TELEMETRY=false turns it off. Declaring that in the README rather than burying it is the behaviour you want from a framework. Apache-2.0. Key handling: Your API key is supplied via an environment variable or local config file. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 88/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 22/25: The key is supplied by an environment variable or local config file you control. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under Apache-2.0, so anyone can check how the key is handled. - Privacy 14/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-29, about 0 month(s) ago. (https://github.com/agno-agi/agno/commits) ### TensorRT-LLM - Page: https://www.byoaik.com/tools/tensorrt-llm/ - Website: https://nvidia.github.io/TensorRT-LLM/ - Source: https://github.com/NVIDIA/TensorRT-LLM - Category: Self-Hosted AI Platforms | Pricing: OPEN_SOURCE | Open source: yes (Apache-2.0) | Trust score: 84/100 - Platforms: Self-hosted, Docker, CLI - Bring-your-own-key providers: Ollama, Custom / OpenAI-compatible - Key storage: ENVIRONMENT_VARIABLE | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit today) TensorRT-LLM compiles language models into optimised TensorRT engines and serves them with in-flight batching and quantisation, aimed at getting maximum throughput from NVIDIA hardware. It runs where your GPUs are, so requests and keys never leave your infrastructure. Licensed Apache 2.0: GitHub reports the repository as unlicensed, but the LICENSE file states plainly that the project is under Apache 2.0 and then reproduces it. Expect a real engineering investment; this is the most demanding option in the category. Key handling: Your API key is supplied via an environment variable or local config file. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 88/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 22/25: The key is supplied by an environment variable or local config file you control. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under Apache-2.0, so anyone can check how the key is handled. - Privacy 14/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-29, about 0 month(s) ago. (https://github.com/NVIDIA/TensorRT-LLM/commits) ### BentoML - Page: https://www.byoaik.com/tools/bentoml/ - Website: https://bentoml.com/ - Source: https://github.com/bentoml/BentoML - Category: Self-Hosted AI Platforms | Pricing: OPEN_SOURCE | Open source: yes (Apache-2.0) | Trust score: 84/100 - Platforms: Self-hosted, Docker, CLI - Bring-your-own-key providers: Ollama, Custom / OpenAI-compatible - Key storage: ENVIRONMENT_VARIABLE | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit 10 days ago) BentoML turns models and inference pipelines into standardised services with batching, autoscaling and an OpenAI-compatible surface for language models, then deploys them to your own Kubernetes, cloud account or hardware. Because the endpoint is yours, a BYOK client can be pointed at it with no third-party key anywhere in the path. The company sells a managed cloud built on the same framework, which is a different arrangement from self-hosting it. Apache-2.0. Key handling: Your API key is supplied via an environment variable or local config file. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 88/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 22/25: The key is supplied by an environment variable or local config file you control. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under Apache-2.0, so anyone can check how the key is handled. - Privacy 14/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-07, about 1 month(s) ago. (https://github.com/bentoml/BentoML/commits) ### LMDeploy - Page: https://www.byoaik.com/tools/lmdeploy/ - Website: https://github.com/InternLM/lmdeploy - Source: https://github.com/InternLM/lmdeploy - Category: Self-Hosted AI Platforms | Pricing: OPEN_SOURCE | Open source: yes (Apache-2.0) | Trust score: 84/100 - Platforms: Self-hosted, Docker, CLI - Bring-your-own-key providers: Ollama, Custom / OpenAI-compatible - Key storage: ENVIRONMENT_VARIABLE | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit 2 days ago) LMDeploy covers quantisation, KV-cache optimisation and high-throughput serving for language and vision-language models, exposing an OpenAI-compatible endpoint from hardware you run. Its quantisation work is the practical point: it is what lets a large model fit on GPUs an organisation already owns rather than ones it has to rent, which is what makes self-hosting a real option rather than a principle. Apache-2.0, from the Shanghai AI Laboratory's InternLM team. Key handling: Your API key is supplied via an environment variable or local config file. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 88/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 22/25: The key is supplied by an environment variable or local config file you control. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under Apache-2.0, so anyone can check how the key is handled. - Privacy 14/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/InternLM/lmdeploy/blob/HEAD/README.md) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-28, about 0 month(s) ago. (https://github.com/InternLM/lmdeploy/commits) ### AppFlowy AI - Page: https://www.byoaik.com/tools/appflowy-ai/ - Website: https://appflowy.com/ - Source: https://github.com/AppFlowy-IO/AppFlowy - Category: Writing Tools | Pricing: FREEMIUM | Open source: yes (AGPL-3.0) | Trust score: 83/100 - Platforms: Desktop, Mobile, Self-hosted, Docker - Bring-your-own-key providers: OpenAI, Anthropic, Ollama, Custom / OpenAI-compatible - Key storage: LOCAL_DEVICE | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit 12 days ago) AppFlowy is an open-source (AGPL-3.0) collaborative workspace and Notion alternative built in Flutter/Rust, available as desktop and mobile apps plus a self-hostable backend. Its AI features (writing assistance, summarization, chat over your docs, generation) are BYOK in several ways: AppFlowy Local AI lets you point the desktop app at your own local models installed through Ollama (gemma3, qwen3, Llama 3.1, DeepSeek R1, Phi4, etc.) so data stays fully on-device, and this was made free for everyone. Beyond local models, self-hosted and cloud setups let you supply your own OpenAI API key, switch to Anthropic Claude in AI Settings, or configure a custom OpenAI-compatible base URL with an optional API key, enabling any OpenAI-compatible provider. Key handling: Your API key is stored locally on your device. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 90/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 23/25: The key is held on your own device. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under AGPL-3.0, so anyone can check how the key is handled. - Privacy 14/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 8/10: Commits within the last six months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-06-26, about 3 month(s) ago. (https://github.com/AppFlowy-IO/AppFlowy/commits) ### SillyTavern - Page: https://www.byoaik.com/tools/sillytavern/ - Website: https://sillytavern.app/ - Source: https://github.com/SillyTavern/SillyTavern - Category: AI Chat Interfaces | Pricing: OPEN_SOURCE | Open source: yes (AGPL-3.0) | Trust score: 83/100 - Platforms: Self-hosted, Web, Docker - Bring-your-own-key providers: OpenAI, Anthropic, Google Gemini, OpenRouter, Mistral, DeepSeek, Cohere, Perplexity, xAI Grok, Groq, Fireworks, Ollama, Custom / OpenAI-compatible - Key storage: SELF_HOSTED_ONLY | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Maintained (last commit about 2 months ago) SillyTavern is a free, open-source (AGPL-3.0), locally-installed/self-hosted web frontend for interacting with large language models, originally forked from TavernAI in 2023 and now a large community project. It provides a single unified interface for both cloud and local LLM backends and is heavily used for character-based chat and roleplay, with advanced features like character cards, world info/lorebooks, group chats, prompt/preset management, extensions, and optional image generation and TTS. It hosts no AI service itself and tracks no user data, so users bring their own API keys (or run fully local models). Under Chat Completion mode users pick a source (OpenAI, Claude/Anthropic, Google AI Studio/Vertex, OpenRouter, Mistral, DeepSeek, Cohere, Perplexity, xAI Grok, Groq, Fireworks, AI21, etc.) and paste their own key, or choose "Custom (OpenAI-compatible)" to set an arbitrary base URL plus key for services like LM Studio, LiteLLM, LocalAI, and Ollama. Actively maintained with frequent releases (v1.18.0 in May 2026). Key handling: You self-host the app, so your API key lives in your own deployment. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 87/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 21/25: The key lives in a deployment you run yourself. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under AGPL-3.0, so anyone can check how the key is handled. - Privacy 14/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [CONFIRMED via SOURCE_SCAN] Ships a container definition, so it can be self-hosted on your own infrastructure. (https://github.com/SillyTavern/SillyTavern/blob/HEAD/Dockerfile) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-14, about 0 month(s) ago. (https://github.com/SillyTavern/SillyTavern/commits) ### Vane - Page: https://www.byoaik.com/tools/perplexica/ - Website: https://github.com/ItzCrazyKns/Perplexica - Source: https://github.com/ItzCrazyKns/Vane - Category: Research Tools | Pricing: OPEN_SOURCE | Open source: yes (MIT) | Trust score: 83/100 - Platforms: Web, Self-hosted, Docker - Bring-your-own-key providers: OpenAI, Anthropic, Google Gemini, Groq, Ollama, Custom / OpenAI-compatible - Key storage: SELF_HOSTED_ONLY | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Maintained (last commit about 2 months ago) Formerly released as Perplexica; the project was renamed to Vane in 2026 and the old repository redirects. Perplexica (rebranded to "Vane" in 2026 by the same maintainer, URL still resolves) is an MIT-licensed, self-hostable AI answering engine and Perplexity AI alternative. It searches the web via a bundled SearXNG instance, optionally reranks results with embeddings, and generates cited answers with an LLM of your choice. It ships as a single Docker image (bundles SearXNG) and is configured entirely through a web setup wizard where you paste your own provider API keys. It is fully provider-agnostic: run local models via Ollama/LM Studio/Lemonade, or bring your own OpenAI, Anthropic Claude, Google Gemini, or Groq key. It also supports any OpenAI-compatible base URL (vLLM, LocalAI, OpenRouter, DeepSeek, etc.). Features include Speed/Balanced/Quality search modes, file uploads, search history, image/video search, and it can be used as a browser search engine. Note: it ships with no auth layer, so an exposed instance can let anyone consume your API credits. Key handling: You self-host the app, so your API key lives in your own deployment. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 87/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 21/25: The key lives in a deployment you run yourself. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under MIT, so anyone can check how the key is handled. - Privacy 14/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/ItzCrazyKns/Vane/blob/HEAD/README.md) - [CONFIRMED via SOURCE_SCAN] Ships a container definition, so it can be self-hosted on your own infrastructure. (https://github.com/ItzCrazyKns/Vane/blob/HEAD/docker-compose.yaml) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-01, about 1 month(s) ago. (https://github.com/ItzCrazyKns/Vane/commits) ### Mastra - Page: https://www.byoaik.com/tools/mastra/ - Website: https://mastra.ai/ - Source: https://github.com/mastra-ai/mastra - Category: Agent Frameworks | Pricing: OPEN_SOURCE | Open source: yes (Apache-2.0 (core); Mastra Enterprise License (enterprise add-ons)) | Trust score: 83/100 - Platforms: Library/SDK, Self-hosted, Docker - Bring-your-own-key providers: OpenAI, Anthropic, Google Gemini, Azure OpenAI, OpenRouter, Groq, Mistral, DeepSeek, xAI Grok, Ollama, Together AI, Fireworks, Custom / OpenAI-compatible - Key storage: SELF_HOSTED_ONLY | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit today) Mastra is an open-source TypeScript framework for building AI agents, copilots, and graph-based workflows with human-in-the-loop, memory, RAG, evals, and observability. Created by the Gatsby team (YC W25), it reached v1.0 in January 2026 and integrates with React/Next.js/Node or deploys as a standalone Hono server. Its model routing connects to 90+ providers through one standard interface (built on the Vercel AI SDK plus Mastra's own model gateway), so developers supply their own provider API keys (typically via environment variables) for OpenAI, Anthropic, Gemini and many others. It supports configurable OpenAI-compatible base URLs and local models via Ollama, making BYOK first-class. The core framework is Apache 2.0; only enterprise add-ons are source-available under a separate license. Key handling: You self-host the app, so your API key lives in your own deployment. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 87/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 21/25: The key lives in a deployment you run yourself. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under Apache-2.0 (core); Mastra Enterprise License (enterprise add-ons), so anyone can check how the key is handled. - Privacy 14/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-29, about 0 month(s) ago. (https://github.com/mastra-ai/mastra/commits) ### Mattermost Agents - Page: https://www.byoaik.com/tools/mattermost-agents/ - Website: https://github.com/mattermost/mattermost-plugin-agents - Source: https://github.com/mattermost/mattermost-plugin-agents - Category: Enterprise AI Tools | Pricing: OPEN_SOURCE | Open source: yes (Apache-2.0) | Trust score: 83/100 - Platforms: Self-hosted, Docker, Web - Bring-your-own-key providers: OpenAI, Anthropic, Azure OpenAI, Ollama, Custom / OpenAI-compatible - Key storage: SELF_HOSTED_ONLY | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit 4 days ago) The Agents plugin adds AI assistants to a self-hosted Mattermost deployment, with the administrator supplying the model access. It supports local models through Ollama or vLLM, cloud providers including OpenAI, Anthropic and Azure OpenAI, and any OpenAI-compatible endpoint, with multiple assistants each carrying their own prompt. Because Mattermost is self-hosted, both the collaboration data and the provider credentials stay inside infrastructure the organisation controls. Apache-2.0. Key handling: You self-host the app, so your API key lives in your own deployment. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 87/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 21/25: The key lives in a deployment you run yourself. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under Apache-2.0, so anyone can check how the key is handled. - Privacy 14/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/mattermost/mattermost-plugin-agents/blob/HEAD/README.md) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-29, about 0 month(s) ago. (https://github.com/mattermost/mattermost-plugin-agents/commits) ### RAGFlow - Page: https://www.byoaik.com/tools/ragflow/ - Website: https://ragflow.io - Source: https://github.com/infiniflow/ragflow - Category: Document Analysis | Pricing: OPEN_SOURCE | Open source: yes (Apache-2.0) | Trust score: 83/100 - Platforms: Web, Self-hosted, Docker - Bring-your-own-key providers: OpenAI, Anthropic, Azure OpenAI, Google Gemini, DeepSeek, Groq, Mistral, Cohere, Perplexity, xAI Grok, Together AI, Ollama, Hugging Face, Custom / OpenAI-compatible - Key storage: SELF_HOSTED_ONLY | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit today) RAGFlow is a leading open-source Retrieval-Augmented Generation engine built on deep document understanding (OCR, layout and table parsing) that delivers grounded, citation-backed question answering over your documents, fused with agent capabilities. Users self-host via Docker and configure AI providers through an in-app Model providers page (avatar > Model providers), pasting their own API key and optional base URL. It supports 60+ model vendors, so end users bring their own keys for OpenAI, Anthropic, Azure OpenAI, DeepSeek, Groq, Mistral, Gemini, Cohere, Together, xAI, Perplexity, local Ollama/Xinference/LocalAI, and any OpenAI-compatible endpoint. A managed cloud tier (cloud.ragflow.io) also exists, but the self-hosted open-source build is fully BYOK. Key handling: You self-host the app, so your API key lives in your own deployment. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 87/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 21/25: The key lives in a deployment you run yourself. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under Apache-2.0, so anyone can check how the key is handled. - Privacy 14/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [CONFIRMED via SOURCE_SCAN] Ships a container definition, so it can be self-hosted on your own infrastructure. (https://github.com/infiniflow/ragflow/blob/HEAD/Dockerfile) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-29, about 0 month(s) ago. (https://github.com/infiniflow/ragflow/commits) ### Omni - Page: https://www.byoaik.com/tools/omni/ - Website: https://github.com/getomnico/omni - Source: https://github.com/getomnico/omni - Category: Enterprise AI Tools | Pricing: OPEN_SOURCE | Open source: yes (Apache-2.0) | Trust score: 83/100 - Platforms: Self-hosted, Docker, Web - Bring-your-own-key providers: OpenAI, Anthropic, Google Gemini, AWS Bedrock, Azure OpenAI, Ollama, LM Studio, Custom / OpenAI-compatible - Key storage: SELF_HOSTED_ONLY | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit 4 days ago) Omni is a self-hosted AI agent for a company's workplace tools, finding context, analysing information and carrying out tasks across connected systems. It runs entirely in your cloud, on-premises or in an isolated environment, and supports Anthropic, OpenAI, Gemini, AWS Bedrock, Vertex AI, Azure AI Foundry or any OpenAI-compatible endpoint including vLLM, Ollama, LM Studio and LiteLLM. Extensible through MCP and connector SDKs in Python or TypeScript. Apache-2.0. Key handling: You self-host the app, so your API key lives in your own deployment. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 87/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 21/25: The key lives in a deployment you run yourself. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under Apache-2.0, so anyone can check how the key is handled. - Privacy 14/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/getomnico/omni/blob/HEAD/README.md) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-28, about 0 month(s) ago. (https://github.com/getomnico/omni/commits) ### Botpress - Page: https://www.byoaik.com/tools/botpress/ - Website: https://botpress.com - Source: https://github.com/botpress/botpress - Category: Automation Tools | Pricing: OPEN_SOURCE | Open source: yes (MIT) | Trust score: 83/100 - Platforms: Web, Self-hosted, Docker - Bring-your-own-key providers: OpenAI, Anthropic, Google Gemini, Ollama - Key storage: SELF_HOSTED_ONLY | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit 2 days ago) Botpress is an open-source platform for building and deploying LLM-powered chatbots and agents, offered as a hosted cloud and a self-hostable version. In self-hosted deployments an administrator adds their own provider key, OpenAI, Anthropic, Google Gemini or a local model, in the bot's settings, giving control over model choice and token cost instead of a bundled per-message rate. Deployable through Docker on your own infrastructure. MIT licensed, 14,869 stars, last commit within a day of review. Key handling: You self-host the app, so your API key lives in your own deployment. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 87/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 21/25: The key lives in a deployment you run yourself. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under MIT, so anyone can check how the key is handled. - Privacy 14/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [CONFIRMED via SOURCE_SCAN] Ships a container definition, so it can be self-hosted on your own infrastructure. (https://github.com/botpress/botpress/blob/HEAD/Dockerfile) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-29, about 0 month(s) ago. (https://github.com/botpress/botpress/commits) ### Inbox Zero - Page: https://www.byoaik.com/tools/inbox-zero/ - Website: https://github.com/elie222/inbox-zero - Source: https://github.com/elie222/inbox-zero - Category: Productivity Tools | Pricing: OPEN_SOURCE | Open source: yes (AGPL-3.0 (with additional commercial/enterprise restrictions)) | Trust score: 83/100 - Platforms: Web, Self-hosted, Docker - Bring-your-own-key providers: OpenAI, Anthropic, Google Gemini, Azure OpenAI, OpenRouter, Groq, Ollama, Custom / OpenAI-compatible - Key storage: SELF_HOSTED_ONLY | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit today) Inbox Zero (elie222/inbox-zero) is a source-available, self-hostable AI personal assistant for email. It connects to Gmail/Outlook and uses LLMs to organize mail, pre-draft replies, automate plain-English rules, track replies, bulk unsubscribe/archive, block cold emails, and generate analytics. It is a Next.js/TypeScript app with Postgres/Prisma, deployable via Docker, Vercel, or local dev. BYOK is confirmed: when self-hosting, the end user supplies their own LLM provider API key. The repo's apps/web/.env.example exposes per-role provider:model config (DEFAULT_LLMS, ECONOMY_LLMS, CHAT_LLMS, NANO_LLMS, DRAFT_LLMS) plus LLM_API_KEY and provider-specific credentials for Azure OpenAI (AZURE_RESOURCE_NAME/AZURE_FOUNDRY_*), Google Vertex (GOOGLE_VERTEX_*), AWS Bedrock (BEDROCK_ACCESS_KEY/SECRET_KEY), Ollama (OLLAMA_BASE_URL/MODEL), and a generic OpenAI-compatible endpoint (OPENAI_COMPATIBLE_BASE_URL/MODEL/AUTH_HEADER). Documented providers include OpenAI, Anthropic, Google Gemini/Vertex, Azure OpenAI, OpenRouter, Groq, AWS Bedrock, Ollama, and any OpenAI-compatible API. A hosted SaaS at getinboxzero.com also exists (paid), but the self-hosted path is fully BYOK. License is AGPL-3.0 with added commercial-use/enterprise restrictions (org with 5+ business users needs an enterprise license; no reselling), so it is source-available rather than permissively open source. Key handling: You self-host the app, so your API key lives in your own deployment. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 87/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 21/25: The key lives in a deployment you run yourself. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under AGPL-3.0 (with additional commercial/enterprise restrictions), so anyone can check how the key is handled. - Privacy 14/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [CONFIRMED via SOURCE_SCAN] Ships a container definition, so it can be self-hosted on your own infrastructure. (https://github.com/elie222/inbox-zero/blob/HEAD/docker-compose.yml) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-29, about 0 month(s) ago. (https://github.com/elie222/inbox-zero/commits) ### CodeCompanion.nvim - Page: https://www.byoaik.com/tools/codecompanion-nvim/ - Website: https://github.com/olimorris/codecompanion.nvim - Source: https://github.com/olimorris/codecompanion.nvim - Category: Coding Assistants | Pricing: OPEN_SOURCE | Open source: yes (Apache-2.0) | Trust score: 83/100 - Platforms: Library/SDK, Self-hosted, Docker - Bring-your-own-key providers: OpenAI, Anthropic, Google Gemini, Azure OpenAI, OpenRouter, Mistral, DeepSeek, xAI Grok, Hugging Face, Ollama, Custom / OpenAI-compatible - Key storage: SELF_HOSTED_ONLY | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit today) CodeCompanion.nvim is a Neovim plugin that brings LLM-powered coding into the editor with a chat buffer, inline code transformations, and agentic workflows/tools. It is built around a bring-your-own-LLM adapter system: users supply their own API keys (via environment variables or shell commands) and can target any OpenAI-compatible endpoint through a dedicated openai_compatible adapter with a configurable base URL, or write fully custom adapters. Ships with out-of-the-box adapters for Anthropic, OpenAI, Azure OpenAI, Google Gemini, DeepSeek, Mistral, OpenRouter, xAI, Hugging Face, GitHub Copilot, GitHub Models, Novita, and Ollama/local models. Key handling: You self-host the app, so your API key lives in your own deployment. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 87/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 21/25: The key lives in a deployment you run yourself. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under Apache-2.0, so anyone can check how the key is handled. - Privacy 14/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/olimorris/codecompanion.nvim/blob/HEAD/README.md) - [CONFIRMED via SOURCE_SCAN] Ships a container definition, so it can be self-hosted on your own infrastructure. (https://github.com/olimorris/codecompanion.nvim/blob/HEAD/Dockerfile) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-29, about 0 month(s) ago. (https://github.com/olimorris/codecompanion.nvim/commits) ### Jupyter AI - Page: https://www.byoaik.com/tools/jupyter-ai/ - Website: https://jupyter-ai.readthedocs.io/ - Source: https://github.com/jupyterlab/jupyter-ai - Category: Education Tools | Pricing: OPEN_SOURCE | Open source: yes (BSD-3-Clause) | Trust score: 83/100 - Platforms: Web, Local-only, Self-hosted, Docker - Bring-your-own-key providers: OpenAI, Anthropic, Hugging Face, Ollama, Custom / OpenAI-compatible - Key storage: SELF_HOSTED_ONLY | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit 2 days ago) Jupyter AI brings a chat panel and magic commands into JupyterLab and Notebook, with a provider abstraction so the model behind them is configurable rather than fixed. It matters for this directory because Jupyter is where a great deal of teaching, coursework and research analysis happens, and those notebooks routinely contain unpublished data. Official Project Jupyter subproject, BSD-3-Clause. Key handling: You self-host the app, so your API key lives in your own deployment. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 87/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 21/25: The key lives in a deployment you run yourself. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under BSD-3-Clause, so anyone can check how the key is handled. - Privacy 14/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-29, about 0 month(s) ago. (https://github.com/jupyterlab/jupyter-ai/commits) ### Langflow - Page: https://www.byoaik.com/tools/langflow/ - Website: https://www.langflow.org/ - Source: https://github.com/langflow-ai/langflow - Category: Agent Frameworks | Pricing: OPEN_SOURCE | Open source: yes (MIT) | Trust score: 83/100 - Platforms: Web, Self-hosted, Library/SDK, Docker - Bring-your-own-key providers: OpenAI, Anthropic, Google Gemini, Azure OpenAI, Groq, Mistral, Cohere, Ollama, Custom / OpenAI-compatible - Key storage: SELF_HOSTED_ONLY | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit 5 days ago) Langflow is an open-source, low-code visual framework for building multi-agent and RAG applications by wiring components together on a drag-and-drop node canvas. Users bring their own AI provider keys: each model/agent component has an API Key field where you paste your provider's key and pick the model, and v1.8+ added a centralized Model Providers settings pane to configure a provider and key once for the whole instance. Keys can also be stored as reusable "Credential" global variables or supplied via environment variables. It supports OpenAI, Anthropic, and Google natively in the model dropdown, local models via Ollama and LM Studio, additional providers (Azure OpenAI, Cohere, Mistral, NVIDIA, etc.) through bundles, and any OpenAI-compatible endpoint via custom components. It can be self-hosted (pip/Docker), run as a Python library/SDK, and exposes flows as APIs. Key handling: You self-host the app, so your API key lives in your own deployment. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 87/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 21/25: The key lives in a deployment you run yourself. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under MIT, so anyone can check how the key is handled. - Privacy 14/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/langflow-ai/langflow/blob/HEAD/pyproject.toml) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-29, about 0 month(s) ago. (https://github.com/langflow-ai/langflow/commits) ### Sim - Page: https://www.byoaik.com/tools/sim/ - Website: https://www.sim.ai/ - Source: https://github.com/simstudioai/sim - Category: Agent Frameworks | Pricing: FREEMIUM | Open source: yes (Apache-2.0) | Trust score: 83/100 - Platforms: Web, Self-hosted, Docker - Bring-your-own-key providers: OpenAI, Anthropic, Google Gemini, Azure OpenAI, OpenRouter, Groq, Mistral, Cohere, DeepSeek, xAI Grok, Ollama, Fireworks, Custom / OpenAI-compatible - Key storage: SELF_HOSTED_ONLY | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit today) Sim (formerly Sim Studio) is an open-source AI agent platform where you build, deploy, and orchestrate agents on a lightweight visual canvas by connecting blocks, or conversationally and via code. It supports 15+ LLM providers and explicitly documents Bring Your Own Key: "you can use your own API keys at base provider pricing with no markup." Users plug in their own keys for OpenAI, Anthropic, Google Gemini, Mistral, xAI Grok, Groq, DeepSeek, OpenRouter, Cohere, Fireworks, and others, and can point at self-hosted OpenAI-compatible endpoints via Ollama (OLLAMA_URL), vLLM (VLLM_BASE_URL), and LiteLLM (LITELLM_BASE_URL), plus Azure OpenAI/Anthropic endpoints. It runs as a hosted cloud product (free Community tier with paid Pro/Max/Enterprise) or fully self-hosted via `npx simstudio` or Docker Compose, requiring PostgreSQL with pgvector. Apache-2.0 licensed, ~28k GitHub stars, very actively maintained (v0.7.10 released June 2026, 5,000+ commits). Key handling: You self-host the app, so your API key lives in your own deployment. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 87/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 21/25: The key lives in a deployment you run yourself. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under Apache-2.0, so anyone can check how the key is handled. - Privacy 14/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/simstudioai/sim/blob/HEAD/README.md) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-29, about 0 month(s) ago. (https://github.com/simstudioai/sim/commits) ### Xinference - Page: https://www.byoaik.com/tools/xinference/ - Website: https://github.com/xorbitsai/inference - Source: https://github.com/xorbitsai/inference - Category: Self-Hosted AI Platforms | Pricing: OPEN_SOURCE | Open source: yes (Apache-2.0) | Trust score: 83/100 - Platforms: Self-hosted, Docker, Local-only - Bring-your-own-key providers: OpenAI, Custom / OpenAI-compatible, Ollama - Key storage: SELF_HOSTED_ONLY | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit today) Xorbits Inference runs language, embedding, reranking, image and speech models on your own infrastructure and exposes them through an OpenAI-compatible API, plus a web UI and clients. Because the models run where you put them, a BYOK client can be pointed at it with no cloud provider key in the path. It can also front third-party providers when you want them, using credentials you supply. Apache-2.0. Key handling: You self-host the app, so your API key lives in your own deployment. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 87/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 21/25: The key lives in a deployment you run yourself. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under Apache-2.0, so anyone can check how the key is handled. - Privacy 14/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/xorbitsai/inference/blob/HEAD/README.md) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-29, about 0 month(s) ago. (https://github.com/xorbitsai/inference/commits) ### LlamaIndex - Page: https://www.byoaik.com/tools/llamaindex/ - Website: https://www.llamaindex.ai - Source: https://github.com/run-llama/llama_index - Category: Agent Frameworks | Pricing: OPEN_SOURCE | Open source: yes (MIT) | Trust score: 83/100 - Platforms: Library/SDK, Self-hosted - Bring-your-own-key providers: OpenAI, Anthropic, Google Gemini, OpenRouter, Ollama, Custom / OpenAI-compatible, AWS Bedrock, Mistral, Cohere, Hugging Face - Key storage: ENVIRONMENT_VARIABLE | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit today) LlamaIndex is a framework for connecting LLMs to your own data and building retrieval-augmented agents, including coding-assistant and research use cases. Each LLM integration is configured by the user either via an environment variable such as ANTHROPIC_API_KEY or by passing api_key directly to the client constructor, so LlamaIndex never brokers or pools a key on the user's behalf. Works with standard provider APIs plus Bedrock and Vertex AI deployments. MIT licensed with 50,000+ stars. Key handling: Your API key is supplied via an environment variable or local config file. Requests are sent directly to the AI provider. Scores: key handling 77/100 (assessable for any licence), verifiability 90/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 22/25: The key is supplied by an environment variable or local config file you control. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under MIT, so anyone can check how the key is handled. Key handling was located in the published source. - Privacy 7/15: Nothing specific is known about its data handling beyond the key itself. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 7/10: Key handling was found in the published source by BYOAIK's scanner. What was checked (verification tier SOURCE_VERIFIED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [STRONG via SOURCE_SCAN] The user supplies their own OpenAI key: the project references its API key variable. (https://github.com/run-llama/llama_index/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/run-llama/llama_index/blob/HEAD/README.md) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-29, about 0 month(s) ago. (https://github.com/run-llama/llama_index/commits) ### OpenAgent - Page: https://www.byoaik.com/tools/openagent/ - Website: https://github.com/the-open-agent/openagent - Source: https://github.com/the-open-agent/openagent - Category: Knowledge Management | Pricing: OPEN_SOURCE | Open source: yes (Apache-2.0) | Trust score: 83/100 - Platforms: Web, Self-hosted, Docker - Bring-your-own-key providers: OpenAI, Anthropic, Google Gemini, DeepSeek, Mistral, Cohere, xAI Grok, Azure OpenAI, AWS Bedrock, OpenRouter, Ollama, Custom / OpenAI-compatible - Key storage: SELF_HOSTED_ONLY | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit 2 days ago) OpenAgent is a self-hostable personal AI assistant combining language models, a RAG knowledge base built from your own documents, and autonomous agent loops that can browse the web, run code and call MCP-compatible tools. The README documents more than thirty model providers, switchable per conversation without code changes, covering OpenAI, Anthropic, Gemini, DeepSeek, Mistral, Cohere, Qwen, Grok, Azure, Bedrock, OpenRouter and Ollama. It also tracks token consumption and cost per provider, model and user, so spend against your own key stays visible. Apache-2.0. Key handling: You self-host the app, so your API key lives in your own deployment. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 87/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 21/25: The key lives in a deployment you run yourself. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under Apache-2.0, so anyone can check how the key is handled. - Privacy 14/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/the-open-agent/openagent/blob/HEAD/README.md) - [CONFIRMED via SOURCE_SCAN] Ships a container definition, so it can be self-hosted on your own infrastructure. (https://github.com/the-open-agent/openagent/blob/HEAD/docker-compose.yml) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-29, about 0 month(s) ago. (https://github.com/the-open-agent/openagent/commits) ### parrot.nvim - Page: https://www.byoaik.com/tools/parrot-nvim/ - Website: https://github.com/frankroeder/parrot.nvim - Source: https://github.com/frankroeder/parrot.nvim - Category: IDE Plugins | Pricing: OPEN_SOURCE | Open source: yes (MIT) | Trust score: 83/100 - Platforms: CLI, Local-only - Bring-your-own-key providers: OpenAI, Anthropic, Google Gemini, OpenRouter, Ollama, Custom / OpenAI-compatible - Key storage: ENVIRONMENT_VARIABLE | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit 28 days ago) parrot.nvim brings chat, rewriting and code actions into Neovim across Anthropic, OpenAI, Gemini, Mistral, Groq, Ollama and OpenAI-compatible endpoints. Its useful detail for key handling is that an API key can be supplied by a shell command rather than a literal string, so the secret can live in a password manager or keychain instead of a dotfile that gets committed by accident. MIT, established by reading the LICENSE because GitHub reports the repository as unlicensed. Key handling: Your API key is supplied via an environment variable or local config file. Requests are sent directly to the AI provider. Scores: key handling 77/100 (assessable for any licence), verifiability 90/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 22/25: The key is supplied by an environment variable or local config file you control. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under MIT, so anyone can check how the key is handled. Key handling was located in the published source. - Privacy 7/15: Nothing specific is known about its data handling beyond the key itself. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 7/10: Key handling was found in the published source by BYOAIK's scanner. What was checked (verification tier SOURCE_VERIFIED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [STRONG via SOURCE_SCAN] The user supplies their own OpenAI key: the project references its API key variable. (https://github.com/frankroeder/parrot.nvim/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] The user supplies their own Anthropic key: the project references its API key variable. (https://github.com/frankroeder/parrot.nvim/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] The user supplies their own Google Gemini key: the project references its API key variable. (https://github.com/frankroeder/parrot.nvim/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] The user supplies their own xAI Grok key: the project references its API key variable. (https://github.com/frankroeder/parrot.nvim/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] The user supplies their own Perplexity key: the project references its API key variable. (https://github.com/frankroeder/parrot.nvim/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] Reads the API key from an environment variable at runtime. (https://github.com/frankroeder/parrot.nvim/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/frankroeder/parrot.nvim/blob/HEAD/README.md) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-08-24, about 1 month(s) ago. (https://github.com/frankroeder/parrot.nvim/commits) ### Opik - Page: https://www.byoaik.com/tools/opik/ - Website: https://www.comet.com/site/products/opik/ - Source: https://github.com/comet-ml/opik - Category: Prompt Engineering Tools | Pricing: OPEN_SOURCE | Open source: yes (Apache-2.0) | Trust score: 83/100 - Platforms: Self-hosted, Docker, Web - Bring-your-own-key providers: OpenAI, Anthropic, Google Gemini, OpenRouter, Ollama, Custom / OpenAI-compatible - Key storage: SELF_HOSTED_ONLY | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit 2 days ago) Opik traces, evaluates and monitors LLM applications, with datasets, LLM-as-a-judge metrics and a production dashboard. It matters here because tracing means recording prompts and outputs: run it self-hosted and those stay in your deployment, run the vendor's cloud and they do not. Judge models use credentials you supply. Apache-2.0, from Comet, which also sells the hosted version. Key handling: You self-host the app, so your API key lives in your own deployment. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 87/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 21/25: The key lives in a deployment you run yourself. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under Apache-2.0, so anyone can check how the key is handled. - Privacy 14/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/comet-ml/opik/blob/HEAD/README.md) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-29, about 0 month(s) ago. (https://github.com/comet-ml/opik/commits) ### WrenAI - Page: https://www.byoaik.com/tools/wrenai/ - Website: https://getwren.ai/ - Source: https://github.com/Canner/WrenAI - Category: Document Analysis | Pricing: OPEN_SOURCE | Open source: yes (Apache-2.0 (multi-licensed per path)) | Trust score: 83/100 - Platforms: Web, Self-hosted, Docker - Bring-your-own-key providers: OpenAI, Anthropic, Google Gemini, Ollama, Custom / OpenAI-compatible - Key storage: SELF_HOSTED_ONLY | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit today) WrenAI puts a semantic layer in front of your databases so questions can be answered in SQL that respects definitions your team agreed on, with the model reached on your own credentials and 22-plus connectors for the data side. The engine in the repository, covering the semantic layer, text-to-SQL, MCP server, CLI and connectors, is Apache-2.0 and self-hostable including air-gapped; a cloud and an enterprise tier are sold separately. Licensing is per-path rather than one blanket grant, which GitHub reports as unlicensed: the LICENSE file was read directly to establish that core, SDK, skills, examples and root files are Apache-2.0 and documentation is CC BY 4.0. Key handling: You self-host the app, so your API key lives in your own deployment. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 87/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 21/25: The key lives in a deployment you run yourself. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under Apache-2.0 (multi-licensed per path), so anyone can check how the key is handled. - Privacy 14/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-29, about 0 month(s) ago. (https://github.com/Canner/WrenAI/commits) ### KServe - Page: https://www.byoaik.com/tools/kserve/ - Website: https://kserve.github.io/website/ - Source: https://github.com/kserve/kserve - Category: Self-Hosted AI Platforms | Pricing: OPEN_SOURCE | Open source: yes (Apache-2.0) | Trust score: 83/100 - Platforms: Self-hosted, Docker - Bring-your-own-key providers: Ollama, Custom / OpenAI-compatible - Key storage: SELF_HOSTED_ONLY | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit 2 days ago) KServe is the CNCF standard for serving models on Kubernetes, covering autoscaling to zero, canary rollouts and an OpenAI-compatible runtime for language models. It matters to this directory as the layer that turns a self-hosted model from a demonstration into something an organisation can actually depend on, with the endpoint inside its own cluster and no provider key anywhere. Apache-2.0, vendor-neutral under the CNCF rather than a single company. Key handling: You self-host the app, so your API key lives in your own deployment. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 87/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 21/25: The key lives in a deployment you run yourself. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under Apache-2.0, so anyone can check how the key is handled. - Privacy 14/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/kserve/kserve/blob/HEAD/README.md) - [CONFIRMED via SOURCE_SCAN] Ships a container definition, so it can be self-hosted on your own infrastructure. (https://github.com/kserve/kserve/blob/HEAD/Dockerfile) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-29, about 0 month(s) ago. (https://github.com/kserve/kserve/commits) ### Napkins - Page: https://www.byoaik.com/tools/napkins/ - Website: https://www.napkins.dev/ - Source: https://github.com/Nutlope/napkins - Category: App Builders | Pricing: OPEN_SOURCE | Open source: yes (MIT) | Trust score: 83/100 - Platforms: Web, Self-hosted, Local-only, Docker - Bring-your-own-key providers: Together AI, Custom / OpenAI-compatible - Key storage: SELF_HOSTED_ONLY | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit 16 days ago) Napkins takes a sketch, wireframe or screenshot and produces a working web application, which is the input most people actually have when they want something built. Like its sibling LlamaCoder it runs on a TOGETHER_API_KEY you supply and self-hosts, so design mockups, which often contain unreleased product work, do not have to be uploaded to a service. MIT. Key handling: You self-host the app, so your API key lives in your own deployment. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 80/100 (assessable for any licence), verifiability 90/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 21/25: The key lives in a deployment you run yourself. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under MIT, so anyone can check how the key is handled. Key handling was located in the published source. - Privacy 10/15: Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 8/10: Commits within the last six months. - Verification Confidence 7/10: Key handling was found in the published source by BYOAIK's scanner. What was checked (verification tier SOURCE_VERIFIED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [STRONG via SOURCE_SCAN] The user supplies their own Together AI key: the project references its API key variable. (https://github.com/Nutlope/napkins/blob/HEAD/README.md) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-06-26, about 3 month(s) ago. (https://github.com/Nutlope/napkins/commits) ### minuet-ai.nvim - Page: https://www.byoaik.com/tools/minuet-ai-nvim/ - Website: https://github.com/milanglacier/minuet-ai.nvim - Source: https://github.com/milanglacier/minuet-ai.nvim - Category: IDE Plugins | Pricing: OPEN_SOURCE | Open source: yes (GPL-3.0) | Trust score: 83/100 - Platforms: CLI, Local-only - Bring-your-own-key providers: OpenAI, Anthropic, Google Gemini, OpenRouter, Ollama, Custom / OpenAI-compatible - Key storage: ENVIRONMENT_VARIABLE | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Maintained (last commit about 1 month ago) minuet-ai.nvim puts AI code completion into Neovim through blink.cmp, nvim-cmp or virtual text, and is unusually broad about providers: OpenAI, Anthropic, Gemini, OpenAI-compatible endpoints and Ollama are all first-class, configured with environment variables you control. The Ollama path means completion can run with no cloud key and nothing leaving the machine, which is a rarer property in completion plugins than in chat ones, since completion sends buffer context continuously rather than on request. GPL-3.0. Key handling: Your API key is supplied via an environment variable or local config file. Requests are sent directly to the AI provider. Scores: key handling 77/100 (assessable for any licence), verifiability 90/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 22/25: The key is supplied by an environment variable or local config file you control. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under GPL-3.0, so anyone can check how the key is handled. Key handling was located in the published source. - Privacy 7/15: Nothing specific is known about its data handling beyond the key itself. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 7/10: Key handling was found in the published source by BYOAIK's scanner. What was checked (verification tier SOURCE_VERIFIED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [STRONG via SOURCE_SCAN] The user supplies their own OpenAI key: the project references its API key variable. (https://github.com/milanglacier/minuet-ai.nvim/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] The user supplies their own Anthropic key: the project references its API key variable. (https://github.com/milanglacier/minuet-ai.nvim/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] The user supplies their own Google Gemini key: the project references its API key variable. (https://github.com/milanglacier/minuet-ai.nvim/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] The user supplies their own OpenRouter key: the project references its API key variable. (https://github.com/milanglacier/minuet-ai.nvim/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] The user supplies their own Groq key: the project references its API key variable. (https://github.com/milanglacier/minuet-ai.nvim/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] The user supplies their own Mistral key: the project references its API key variable. (https://github.com/milanglacier/minuet-ai.nvim/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] The user supplies their own DeepSeek key: the project references its API key variable. (https://github.com/milanglacier/minuet-ai.nvim/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] The user supplies their own Fireworks key: the project references its API key variable. (https://github.com/milanglacier/minuet-ai.nvim/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/milanglacier/minuet-ai.nvim/blob/HEAD/README.md) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-08-14, about 2 month(s) ago. (https://github.com/milanglacier/minuet-ai.nvim/commits) ### LearnHouse - Page: https://www.byoaik.com/tools/learnhouse/ - Website: https://learnhouse.app/ - Source: https://github.com/learnhouse/learnhouse - Category: Education Tools | Pricing: OPEN_SOURCE | Open source: yes (AGPL-3.0) | Trust score: 83/100 - Platforms: Web, Self-hosted, Docker - Bring-your-own-key providers: OpenAI, Ollama, Custom / OpenAI-compatible - Key storage: SELF_HOSTED_ONLY | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit today) LearnHouse is an open-source learning platform covering courses, assignments, discussions, code execution with auto-grading and collaborative whiteboards, with AI features through the material rather than bolted on. It self-hosts, so course content and student submissions stay on infrastructure the institution runs. AGPL-3.0, which matters if you intend to offer a modified version as a service. Key handling: You self-host the app, so your API key lives in your own deployment. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 87/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 21/25: The key lives in a deployment you run yourself. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under AGPL-3.0, so anyone can check how the key is handled. - Privacy 14/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [CONFIRMED via SOURCE_SCAN] Ships a container definition, so it can be self-hosted on your own infrastructure. (https://github.com/learnhouse/learnhouse/blob/HEAD/Dockerfile) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-24, about 0 month(s) ago. (https://github.com/learnhouse/learnhouse/commits) ### ChatterUI - Page: https://www.byoaik.com/tools/chatterui/ - Website: https://github.com/Vali-98/ChatterUI - Source: https://github.com/Vali-98/ChatterUI - Category: AI Chat Interfaces | Pricing: OPEN_SOURCE | Open source: yes (AGPL-3.0) | Trust score: 82/100 - Platforms: Mobile - Bring-your-own-key providers: OpenAI, Anthropic, OpenRouter, Cohere, Ollama, Custom / OpenAI-compatible - Key storage: LOCAL_DEVICE | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Maintained (last commit about 1 month ago) ChatterUI is a native Android frontend for LLMs, described by its maintainer as supporting bring-your-own-key, local native inference and local network modes. It runs GGUF models fully on-device via llama.cpp, or connects to remote services including OpenAI, Claude, Cohere and OpenRouter, plus self-hosted backends such as Ollama, KoboldCPP and text-generation-webui, using API keys the user enters in-app. Distributed as a free APK from GitHub under AGPL-3.0. Key handling: Your API key is stored locally on your device. Requests are sent directly to the AI provider. Scores: key handling 85/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 23/25: The key is held on your own device. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under AGPL-3.0, so anyone can check how the key is handled. - Privacy 11/15: Local-first: it works without sending your data anywhere. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/Vali-98/ChatterUI/blob/HEAD/README.md) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-22, about 0 month(s) ago. (https://github.com/Vali-98/ChatterUI/commits) ### Zed - Page: https://www.byoaik.com/tools/zed/ - Website: https://zed.dev - Source: https://github.com/zed-industries/zed - Category: Coding Assistants | Pricing: OPEN_SOURCE | Open source: yes (GPL-3.0 / Apache-2.0 (mixed; GitHub reports NOASSERTION)) | Trust score: 82/100 - Platforms: Desktop - Bring-your-own-key providers: Anthropic, OpenAI, Google Gemini, Mistral, DeepSeek, xAI Grok, OpenRouter, Ollama, Custom / OpenAI-compatible - Key storage: LOCAL_DEVICE | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit today) Zed is an open-source, high-performance code editor written in Rust by the creators of Atom and Tree-sitter. Its Agent Panel and inline assist support multiple LLM providers where the end user pastes their own API key; spend then goes directly to that provider. Confirmed BYOK providers in the docs include Anthropic, OpenAI, Google AI (Gemini), Mistral, DeepSeek, and xAI, plus gateways like OpenRouter and local models via Ollama/LM Studio. Zed also supports adding a custom OpenAI-compatible provider by specifying a custom base URL, model ID, and API key. Keys entered in Zed are stored in the OS system keychain rather than in plaintext settings, and provider-specific environment variables are also honored. Key handling: Your API key is stored locally on your device. Requests are sent directly to the AI provider. Scores: key handling 85/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 23/25: The key is held on your own device. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under GPL-3.0 / Apache-2.0 (mixed; GitHub reports NOASSERTION), so anyone can check how the key is handled. - Privacy 11/15: Local-first: it works without sending your data anywhere. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-29, about 0 month(s) ago. (https://github.com/zed-industries/zed/commits) ### OpenCode - Page: https://www.byoaik.com/tools/opencode/ - Website: https://opencode.ai - Source: https://github.com/anomalyco/opencode - Category: Coding Assistants | Pricing: OPEN_SOURCE | Open source: yes (MIT) | Trust score: 82/100 - Platforms: CLI, Desktop, Self-hosted, Docker - Bring-your-own-key providers: Anthropic, OpenAI, Google Gemini, Azure OpenAI, OpenRouter, Groq, DeepSeek, xAI Grok, Together AI, Fireworks, Hugging Face, Ollama, Custom / OpenAI-compatible - Key storage: LOCAL_DEVICE | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit today) OpenCode is an open-source (MIT) AI coding agent built primarily as a terminal UI (TUI), with a beta desktop app. It is fully BYOK and provider-agnostic: via the AI SDK and Models.dev it supports 75+ LLM providers, and users add their own API keys with the `/connect` command, stored locally in `~/.local/share/opencode/auth.json`. Any provider's `baseURL` can be customized for proxies, custom endpoints, or local models (Ollama, LM Studio, llama.cpp). It is one of the most popular open-source coding agents, with ~176.6k GitHub stars and active daily development by the SST team. Key handling: Your API key is stored locally on your device. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 85/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 23/25: The key is held on your own device. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under MIT, so anyone can check how the key is handled. - Privacy 11/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. Analytics or error-reporting libraries are present; what they send was not established. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [REPORTED via SOURCE_SCAN] References Sentry, so some analytics or error reporting is present. This scan cannot tell whether it is opt-in or what it sends. (https://github.com/anomalyco/opencode/blob/HEAD/package.json) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-29, about 0 month(s) ago. (https://github.com/anomalyco/opencode/commits) ### Open Deep Research (LangChain) - Page: https://www.byoaik.com/tools/open-deep-research-langchain/ - Website: https://github.com/langchain-ai/open_deep_research - Source: https://github.com/langchain-ai/open_deep_research - Category: Research Tools | Pricing: OPEN_SOURCE | Open source: yes (MIT) | Trust score: 82/100 - Platforms: Self-hosted, Library/SDK, Docker - Bring-your-own-key providers: OpenAI, Anthropic, Google Gemini, Azure OpenAI, OpenRouter, Groq, Mistral, Cohere, DeepSeek, Together AI, Ollama - Key storage: SELF_HOSTED_ONLY | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Maintained (last commit about 3 months ago) Open Deep Research is LangChain's fully open-source, configurable deep research agent built on LangGraph that automates multi-step web research and generates long-form, cited reports (ranked among the top systems on Deep Research Bench). You bring your own provider API keys via a .env file, and because it uses LangChain's init_chat_model() with provider:model strings, it works across a wide range of LLM providers (OpenAI and Anthropic are tested defaults; Google Gemini, Ollama local models, OpenRouter, plus other LangChain-supported providers like Azure OpenAI, Cohere, Mistral, Groq, Together, Bedrock, and DeepSeek). Search is pluggable: Tavily by default, plus native web search for Anthropic and OpenAI, and full Model Context Protocol (MCP) server compatibility. It runs self-hosted via a local LangGraph server / LangGraph Studio and can be deployed on LangGraph Platform or used as a Python library. MIT licensed. Key handling: You self-host the app, so your API key lives in your own deployment. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 92/100 (assessable for any licence), verifiability 90/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 21/25: The key lives in a deployment you run yourself. - Request Routing 20/20: Requests go straight from you to the AI provider. The provider endpoint is configurable, so you can point it at the provider or your own gateway. - Transparency 20/20: Source is public under MIT, so anyone can check how the key is handled. Key handling was located in the published source. - Privacy 14/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 0/10: The repository is archived and no longer maintained. - Verification Confidence 7/10: Key handling was found in the published source by BYOAIK's scanner. What was checked (verification tier SOURCE_VERIFIED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/langchain-ai/open_deep_research/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] The provider endpoint is configurable, so requests can be pointed straight at the provider or at your own gateway. (https://github.com/langchain-ai/open_deep_research/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] The user supplies their own OpenAI key: the project references its API key variable. (https://github.com/langchain-ai/open_deep_research/blob/HEAD/.env.example) - [STRONG via SOURCE_SCAN] The user supplies their own Anthropic key: the project references its API key variable. (https://github.com/langchain-ai/open_deep_research/blob/HEAD/.env.example) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-08-10, about 2 month(s) ago. (https://github.com/langchain-ai/open_deep_research/commits) ### Cline - Page: https://www.byoaik.com/tools/cline/ - Website: https://cline.bot/ - Source: https://github.com/cline/cline - Category: Coding Assistants | Pricing: OPEN_SOURCE | Open source: yes (Apache-2.0) | Trust score: 82/100 - Platforms: VS Code Extension, JetBrains Plugin, CLI, Desktop - Bring-your-own-key providers: Anthropic, OpenAI, Google Gemini, Azure OpenAI, OpenRouter, Groq, Mistral, DeepSeek, xAI Grok, Together AI, Fireworks, Perplexity, Ollama, Custom / OpenAI-compatible, Hugging Face - Key storage: LOCAL_DEVICE | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit today) Cline is an open-source (Apache 2.0) autonomous coding agent that runs as a VS Code/JetBrains extension, CLI, and SDK. It edits files across a project, executes terminal commands with human-in-the-loop approval (Plan/Act modes), and integrates MCP servers. It is bring-your-own-key: users plug in their own provider credentials and pay only for inference, with no subscription. The settings panel exposes API key and custom OpenAI-compatible base URL fields, and the pricing FAQ explicitly confirms users can bring their own keys from OpenAI, Anthropic, Google, and others. Local models via Ollama/LM Studio are supported. The project is very actively maintained (63.6k stars, 250+ contributors, releases into June 2026) and reports 5M+ installs. Key handling: Your API key is stored locally on your device. Requests are sent directly to the AI provider. Scores: key handling 85/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 23/25: The key is held on your own device. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under Apache-2.0, so anyone can check how the key is handled. - Privacy 11/15: Local-first: it works without sending your data anywhere. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/cline/cline/blob/HEAD/README.md) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-29, about 0 month(s) ago. (https://github.com/cline/cline/commits) ### AnythingLLM - Page: https://www.byoaik.com/tools/anythingllm/ - Website: https://anythingllm.com/ - Source: https://github.com/Mintplex-Labs/anything-llm - Category: Knowledge Management | Pricing: FREEMIUM | Open source: yes (MIT) | Trust score: 82/100 - Platforms: Desktop, Self-hosted, Docker - Bring-your-own-key providers: OpenAI, Anthropic, Google Gemini, Azure OpenAI, OpenRouter, Groq, Mistral, Cohere, Perplexity, DeepSeek, xAI Grok, Ollama, Together AI, Fireworks, Custom / OpenAI-compatible - Key storage: LOCAL_DEVICE | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit 3 days ago) AnythingLLM is an open-source (MIT) all-in-one AI application that lets users chat with their documents, build no-code AI agents, and run a private RAG workspace either as a free desktop app (Mac/Windows/Linux) or self-hosted via Docker. It is bring-your-own-key: users obtain their own API keys from each provider and enter them in AnythingLLM's settings, or point it at local engines (Ollama, LM Studio, llama.cpp) or any custom OpenAI-compatible base URL. Confirmed providers where the user supplies their own key include OpenAI, Anthropic, Google Gemini, Azure OpenAI, AWS Bedrock, Groq, Mistral, DeepSeek, xAI Grok, OpenRouter, Perplexity, Cohere, Together AI, Fireworks AI, and generic OpenAI-compatible endpoints. The project is highly popular (~62k GitHub stars) and actively maintained (v1.14.1 released June 2026). A paid hosted Cloud tier and enterprise plans also exist, but the desktop and self-hosted BYOK paths are free. Key handling: Your API key is stored locally on your device. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 85/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 23/25: The key is held on your own device. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under MIT, so anyone can check how the key is handled. - Privacy 11/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. Analytics or error-reporting libraries are present; what they send was not established. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/Mintplex-Labs/anything-llm/blob/HEAD/README.md) - [REPORTED via SOURCE_SCAN] References PostHog, so some analytics or error reporting is present. This scan cannot tell whether it is opt-in or what it sends. (https://github.com/Mintplex-Labs/anything-llm/blob/HEAD/README.md) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-29, about 0 month(s) ago. (https://github.com/Mintplex-Labs/anything-llm/commits) ### Crush - Page: https://www.byoaik.com/tools/crush/ - Website: https://github.com/charmbracelet/crush - Source: https://github.com/charmbracelet/crush - Category: Coding Assistants | Pricing: OPEN_SOURCE | Open source: yes (FSL-1.1-MIT) | Trust score: 82/100 - Platforms: CLI - Bring-your-own-key providers: OpenAI, Anthropic, Google Gemini, Azure OpenAI, OpenRouter, Groq, Hugging Face, Ollama, Custom / OpenAI-compatible - Key storage: LOCAL_DEVICE | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit today) Crush is a terminal-based agentic AI coding assistant from Charm that bridges your terminal, your codebase, and the LLM of your choice. It uses LSP for additional code context, supports MCP servers over stdio/HTTP/SSE, manages per-project sessions, and lets you switch models mid-session while preserving conversation context. Users bring their own LLM provider API key, choosing from many named providers or connecting any OpenAI- or Anthropic-compatible endpoint (including local models) via configurable base URLs. Key handling: Your API key is stored locally on your device. Requests are sent directly to the AI provider. Scores: key handling 83/100 (assessable for any licence), verifiability 73/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 23/25: The key is held on your own device. - Request Routing 20/20: Requests go straight from you to the AI provider. The provider endpoint is configurable, so you can point it at the provider or your own gateway. - Transparency 15/20: Source is public, though the licence is unclear or non-standard. Key handling was located in the published source. - Privacy 7/15: Nothing specific is known about its data handling beyond the key itself. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 7/10: Key handling was found in the published source by BYOAIK's scanner. What was checked (verification tier SOURCE_VERIFIED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [STRONG via SOURCE_SCAN] The user supplies their own OpenAI key: the project references its API key variable. (https://github.com/charmbracelet/crush/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] The user supplies their own Anthropic key: the project references its API key variable. (https://github.com/charmbracelet/crush/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] The user supplies their own Google Gemini key: the project references its API key variable. (https://github.com/charmbracelet/crush/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] The user supplies their own OpenRouter key: the project references its API key variable. (https://github.com/charmbracelet/crush/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] The user supplies their own Groq key: the project references its API key variable. (https://github.com/charmbracelet/crush/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] The user supplies their own DeepSeek key: the project references its API key variable. (https://github.com/charmbracelet/crush/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] The user supplies their own Azure OpenAI key: the project references its API key variable. (https://github.com/charmbracelet/crush/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] The user supplies their own Hugging Face key: the project references its API key variable. (https://github.com/charmbracelet/crush/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/charmbracelet/crush/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] The provider endpoint is configurable, so requests can be pointed straight at the provider or at your own gateway. (https://github.com/charmbracelet/crush/blob/HEAD/README.md) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-29, about 0 month(s) ago. (https://github.com/charmbracelet/crush/commits) ### Subtitle Edit - Page: https://www.byoaik.com/tools/subtitle-edit/ - Website: https://www.nikse.dk/subtitleedit - Source: https://github.com/SubtitleEdit/subtitleedit - Category: Voice and Transcription Tools | Pricing: OPEN_SOURCE | Open source: yes (MIT) | Trust score: 82/100 - Platforms: Desktop - Bring-your-own-key providers: OpenAI, Anthropic, Google Gemini, Groq, OpenRouter, Perplexity, Mistral, DeepSeek, Ollama, Custom / OpenAI-compatible - Key storage: LOCAL_DEVICE | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit today) Subtitle Edit is a long-running, MIT-licensed open-source subtitle/caption editor (originally Windows-native, now cross-platform via v4) for creating, editing, syncing, and translating subtitles for video. Its Auto Translate feature includes a broad set of AI translation engines that the end user drives with their own API key: ChatGPT/OpenAI, Anthropic Claude, Google Gemini, Groq, OpenRouter, Perplexity, Mistral AI, DeepSeek, and NVIDIA, plus OpenAI-compatible local LLM backends (Ollama, LM Studio, llama.cpp) requiring no external key. The official docs explicitly mark these engines as "requires API key," and GitHub issue #7850 shows users supplying their own OpenAI secret key and endpoint. Separately, offline speech-to-text is handled by bundled Whisper engines (OpenAI Whisper, Faster Whisper, whisper.cpp, Const-me, etc.) running entirely on-device: not a BYOK cloud feature. The app is privacy-focused and offline-first; only the optional online translation/STT engines send data to the selected provider. Key handling: Your API key is stored locally on your device. Requests are sent directly to the AI provider. Scores: key handling 85/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 23/25: The key is held on your own device. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under MIT, so anyone can check how the key is handled. - Privacy 11/15: Local-first: it works without sending your data anywhere. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-29, about 0 month(s) ago. (https://github.com/SubtitleEdit/subtitleedit/commits) ### Scira - Page: https://www.byoaik.com/tools/scira/ - Website: https://github.com/zaidmukaddam/scira - Source: https://github.com/zaidmukaddam/scira - Category: Research Tools | Pricing: OPEN_SOURCE | Open source: yes (AGPL-3.0) | Trust score: 82/100 - Platforms: Web, Self-hosted, Docker - Bring-your-own-key providers: xAI Grok, OpenAI, Anthropic, Google Gemini, Mistral, Groq, Cohere, DeepSeek, OpenRouter, Custom / OpenAI-compatible - Key storage: SELF_HOSTED_ONLY | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Maintained (last commit about 3 months ago) Scira (formerly MiniPerplx) is an open-source, agentic AI-powered search engine built on the Vercel AI SDK. It decomposes a question into sub-tasks, runs live web/academic/social searches (Exa, Firecrawl, Tavily, Parallel), cross-checks evidence, and returns answers with inline citations. The hosted product runs at scira.ai (1M+ monthly searches, 60k+ users per third-party reporting), but the full app is self-hostable under AGPL-3.0 via Docker Compose, Vercel, or Node/pnpm. When self-hosting, the operator brings their own provider API keys: the .env.example defines XAI_API_KEY, OPENAI_API_KEY, ANTHROPIC_API_KEY, GROQ_API_KEY, GOOGLE_GENERATIVE_AI_API_KEY (plus Inception) and search keys EXA_API_KEY, FIRECRAWL_API_KEY, TAVILY_API_KEY. The ai/providers.ts wires direct SDKs for OpenAI, Anthropic, Google Gemini, Mistral, Groq, xAI Grok, and Cohere, an OpenRouter custom provider, OpenAI-compatible providers (Z.AI/GLM, HuggingFace, Novita, MiniMax, etc.), and Vercel AI Gateway routing for DeepSeek, Qwen/Alibaba, Kimi and others. Key handling: You self-host the app, so your API key lives in your own deployment. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 75/100 (assessable for any licence), verifiability 90/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 21/25: The key lives in a deployment you run yourself. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under AGPL-3.0, so anyone can check how the key is handled. Key handling was located in the published source. - Privacy 7/15: Can be self-hosted, so the data path stays inside infrastructure you control. Analytics or error-reporting libraries are present; what they send was not established. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 7/10: Key handling was found in the published source by BYOAIK's scanner. What was checked (verification tier SOURCE_VERIFIED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [STRONG via SOURCE_SCAN] The user supplies their own OpenAI key: the project references its API key variable. (https://github.com/zaidmukaddam/scira/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] The user supplies their own Anthropic key: the project references its API key variable. (https://github.com/zaidmukaddam/scira/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] The user supplies their own Groq key: the project references its API key variable. (https://github.com/zaidmukaddam/scira/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] The user supplies their own xAI Grok key: the project references its API key variable. (https://github.com/zaidmukaddam/scira/blob/HEAD/README.md) - [REPORTED via SOURCE_SCAN] References PostHog, so some analytics or error reporting is present. This scan cannot tell whether it is opt-in or what it sends. (https://github.com/zaidmukaddam/scira/blob/HEAD/README.md) - [CONFIRMED via SOURCE_SCAN] Ships a container definition, so it can be self-hosted on your own infrastructure. (https://github.com/zaidmukaddam/scira/blob/HEAD/docker-compose.yml) - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/zaidmukaddam/scira/blob/HEAD/package.json) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-08-12, about 2 month(s) ago. (https://github.com/zaidmukaddam/scira/commits) ### Anarlog - Page: https://www.byoaik.com/tools/anarlog/ - Website: https://anarlog.so/ - Source: https://github.com/fastrepl/anarlog - Category: Voice and Transcription Tools | Pricing: OPEN_SOURCE | Open source: yes (MIT) | Trust score: 82/100 - Platforms: Desktop - Bring-your-own-key providers: OpenAI, Anthropic, Google Gemini, OpenRouter, Ollama, Custom / OpenAI-compatible - Key storage: LOCAL_DEVICE | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit 3 days ago) Anarlog (formerly Hyprnote/Char) is an open-source, local-first AI notepad for meetings built on Rust/Tauri. Transcription runs on-device so audio never leaves the machine, and notes/transcripts are stored as plain markdown files on disk with no cloud backend. For LLM-powered summarization and note-merging, the end user supplies their own provider: the README states "Bring your own LLM: OpenAI, Anthropic, Gemini, OpenRouter, Ollama, LM Studio, or anything OpenAI-compatible." It works with any meeting platform (Zoom, Teams, Meet, in-person). Currently macOS 14.2+ only (Apple Silicon and Intel); Windows and Linux are planned for Q2 2026. Free for local transcription and BYOK. Note: the candidate's listed URL (anarlog.app) is dead and the listed repo (anarlog/anarlog) 404s; the correct sources are anarlog.so and github.com/fastrepl/anarlog. License is MIT (confirmed via GitHub API and README); one marketing testimonial incorrectly referenced "GPL". Key handling: Your API key is stored locally on your device. Requests are sent directly to the AI provider. Scores: key handling 85/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 23/25: The key is held on your own device. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under MIT, so anyone can check how the key is handled. - Privacy 11/15: Local-first: it works without sending your data anywhere. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/fastrepl/anarlog/blob/HEAD/README.md) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-29, about 0 month(s) ago. (https://github.com/fastrepl/anarlog/commits) ### PowerToys Advanced Paste - Page: https://www.byoaik.com/tools/powertoys-advanced-paste/ - Website: https://learn.microsoft.com/en-us/windows/powertoys/advanced-paste - Source: https://github.com/microsoft/PowerToys - Category: Productivity Tools | Pricing: OPEN_SOURCE | Open source: yes (MIT) | Trust score: 82/100 - Platforms: Desktop - Bring-your-own-key providers: OpenAI, Azure OpenAI, Google Gemini, Mistral, Ollama - Key storage: LOCAL_DEVICE | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit 3 days ago) Advanced Paste is a module in Microsoft's open-source PowerToys suite for Windows that reformats clipboard content into Markdown, JSON or a custom prompt result using an AI provider you configure yourself. In settings you add a model provider and, per Microsoft's documentation, usually supply the API key, endpoint URL and any extra options for OpenAI, Azure OpenAI, Mistral or Google, or point at a local Foundry Local or Ollama model requiring no key at all. PowerToys is MIT licensed and free; you pay only your chosen provider. 137,669 stars on the monorepo, continuous development. Key handling: Your API key is stored locally on your device. Requests are sent directly to the AI provider. Scores: key handling 85/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 23/25: The key is held on your own device. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under MIT, so anyone can check how the key is handled. - Privacy 11/15: Local-first: it works without sending your data anywhere. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-29, about 0 month(s) ago. (https://github.com/microsoft/PowerToys/commits) ### Tabby - Page: https://www.byoaik.com/tools/tabby/ - Website: https://www.tabbyml.com - Source: https://github.com/TabbyML/tabby - Category: Coding Assistants | Pricing: OPEN_SOURCE | Open source: yes (Apache-2.0 (core); separate enterprise license for the ee/ directory) | Trust score: 82/100 - Platforms: Self-hosted, VS Code Extension, JetBrains Plugin, CLI, Docker - Bring-your-own-key providers: OpenAI, Anthropic, Azure OpenAI, OpenRouter, Mistral, Perplexity, DeepSeek, Fireworks, Hugging Face, Ollama, Custom / OpenAI-compatible - Key storage: ENVIRONMENT_VARIABLE | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Slowing down (last commit about 4 months ago) Tabby is a self-hosted AI coding assistant that runs as a server (binary, Docker, or Homebrew) and provides code completion and chat as an open-source, on-premises alternative to GitHub Copilot, with IDE extensions for VS Code, JetBrains/IntelliJ, and Vim/Neovim. It is BYOK: via ~/.tabby/config.toml's HTTP Model Connector you configure completion, chat, and embedding models against external providers using your own api_key and api_endpoint. The docs include dedicated configuration pages for OpenAI, Azure OpenAI, Anthropic Claude, Mistral AI, OpenRouter, DeepSeek, Perplexity, Fireworks, Hugging Face, Amazon Bedrock, plus local engines (Ollama, vLLM, llama.cpp, LM Studio) and arbitrary OpenAI-compatible endpoints. Core is Apache 2.0; an "ee/" enterprise directory carries a separate license. ~33.6k GitHub stars, actively maintained (v0.32.0, Jan 2026). Key handling: Your API key is supplied via an environment variable or local config file. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 88/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 22/25: The key is supplied by an environment variable or local config file you control. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under Apache-2.0 (core); separate enterprise license for the ee/ directory, so anyone can check how the key is handled. - Privacy 14/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 8/10: Commits within the last six months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-06-30, about 3 month(s) ago. (https://github.com/TabbyML/tabby/commits) ### Copilot for Obsidian - Page: https://www.byoaik.com/tools/copilot-for-obsidian/ - Website: https://www.obsidiancopilot.com/ - Source: https://github.com/logancyang/obsidian-copilot - Category: Writing Tools | Pricing: OPEN_SOURCE | Open source: yes (AGPL-3.0) | Trust score: 82/100 - Platforms: Desktop, Mobile, Self-hosted, Docker - Bring-your-own-key providers: OpenAI, Anthropic, Google Gemini, Cohere, Groq, OpenRouter, Perplexity, Ollama, Custom / OpenAI-compatible - Key storage: LOCAL_DEVICE | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit today) Copilot for Obsidian (logancyang/obsidian-copilot) is an open-source community plugin that embeds an AI assistant inside the Obsidian note-taking app. It provides AI chat, vault-wide semantic search and Q&A over your notes, web/YouTube context, and agentic content generation, with a stated philosophy of local data control and no provider lock-in. The free plugin is bring-your-own-key: users paste their own API keys for OpenAI, Anthropic, Google, Cohere, Groq, or OpenRouter, point at any OpenAI-compatible endpoint via a custom Base URL (e.g. Perplexity), or run fully local via Ollama or LM Studio. An optional one-time "Copilot Plus" license adds a built-in (managed) chat model and embeddings for users who do not want to supply their own keys. Licensed AGPL-3.0, ~7.3k GitHub stars, written in TypeScript, actively maintained (latest release v3.3.3, May 2026). Key handling: Your API key is stored locally on your device. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 85/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 23/25: The key is held on your own device. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under AGPL-3.0, so anyone can check how the key is handled. - Privacy 11/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. Analytics or error-reporting libraries are present; what they send was not established. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [REPORTED via SOURCE_SCAN] References PostHog, so some analytics or error reporting is present. This scan cannot tell whether it is opt-in or what it sends. (https://github.com/logancyang/obsidian-copilot/blob/HEAD/docs/README.md) - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/logancyang/obsidian-copilot/blob/HEAD/package.json) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-29, about 0 month(s) ago. (https://github.com/logancyang/obsidian-copilot/commits) ### Enchanted - Page: https://www.byoaik.com/tools/enchanted/ - Website: https://github.com/gluonfield/enchanted - Source: https://github.com/gluonfield/enchanted - Category: AI Chat Interfaces | Pricing: OPEN_SOURCE | Open source: yes (Apache-2.0) | Trust score: 82/100 - Platforms: Mobile, Desktop - Bring-your-own-key providers: Ollama - Key storage: LOCAL_DEVICE | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Slowing down (last commit about 4 months ago) Enchanted is a native Apple-ecosystem chat client that talks to a private Ollama server you run yourself, for models such as Llama and Mistral. Setup requires the user to run their own Ollama server and specify that server endpoint in app settings, so no vendor account or pooled key is involved and no conversation data leaves infrastructure you control. Apache-2.0 licensed, free on the App Store, with around 6,000 stars and active maintenance. Key handling: Your API key is stored locally on your device. Requests are sent directly to the AI provider. Scores: key handling 85/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 23/25: The key is held on your own device. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under Apache-2.0, so anyone can check how the key is handled. - Privacy 11/15: Local-first: it works without sending your data anywhere. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/gluonfield/enchanted/blob/HEAD/README.md) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-07-07, about 3 month(s) ago. (https://github.com/gluonfield/enchanted/commits) ### Text Generator (Obsidian) - Page: https://www.byoaik.com/tools/text-generator-obsidian/ - Website: https://github.com/nhaouari/obsidian-textgenerator-plugin - Source: https://github.com/nhaouari/obsidian-textgenerator-plugin - Category: Writing Tools | Pricing: OPEN_SOURCE | Open source: yes (MIT) | Trust score: 82/100 - Platforms: Desktop, Mobile - Bring-your-own-key providers: OpenAI, Anthropic, Google Gemini, Azure OpenAI, Mistral, Perplexity, Hugging Face, Together AI, Ollama, Custom / OpenAI-compatible - Key storage: LOCAL_DEVICE | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Slowing down (last commit about 3 months ago) Text Generator is an open-source Obsidian community plugin that generates text content (ideas, titles, summaries, outlines, paragraphs) directly in your notes using AI. It centers on a customizable template engine with community-shared prompt templates and per-note frontmatter configuration, letting users wire prompts to many AI providers. Users bring their own API keys, configured in plugin settings or frontmatter; there is no built-in hosted model. Provider support is broad, spanning OpenAI, Anthropic, Google Gemini/PaLM, Azure OpenAI, Hugging Face, Mistral, Perplexity, Together, local models via Ollama, and arbitrary custom OpenAI-compatible endpoints. MIT-licensed and actively maintained (last pushed May 2026), it is one of the most popular AI plugins in the Obsidian ecosystem. Key handling: Your API key is stored locally on your device. Requests are sent directly to the AI provider. Scores: key handling 85/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 23/25: The key is held on your own device. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under MIT, so anyone can check how the key is handled. - Privacy 11/15: Local-first: it works without sending your data anywhere. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/nhaouari/obsidian-textgenerator-plugin/blob/HEAD/package.json) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-08-06, about 2 month(s) ago. (https://github.com/nhaouari/obsidian-textgenerator-plugin/commits) ### Whispering - Page: https://www.byoaik.com/tools/whispering/ - Website: https://epicenter.so/whispering/ - Source: https://github.com/EpicenterHQ/epicenter - Category: Voice and Transcription Tools | Pricing: OPEN_SOURCE | Open source: yes (AGPL-3.0 (app); monorepo dual-licensed (MIT toolkit + AGPL-3.0)) | Trust score: 82/100 - Platforms: Desktop, Web, Self-hosted, Docker - Bring-your-own-key providers: OpenAI, Anthropic, Google Gemini, Groq, Ollama, Custom / OpenAI-compatible - Key storage: LOCAL_DEVICE | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit today) Whispering is an open-source, local-first dictation and transcription desktop/web app from Epicenter (YC-backed). You press a shortcut, speak, and it transcribes, optionally runs an AI transformation, then pastes the result at your cursor. It is built on a bring-your-own-key model: you supply your own provider API keys and pay providers directly, with audio going straight from your device to the chosen provider's API (no Epicenter servers in the transcription path). Transcription can run fully locally (Speaches, Whisper C++, Parakeet, Moonshine) or via cloud providers (OpenAI, Groq, ElevenLabs). AI transformations support OpenAI, Anthropic, Google Gemini, Groq, and any custom OpenAI-compatible endpoint via a configurable base URL, so you can point it at self-hosted or private LLMs. Built with Svelte 5 and Tauri, the desktop app is small (~22MB) and a web version is also available. Key handling: Your API key is stored locally on your device. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 85/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 23/25: The key is held on your own device. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under AGPL-3.0 (app); monorepo dual-licensed (MIT toolkit + AGPL-3.0), so anyone can check how the key is handled. - Privacy 11/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. Analytics or error-reporting libraries are present; what they send was not established. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [REPORTED via SOURCE_SCAN] References PostHog, so some analytics or error reporting is present. This scan cannot tell whether it is opt-in or what it sends. (https://github.com/EpicenterHQ/epicenter/blob/HEAD/README.md) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-28, about 0 month(s) ago. (https://github.com/EpicenterHQ/epicenter/commits) ### VoiceInk - Page: https://www.byoaik.com/tools/voiceink/ - Website: https://tryvoiceink.com/ - Source: https://github.com/Beingpax/VoiceInk - Category: Voice and Transcription Tools | Pricing: OPEN_SOURCE | Open source: yes (GPL-3.0) | Trust score: 82/100 - Platforms: Desktop - Bring-your-own-key providers: OpenAI, Anthropic, Google Gemini, Groq, Mistral, OpenRouter, Ollama, Custom / OpenAI-compatible - Key storage: LOCAL_DEVICE | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit 2 days ago) VoiceInk is a native macOS (Apple Silicon, macOS 14.4+) voice-to-text dictation app positioned as a Superwhisper/Wispr Flow alternative. By default it transcribes entirely on-device using whisper.cpp models with no data leaving the machine. Optionally, users can enable cloud speech-to-text and AI text enhancement (cleanup, formatting, custom enhancement modes) using their own API keys. The source confirms a broad provider list in AIService.swift: OpenAI, Anthropic, Google Gemini, Groq, Mistral, OpenRouter, Cerebras, Deepgram, ElevenLabs, Speechmatics, Soniox, AssemblyAI, Ollama (local), plus a 'custom' OpenAI-compatible provider where the user supplies both the base URL and API key. API keys are stored in the macOS Keychain and never sent to VoiceInk servers. Licensed GPLv3; very actively maintained (v1.79 released May 2026, 124 releases, ~1.5k commits, ~5.3k GitHub stars). The app is sold as a one-time license ($25/$39/$49 for 1/2/3 devices) while the source remains available on GitHub. Key handling: Your API key is stored locally on your device. Requests are sent directly to the AI provider. Scores: key handling 85/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 23/25: The key is held on your own device. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under GPL-3.0, so anyone can check how the key is handled. - Privacy 11/15: Local-first: it works without sending your data anywhere. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/Beingpax/VoiceInk/blob/HEAD/README.md) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-29, about 0 month(s) ago. (https://github.com/Beingpax/VoiceInk/commits) ### Text Generator - Page: https://www.byoaik.com/tools/text-generator/ - Website: https://text-gen.com/ - Source: https://github.com/nhaouari/obsidian-textgenerator-plugin - Category: Writing Tools | Pricing: OPEN_SOURCE | Open source: yes (MIT) | Trust score: 82/100 - Platforms: Desktop, Mobile - Bring-your-own-key providers: OpenAI, Anthropic, Google Gemini, Azure OpenAI, Hugging Face, Ollama, Custom / OpenAI-compatible - Key storage: LOCAL_DEVICE | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Slowing down (last commit about 3 months ago) Text Generator is an open-source (MIT) Obsidian community plugin that generates text content from reusable prompt templates directly inside your notes. It supports inline generation, batch processing, custom commands, and frontmatter-based per-note configuration. Users bring their own API keys: in plugin settings you pick a provider (OpenAI, Anthropic, Google/Gemini, Azure OpenAI, HuggingFace, or a custom OpenAI-compatible endpoint such as Ollama/local models) and enter your own key, paying the provider directly. The plugin is widely used (~546K downloads, ~1,950 GitHub stars) and actively maintained, with the latest release (v0.8.7) shipped in April 2026. Key handling: Your API key is stored locally on your device. Requests are sent directly to the AI provider. Scores: key handling 85/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 23/25: The key is held on your own device. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under MIT, so anyone can check how the key is handled. - Privacy 11/15: Local-first: it works without sending your data anywhere. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/nhaouari/obsidian-textgenerator-plugin/blob/HEAD/package.json) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-08-06, about 2 month(s) ago. (https://github.com/nhaouari/obsidian-textgenerator-plugin/commits) ### Anki AI Utils - Page: https://www.byoaik.com/tools/anki-ai-utils/ - Website: https://github.com/thiswillbeyourgithub/AnkiAIUtils - Source: https://github.com/thiswillbeyourgithub/AnkiAIUtils - Category: Education Tools | Pricing: OPEN_SOURCE | Open source: yes (AGPL-3.0) | Trust score: 82/100 - Platforms: CLI, Self-hosted, Docker - Bring-your-own-key providers: OpenAI, Anthropic, Google Gemini, OpenRouter, Ollama, Custom / OpenAI-compatible - Key storage: ENVIRONMENT_VARIABLE | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Slowing down (last commit about 5 months ago) Anki AI Utils is a set of Python scripts that enhance the Anki cards you answer wrong, generating explanations, personalised mnemonics and illustrations while preserving your own mnemonic style. It is provider agnostic, routing calls through LiteLLM, and you place each provider's key in a local keys directory or the matching environment variable, so nothing is proxied through the author's infrastructure. Runs as local scripts alongside Anki. AGPL-3.0, 876 stars, last commit about two months before review. Key handling: Your API key is supplied via an environment variable or local config file. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 88/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 22/25: The key is supplied by an environment variable or local config file you control. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under AGPL-3.0, so anyone can check how the key is handled. - Privacy 14/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 8/10: Commits within the last six months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-06-01, about 4 month(s) ago. (https://github.com/thiswillbeyourgithub/AnkiAIUtils/commits) ### Ellama - Page: https://www.byoaik.com/tools/ellama/ - Website: https://github.com/s-kostyaev/ellama - Source: https://github.com/s-kostyaev/ellama - Category: IDE Plugins | Pricing: OPEN_SOURCE | Open source: yes (GPL-3.0) | Trust score: 82/100 - Platforms: Desktop, Local-only - Bring-your-own-key providers: OpenAI, Anthropic, Google Gemini, Ollama, Custom / OpenAI-compatible - Key storage: LOCAL_DEVICE | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Maintained (last commit about 1 month ago) Ellama brings chat, code review, refactoring, summarising and commit-message generation into Emacs through the GNU ELPA llm package, so it speaks to Ollama, OpenAI-compatible endpoints, OpenAI, Claude, Gemini, Vertex, GPT4All and llama.cpp among others. By default it selects the first available Ollama model, meaning the out-of-the-box path needs no cloud key and no request leaves the machine; hosted providers are opt-in with a credential you supply. GPL-3.0. Key handling: Your API key is stored locally on your device. Requests are sent directly to the AI provider. Scores: key handling 85/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 23/25: The key is held on your own device. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under GPL-3.0, so anyone can check how the key is handled. - Privacy 11/15: Local-first: it works without sending your data anywhere. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/s-kostyaev/ellama/blob/HEAD/README.org) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-29, about 0 month(s) ago. (https://github.com/s-kostyaev/ellama/commits) ### Guidance - Page: https://www.byoaik.com/tools/guidance/ - Website: https://github.com/guidance-ai/guidance - Source: https://github.com/guidance-ai/guidance - Category: Prompt Engineering Tools | Pricing: OPEN_SOURCE | Open source: yes (MIT) | Trust score: 82/100 - Platforms: Library/SDK, Self-hosted, Docker - Bring-your-own-key providers: OpenAI, Azure OpenAI, Hugging Face, Ollama - Key storage: ENVIRONMENT_VARIABLE | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Slowing down (last commit about 6 months ago) Guidance is a library for constrained, structured prompt programming, interleaving generation, logic and control flow to steer output reliably. You authenticate by passing api_key to a model constructor, setting the OPENAI_API_KEY environment variable, or using a local key file; local backends through Transformers or llama.cpp need no key at all. Python library. MIT licensed, 21,714 stars, last commit about two months before review. Key handling: Your API key is supplied via an environment variable or local config file. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 88/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 22/25: The key is supplied by an environment variable or local config file you control. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under MIT, so anyone can check how the key is handled. - Privacy 14/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 8/10: Commits within the last six months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/guidance-ai/guidance/blob/HEAD/README.md) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-05-21, about 4 month(s) ago. (https://github.com/guidance-ai/guidance/commits) ### PyGPT - Page: https://www.byoaik.com/tools/pygpt/ - Website: https://pygpt.net/ - Source: https://github.com/szczyglis-dev/py-gpt - Category: Desktop AI Assistants | Pricing: OPEN_SOURCE | Open source: yes (MIT) | Trust score: 82/100 - Platforms: Desktop, Local-only - Bring-your-own-key providers: OpenAI, Anthropic, Google Gemini, Mistral, Perplexity, xAI Grok, OpenRouter, Ollama, Custom / OpenAI-compatible - Key storage: LOCAL_DEVICE | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Maintained (last commit about 1 month ago) PyGPT is a desktop AI assistant covering chat, agents, web search, Python execution, speech in and out, plugins and long-term memory. The README is unusually direct about the key question: it 'utilizes the user's own API key', connecting to OpenAI, Google, Anthropic, xAI, Perplexity, Mistral, OpenRouter and others with credentials you hold, and it notes that local models served through Ollama need no external key at all. MIT, confirmed by reading the LICENSE rather than trusting GitHub's detector, which reports the repository as unlicensed. Key handling: Your API key is stored locally on your device. Requests are sent directly to the AI provider. Scores: key handling 85/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 23/25: The key is held on your own device. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under MIT, so anyone can check how the key is handled. - Privacy 11/15: Local-first: it works without sending your data anywhere. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/szczyglis-dev/py-gpt/blob/HEAD/README.md) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-29, about 0 month(s) ago. (https://github.com/szczyglis-dev/py-gpt/commits) ### llamafile - Page: https://www.byoaik.com/tools/llamafile/ - Website: https://github.com/mozilla-ai/llamafile - Source: https://github.com/mozilla-ai/llamafile - Category: Local AI Tools | Pricing: OPEN_SOURCE | Open source: yes (Mixed: Apache-2.0 with MIT-derived components, no single SPDX tag) | Trust score: 82/100 - Platforms: CLI, Self-hosted, Docker - Bring-your-own-key providers: Ollama - Key storage: NO_KEY_LOCAL_MODEL | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit 3 days ago) llamafile, a Mozilla.ai project, combines llama.cpp with Cosmopolitan Libc to turn a model into a single cross-platform executable: download the file, make it executable and run it, and it opens a local chat interface plus an OpenAI-compatible API with no installation, account or key. The same file runs on Windows, macOS, Linux and several CPU architectures, and a sibling tool covers local speech-to-text. 25,590 stars, pushed within a month of review. The repository reports no single SPDX licence: it is Apache-2.0 with MIT-derived llama.cpp and whisper.cpp portions. Key handling: Key handling varies. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 93/100 (assessable for any licence), verifiability 53/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 25/25: No cloud key is needed at all. The model runs locally. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 15/20: Source is public, though the licence is unclear or non-standard. - Privacy 14/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/mozilla-ai/llamafile/blob/HEAD/README.md) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-16, about 0 month(s) ago. (https://github.com/mozilla-ai/llamafile/commits) ### Helicone - Page: https://www.byoaik.com/tools/helicone/ - Website: https://www.helicone.ai/ - Source: https://github.com/Helicone/helicone - Category: Developer Infrastructure | Pricing: OPEN_SOURCE | Open source: yes (Apache-2.0) | Trust score: 82/100 - Platforms: Self-hosted, Docker, Web - Bring-your-own-key providers: OpenAI, Anthropic, Google Gemini, Azure OpenAI, OpenRouter, Custom / OpenAI-compatible - Key storage: SELF_HOSTED_ONLY | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit 11 days ago) Helicone sits in front of your model provider and records requests, latency, cost and errors, so you can see what an application is actually spending and sending. Be clear about the trade that involves: a logging proxy reads every prompt and response and holds the upstream key while forwarding. Running it yourself is what decides whether those logs sit with you or with a vendor, and this one is Apache-2.0 and self-hostable, alongside a hosted service from the same company. Key handling: You self-host the app, so your API key lives in your own deployment. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 75/100 (assessable for any licence), verifiability 90/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 21/25: The key lives in a deployment you run yourself. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under Apache-2.0, so anyone can check how the key is handled. - Privacy 7/15: Can be self-hosted, so the data path stays inside infrastructure you control. Analytics or error-reporting libraries are present; what they send was not established. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 7/10: Key handling was found in the published source by BYOAIK's scanner. What was checked (verification tier SOURCE_VERIFIED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [STRONG via SOURCE_SCAN] Reads the API key from an environment variable at runtime. (https://github.com/Helicone/helicone/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/Helicone/helicone/blob/HEAD/README.md) - [REPORTED via SOURCE_SCAN] References PostHog, so some analytics or error reporting is present. This scan cannot tell whether it is opt-in or what it sends. (https://github.com/Helicone/helicone/blob/HEAD/README.md) - [CONFIRMED via SOURCE_SCAN] Ships a container definition, so it can be self-hosted on your own infrastructure. (https://github.com/Helicone/helicone/blob/HEAD/Dockerfile) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-16, about 0 month(s) ago. (https://github.com/Helicone/helicone/commits) ### Postiz - Page: https://www.byoaik.com/tools/postiz/ - Website: https://postiz.com - Source: https://github.com/gitroomhq/postiz-app - Category: Marketing and SEO Tools | Pricing: OPEN_SOURCE | Open source: yes (AGPL-3.0) | Trust score: 82/100 - Platforms: Self-hosted, Docker, Web - Bring-your-own-key providers: OpenAI - Key storage: SELF_HOSTED_ONLY | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit today) Postiz is an open-source alternative to Buffer and Hootsuite for scheduling and analysing posts across more than 35 platforms. Self-hosted deployments set an OPENAI_API_KEY environment variable, documented as the key used for the copilot and AI image generation, which powers the in-app writing assistant. There is no separate paid AI tier in the self-hosted edition. Ships as a Docker Compose stack you run on your own server, so you pay for hosting and your own OpenAI usage. AGPL-3.0, 34,530 stars, daily commits. A managed cloud plan exists separately. Key handling: You self-host the app, so your API key lives in your own deployment. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 75/100 (assessable for any licence), verifiability 90/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 21/25: The key lives in a deployment you run yourself. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under AGPL-3.0, so anyone can check how the key is handled. Key handling was located in the published source. - Privacy 7/15: Can be self-hosted, so the data path stays inside infrastructure you control. Analytics or error-reporting libraries are present; what they send was not established. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 7/10: Key handling was found in the published source by BYOAIK's scanner. What was checked (verification tier SOURCE_VERIFIED, derived from this evidence): - [STRONG via SOURCE_SCAN] The user supplies their own OpenAI key: the project references its API key variable. (https://github.com/gitroomhq/postiz-app/blob/HEAD/.env.example) - [CONFIRMED via SOURCE_SCAN] Ships a container definition, so it can be self-hosted on your own infrastructure. (https://github.com/gitroomhq/postiz-app/blob/HEAD/docker-compose.yaml) - [REPORTED via SOURCE_SCAN] References PostHog, so some analytics or error reporting is present. This scan cannot tell whether it is opt-in or what it sends. (https://github.com/gitroomhq/postiz-app/blob/HEAD/package.json) - [REPORTED via SOURCE_SCAN] References Sentry, so some analytics or error reporting is present. This scan cannot tell whether it is opt-in or what it sends. (https://github.com/gitroomhq/postiz-app/blob/HEAD/package.json) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-29, about 0 month(s) ago. (https://github.com/gitroomhq/postiz-app/commits) ### TryPost - Page: https://www.byoaik.com/tools/trypost/ - Website: https://trypost.it/ - Source: https://github.com/trypostit/trypost - Category: Marketing and SEO Tools | Pricing: OPEN_SOURCE | Open source: yes (AGPL-3.0) | Trust score: 82/100 - Platforms: Web, Self-hosted, Docker - Bring-your-own-key providers: OpenAI, Anthropic, Google Gemini, OpenRouter, xAI Grok, Custom / OpenAI-compatible - Key storage: SELF_HOSTED_ONLY | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit 4 days ago) TryPost schedules and publishes to a dozen social networks with an AI copilot for drafting, and exposes an MCP server so your own assistant can post through it. Self-hosted, the posts, drafts and metrics stay on your infrastructure, and the AI side runs on credentials you supply: the project's .env.example carries OPENAI_API_KEY, ANTHROPIC_API_KEY, GEMINI_API_KEY, OPENROUTER_API_KEY and an XAI_API_KEY line. AGPL-3.0, with a hosted option for people who would rather not run servers. Key handling: You self-host the app, so your API key lives in your own deployment. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 75/100 (assessable for any licence), verifiability 90/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 21/25: The key lives in a deployment you run yourself. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under AGPL-3.0, so anyone can check how the key is handled. Key handling was located in the published source. - Privacy 7/15: Can be self-hosted, so the data path stays inside infrastructure you control. Analytics or error-reporting libraries are present; what they send was not established. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 7/10: Key handling was found in the published source by BYOAIK's scanner. What was checked (verification tier SOURCE_VERIFIED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [STRONG via SOURCE_SCAN] The user supplies their own OpenAI key: the project references its API key variable. (https://github.com/trypostit/trypost/blob/HEAD/.env.example) - [STRONG via SOURCE_SCAN] The user supplies their own Anthropic key: the project references its API key variable. (https://github.com/trypostit/trypost/blob/HEAD/.env.example) - [STRONG via SOURCE_SCAN] The user supplies their own Google Gemini key: the project references its API key variable. (https://github.com/trypostit/trypost/blob/HEAD/.env.example) - [STRONG via SOURCE_SCAN] The user supplies their own OpenRouter key: the project references its API key variable. (https://github.com/trypostit/trypost/blob/HEAD/.env.example) - [STRONG via SOURCE_SCAN] The user supplies their own Groq key: the project references its API key variable. (https://github.com/trypostit/trypost/blob/HEAD/.env.example) - [STRONG via SOURCE_SCAN] The user supplies their own Mistral key: the project references its API key variable. (https://github.com/trypostit/trypost/blob/HEAD/.env.example) - [STRONG via SOURCE_SCAN] The user supplies their own DeepSeek key: the project references its API key variable. (https://github.com/trypostit/trypost/blob/HEAD/.env.example) - [STRONG via SOURCE_SCAN] The user supplies their own xAI Grok key: the project references its API key variable. (https://github.com/trypostit/trypost/blob/HEAD/.env.example) - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/trypostit/trypost/blob/HEAD/.env.example) - [REPORTED via SOURCE_SCAN] References PostHog, so some analytics or error reporting is present. This scan cannot tell whether it is opt-in or what it sends. (https://github.com/trypostit/trypost/blob/HEAD/.env.example) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-28, about 0 month(s) ago. (https://github.com/trypostit/trypost/commits) ### OpenHands - Page: https://www.byoaik.com/tools/openhands/ - Website: https://www.all-hands.dev - Source: https://github.com/OpenHands/OpenHands - Category: Coding Assistants | Pricing: OPEN_SOURCE | Open source: yes (MIT) | Trust score: 81/100 - Platforms: Self-hosted, CLI, Web, Library/SDK, Docker - Bring-your-own-key providers: OpenAI, Anthropic, Google Gemini, Azure OpenAI, Mistral, Ollama, Custom / OpenAI-compatible - Key storage: ENVIRONMENT_VARIABLE | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit today) OpenHands (formerly OpenDevin) is an open-source, self-hostable autonomous coding agent that can write code, run commands, browse the web, and call APIs to complete software-engineering tasks. It is BYOK by design: the LLM settings let users bring their own model and API key, with a configurable Base URL for custom and OpenAI-compatible endpoints, plus CLI/env config via LLM_MODEL, LLM_API_KEY, and LLM_BASE_URL. Because it is built on LiteLLM, it works with virtually any provider; Anthropic, OpenAI, and Mistral are explicitly verified, and the custom Base URL covers Azure OpenAI, Gemini, and local servers like Ollama and vLLM. Available as a self-hosted web UI, a CLI, and a Python SDK/library. MIT-licensed with ~77.8k GitHub stars and active releases (v1.8.0, June 2026); an optional paid OpenHands Cloud is also offered. Key handling: Your API key is supplied via an environment variable or local config file. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 83/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 22/25: The key is supplied by an environment variable or local config file you control. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under MIT, so anyone can check how the key is handled. - Privacy 11/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. Analytics or error-reporting libraries are present; what they send was not established. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [REPORTED via SOURCE_SCAN] References PostHog, so some analytics or error reporting is present. This scan cannot tell whether it is opt-in or what it sends. (https://github.com/OpenHands/OpenHands/blob/HEAD/.env.sample) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-29, about 0 month(s) ago. (https://github.com/OpenHands/OpenHands/commits) ### Onlook - Page: https://www.byoaik.com/tools/onlook/ - Website: https://onlook.com - Source: https://github.com/onlook-dev/onlook - Category: App Builders | Pricing: OPEN_SOURCE | Open source: yes (Apache-2.0) | Trust score: 81/100 - Platforms: Web, Desktop, Self-hosted, Docker - Bring-your-own-key providers: OpenRouter, OpenAI, Anthropic, Google Gemini - Key storage: LOCAL_DEVICE | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Slowing down (last commit about 4 months ago) Onlook is an open-source, Figma-like visual editor for building real Next.js and Tailwind applications with AI, positioned against hosted builders. You can use the hosted service or self-host it: the self-hosting documentation has you copy the example environment file and set OPENROUTER_API_KEY to your own OpenRouter key, so chat and code generation bill to your account rather than Onlook's. Supports live visual editing with full code access to the generated app. Apache-2.0, 26,494 stars, pushed within a month of review. Bring-your-own-key applies to the self-hosted path; the hosted option uses Onlook's own billing. Key handling: Your API key is stored locally on your device. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 83/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 23/25: The key is held on your own device. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under Apache-2.0, so anyone can check how the key is handled. - Privacy 10/15: Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [CONFIRMED via SOURCE_SCAN] Ships a container definition, so it can be self-hosted on your own infrastructure. (https://github.com/onlook-dev/onlook/blob/HEAD/docker-compose.yml) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-07-22, about 2 month(s) ago. (https://github.com/onlook-dev/onlook/commits) ### InvokeAI - Page: https://www.byoaik.com/tools/invokeai/ - Website: https://invoke.ai/ - Source: https://github.com/invoke-ai/InvokeAI - Category: Image Generation Tools | Pricing: OPEN_SOURCE | Open source: yes (Apache-2.0) | Trust score: 81/100 - Platforms: Self-hosted, Desktop, Web, Docker - Bring-your-own-key providers: OpenAI, Google Gemini, Custom / OpenAI-compatible - Key storage: LOCAL_DEVICE | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit today) InvokeAI (Invoke) is an open-source (Apache-2.0) creative engine for Stable Diffusion, offering a professional web UI with a unified canvas, inpainting/outpainting, node-based workflows, gallery/board management, model installation, and upscaling. It runs local diffusion models (SD 1.5/2/SDXL/3.5, Flux, etc.) self-hosted on your own hardware. It also supports externally-hosted ("API-only") image models from cloud providers using your own API keys: users add credentials to an api_keys.yaml file (external_openai_api_key, external_gemini_api_key, plus optional external_openai_base_url / external_gemini_base_url for OpenAI-compatible proxies or regional endpoints) to install models from OpenAI (GPT Image 1/1.5/mini, DALL-E 3), Google Gemini (Gemini image models), BytePlus Seedream, and Alibaba Cloud DashScope (Qwen Image, Wan). The Community Edition is free and self-hosted; paid cloud generation plans existed but the hosted cloud service was slated for discontinuation in late 2025 while the open-source edition continues. The repo is actively maintained (~27.5k stars, last push June 2026). Key handling: Your API key is stored locally on your device. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 83/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 23/25: The key is held on your own device. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under Apache-2.0, so anyone can check how the key is handled. - Privacy 10/15: Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/invoke-ai/InvokeAI/blob/HEAD/README.md) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-27, about 0 month(s) ago. (https://github.com/invoke-ai/InvokeAI/commits) ### Baserow - Page: https://www.byoaik.com/tools/baserow/ - Website: https://baserow.io - Source: https://github.com/baserow/baserow - Category: Productivity Tools | Pricing: OPEN_SOURCE | Open source: yes (Mixed: MIT core, separate premium/enterprise licence) | Trust score: 81/100 - Platforms: Web, Self-hosted, Docker - Bring-your-own-key providers: OpenAI, Anthropic, Mistral, OpenRouter, Ollama - Key storage: SELF_HOSTED_ONLY | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit 2 days ago) Baserow is a self-hostable no-code database and spreadsheet platform. Its AI field type generates or transforms cell content with an LLM, and self-hosted instances supply their own provider credentials either per workspace under Generative AI settings or instance-wide through environment variables, connecting directly to OpenAI, Anthropic, Mistral, OpenRouter or a local Ollama endpoint. The open-source edition is MIT licensed while premium and enterprise directories carry separate terms, so GitHub reports no single licence for the repository. 5,571 stars, daily commits. Key handling: You self-host the app, so your API key lives in your own deployment. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 82/100 (assessable for any licence), verifiability 73/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 21/25: The key lives in a deployment you run yourself. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 15/20: Source is public, though the licence is unclear or non-standard. Key handling was located in the published source. - Privacy 11/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. Analytics or error-reporting libraries are present; what they send was not established. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 7/10: Key handling was found in the published source by BYOAIK's scanner. What was checked (verification tier SOURCE_VERIFIED, derived from this evidence): - [STRONG via SOURCE_SCAN] The user supplies their own Groq key: the project references its API key variable. (https://github.com/baserow/baserow/blob/HEAD/.env.example) - [STRONG via SOURCE_SCAN] The user supplies their own OpenAI key: the project references its API key variable. (https://github.com/baserow/baserow/blob/HEAD/docker-compose.yml) - [REPORTED via SOURCE_SCAN] References PostHog, so some analytics or error reporting is present. This scan cannot tell whether it is opt-in or what it sends. (https://github.com/baserow/baserow/blob/HEAD/docker-compose.yml) - [CONFIRMED via SOURCE_SCAN] Ships a container definition, so it can be self-hosted on your own infrastructure. (https://github.com/baserow/baserow/blob/HEAD/docker-compose.yml) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-29, about 0 month(s) ago. (https://github.com/baserow/baserow/commits) ### SiYuan - Page: https://www.byoaik.com/tools/siyuan/ - Website: https://b3log.org/siyuan/en/ - Source: https://github.com/siyuan-note/siyuan - Category: Writing Tools | Pricing: OPEN_SOURCE | Open source: yes (AGPL-3.0) | Trust score: 81/100 - Platforms: Desktop, Mobile, Self-hosted, Web, Docker - Bring-your-own-key providers: OpenAI, Azure OpenAI, Custom / OpenAI-compatible - Key storage: LOCAL_DEVICE | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit 11 days ago) SiYuan is an open-source, privacy-first personal knowledge management application built around block-level editing, bidirectional links, databases, and spaced-repetition flashcards. It runs fully self-hosted (desktop, mobile, and Docker/web) with local-first data storage. Its built-in AI writing and Q&A features (writing continuation, translation, grammar correction, chat) are powered by the end user's own credentials: Settings → AI lets you choose an API Provider (OpenAI or Azure), and enter your own API Key, API Base URL ("the URL of the OpenAI-compatible API"), and API Model. Because the base URL is user-configurable, any OpenAI-compatible endpoint works (DeepSeek, OpenRouter, local/self-hosted models via a compatible gateway, etc.). Key handling: Your API key is stored locally on your device. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 83/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 23/25: The key is held on your own device. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under AGPL-3.0, so anyone can check how the key is handled. - Privacy 10/15: Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [CONFIRMED via SOURCE_SCAN] Ships a container definition, so it can be self-hosted on your own infrastructure. (https://github.com/siyuan-note/siyuan/blob/HEAD/Dockerfile) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-29, about 0 month(s) ago. (https://github.com/siyuan-note/siyuan/commits) ### Rivet - Page: https://www.byoaik.com/tools/rivet/ - Website: https://rivet.ironcladapp.com - Source: https://github.com/Ironclad/rivet - Category: Prompt Engineering Tools | Pricing: OPEN_SOURCE | Open source: yes (MIT) | Trust score: 81/100 - Platforms: Desktop, Library/SDK, Self-hosted, Docker - Bring-your-own-key providers: OpenAI, Anthropic - Key storage: LOCAL_DEVICE | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Maintained (last commit about 2 months ago) Rivet is an open-source visual programming environment and TypeScript library for building and debugging AI agent graphs, created by Ironclad. You configure your own OpenAI key and organisation, or Anthropic key, in the settings panel or through environment variables, with no Ironclad-hosted credits involved. Ships as a cross-platform desktop app plus an embeddable library for running exported graphs in your own backend. MIT licensed, 4,662 stars, last commit about three months before review. Key handling: Your API key is stored locally on your device. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 83/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 23/25: The key is held on your own device. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under MIT, so anyone can check how the key is handled. - Privacy 10/15: Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-08-26, about 1 month(s) ago. (https://github.com/Ironclad/rivet/commits) ### Deep Research - Page: https://www.byoaik.com/tools/deep-research/ - Website: https://github.com/u14app/deep-research - Source: https://github.com/u14app/deep-research - Category: Research Tools | Pricing: OPEN_SOURCE | Open source: yes (MIT) | Trust score: 81/100 - Platforms: Web, Self-hosted, Docker - Bring-your-own-key providers: OpenAI, Anthropic, Google Gemini, Azure OpenAI, OpenRouter, Mistral, DeepSeek, xAI Grok, Ollama, Custom / OpenAI-compatible - Key storage: SELF_HOSTED_ONLY | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Slowing down (last commit about 5 months ago) Deep Research is an open-source (MIT) Next.js 15 / Shadcn UI web app that produces comprehensive, multi-step research reports by pairing LLMs with internet search. It splits work between "Thinking" and "Task" models to balance depth and speed, typically completing a report in about two minutes. It runs in two modes: a privacy-first "local API mode" where the end user enters their own provider API key in the in-app settings panel (persisted in browser localStorage via Zustand, with direct browser-to-provider calls so keys never touch a server), and a "proxy/server mode" where keys are set via environment variables behind an access password. It supports a wide range of providers and also exposes an SSE API and an MCP server. Deployable to Vercel, Cloudflare Pages, Docker, and static hosting (GitHub Pages), with PWA support, plus a hosted demo at research.u14.app. Key handling: You self-host the app, so your API key lives in your own deployment. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 87/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 21/25: The key lives in a deployment you run yourself. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under MIT, so anyone can check how the key is handled. - Privacy 14/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 8/10: Commits within the last six months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/u14app/deep-research/blob/HEAD/README.md) - [CONFIRMED via SOURCE_SCAN] Ships a container definition, so it can be self-hosted on your own infrastructure. (https://github.com/u14app/deep-research/blob/HEAD/docker-compose.yml) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-06-18, about 3 month(s) ago. (https://github.com/u14app/deep-research/commits) ### mem0 - Page: https://www.byoaik.com/tools/mem0/ - Website: https://mem0.ai/ - Source: https://github.com/mem0ai/mem0 - Category: Knowledge Management | Pricing: OPEN_SOURCE | Open source: yes (Apache-2.0) | Trust score: 81/100 - Platforms: CLI, Self-hosted, Docker - Bring-your-own-key providers: OpenAI, Anthropic, Google Gemini, Ollama, Custom / OpenAI-compatible - Key storage: ENVIRONMENT_VARIABLE | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit 2 days ago) mem0 gives AI agents persistent memory across sessions, and ships in three forms the README sets out side by side: a library you import, a server you self-host, and a managed cloud. The first two run on your infrastructure with your own provider credentials; the cloud is the company's hosted product with its own key, which is a different arrangement and worth not confusing with the other two. Apache-2.0. Key handling: Your API key is supplied via an environment variable or local config file. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 83/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 22/25: The key is supplied by an environment variable or local config file you control. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under Apache-2.0, so anyone can check how the key is handled. - Privacy 11/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. Analytics or error-reporting libraries are present; what they send was not established. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/mem0ai/mem0/blob/HEAD/pyproject.toml) - [REPORTED via SOURCE_SCAN] References PostHog, so some analytics or error reporting is present. This scan cannot tell whether it is opt-in or what it sends. (https://github.com/mem0ai/mem0/blob/HEAD/pyproject.toml) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-25, about 0 month(s) ago. (https://github.com/mem0ai/mem0/commits) ### Witsy - Page: https://www.byoaik.com/tools/witsy/ - Website: https://github.com/Kochava-Studios/witsy - Source: https://github.com/Kochava-Studios/witsy - Category: Desktop AI Assistants | Pricing: OPEN_SOURCE | Open source: yes (AGPL-3.0) | Trust score: 80/100 - Platforms: Desktop - Bring-your-own-key providers: OpenAI, Anthropic, Google Gemini, Azure OpenAI, OpenRouter, Groq, Mistral, Perplexity, DeepSeek, xAI Grok, Ollama, Hugging Face, Fireworks, Custom / OpenAI-compatible - Key storage: LOCAL_DEVICE | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Stale (last commit about 7 months ago) Witsy is an open-source (AGPL-3.0) Electron/Vue desktop AI assistant and universal Model Context Protocol (MCP) client for macOS, Windows, and Linux. It is explicitly a BYOK app: the README states "Witsy is a BYOK (Bring Your Own Keys) AI application: it means you need to have API keys for the LLM providers you want to use" (or run free local models via Ollama/LM Studio). It offers an unusually wide capability surface, all keyed by the user: chat with vision across OpenAI, Anthropic, Google Gemini, xAI Grok, Meta Llama, Mistral, DeepSeek, OpenRouter, Groq, Cerebras, Azure OpenAI, Ollama, LM Studio, and OpenAI-compatible endpoints; text-to-image and text-to-video (OpenAI, Google, xAI, Replicate, fal.ai, HuggingFace, Stable Diffusion WebUI); text-to-speech (OpenAI, ElevenLabs, Groq, fal.ai); speech-to-text/dictation (OpenAI Whisper, local Whisper, Groq, Fireworks, Gladia, nVidia, Speechmatics, Soniox, fal.ai); and internet search (Perplexity, Tavily, Brave, Exa, local Google). Additional features include RAG over local documents, MCP server integration, voice/realtime mode, Anthropic Computer Use, AI Commands, Prompt Anywhere, Experts, an HTTP API, a CLI, and agent webhooks. The project originated as nbonamy/witsy and has since moved to the Kochava-Studios organization (the old repo's README now reads "Witsy has a new home"); the canonical repo carries ~1,986 stars and 173 releases (latest v3.5.2). The witsyai.com domain currently 301-redirects to the GitHub repo rather than a standalone marketing site. Free and open source; users only pay their chosen providers. Key handling: Your API key is stored locally on your device. Requests are sent directly to the AI provider. Scores: key handling 85/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 23/25: The key is held on your own device. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under AGPL-3.0, so anyone can check how the key is handled. - Privacy 11/15: Local-first: it works without sending your data anywhere. - Maintenance 8/10: Commits within the last six months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/Kochava-Studios/witsy/blob/HEAD/README.md) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-04-09, about 6 month(s) ago. (https://github.com/Kochava-Studios/witsy/commits) ### Onyx - Page: https://www.byoaik.com/tools/onyx/ - Website: https://onyx.app - Source: https://github.com/onyx-dot-app/onyx - Category: Knowledge Management | Pricing: FREEMIUM | Open source: yes (MIT (Community Edition)) | Trust score: 80/100 - Platforms: Web, Self-hosted, Docker - Bring-your-own-key providers: OpenAI, Anthropic, Google Gemini, Azure OpenAI, OpenRouter, Ollama, Custom / OpenAI-compatible - Key storage: SELF_HOSTED_ONLY | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit today) Onyx (formerly Danswer) is an open-source AI platform for enterprise search and chat. It connects to 40+ knowledge sources (Slack, GitHub, Confluence, Google Drive, etc.) and grounds answers using hybrid search and advanced agentic RAG, with deep research, MCP, code interpreter, and web search. Admins configure their own LLM providers in the admin UI by pasting their own API keys, and it works with both proprietary providers (OpenAI, Anthropic, Gemini/Vertex, Azure OpenAI, Bedrock) and self-hosted/local models (Ollama, LM Studio, vLLM, LiteLLM), plus any OpenAI-compatible endpoint via a dedicated Custom Inference Provider config. The Community Edition is MIT-licensed and fully self-hostable via Docker/Kubernetes/Helm; a managed cloud and a feature-richer Enterprise Edition are also offered. Key handling: You self-host the app, so your API key lives in your own deployment. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 82/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 21/25: The key lives in a deployment you run yourself. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under MIT (Community Edition), so anyone can check how the key is handled. - Privacy 11/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. Analytics or error-reporting libraries are present; what they send was not established. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/onyx-dot-app/onyx/blob/HEAD/README.md) - [REPORTED via SOURCE_SCAN] References PostHog, so some analytics or error reporting is present. This scan cannot tell whether it is opt-in or what it sends. (https://github.com/onyx-dot-app/onyx/blob/HEAD/pyproject.toml) - [REPORTED via SOURCE_SCAN] References Sentry, so some analytics or error reporting is present. This scan cannot tell whether it is opt-in or what it sends. (https://github.com/onyx-dot-app/onyx/blob/HEAD/pyproject.toml) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-29, about 0 month(s) ago. (https://github.com/onyx-dot-app/onyx/commits) ### n8n - Page: https://www.byoaik.com/tools/n8n/ - Website: https://n8n.io/ - Source: https://github.com/n8n-io/n8n - Category: Automation Tools | Pricing: FREEMIUM | Open source: yes (Sustainable Use License (fair-code, source-available) + n8n Enterprise License) | Trust score: 80/100 - Platforms: Web, Self-hosted, Desktop, Library/SDK, Docker - Bring-your-own-key providers: OpenAI, Anthropic, Google Gemini, Azure OpenAI, OpenRouter, Groq, Mistral, Cohere, DeepSeek, xAI Grok, Ollama, Custom / OpenAI-compatible - Key storage: LOCAL_DEVICE | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit today) n8n is a source-available (fair-code) workflow automation platform for technical teams, combining a visual node-based builder with custom code. Its advanced-AI feature set is built on LangChain: an AI Agent node connects to chat-model sub-nodes for OpenAI, Anthropic, Google Gemini/Vertex, Azure OpenAI, Mistral, Cohere, Groq, Ollama, OpenRouter, DeepSeek, xAI Grok, AWS Bedrock and more. Each provider has its own credential type where the user pastes their own API key into n8n's encrypted credentials manager and selects it in the model node. Self-hosting keeps keys inside your own instance config and you pay the provider directly with no markup. Available as managed cloud, self-hosted (Docker/npm), and via desktop setups. Key handling: Your API key is stored locally on your device. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 90/100 (assessable for any licence), verifiability 53/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 23/25: The key is held on your own device. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 15/20: Source is public, though the licence is unclear or non-standard. - Privacy 14/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-29, about 0 month(s) ago. (https://github.com/n8n-io/n8n/commits) ### Local GPT (Obsidian) - Page: https://www.byoaik.com/tools/local-gpt-obsidian/ - Website: https://github.com/pfrankov/obsidian-local-gpt - Source: https://github.com/pfrankov/obsidian-local-gpt - Category: Writing Tools | Pricing: OPEN_SOURCE | Open source: yes (MIT) | Trust score: 80/100 - Platforms: Desktop, Mobile - Bring-your-own-key providers: OpenAI, Anthropic, Ollama, Custom / OpenAI-compatible - Key storage: LOCAL_DEVICE | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit 13 days ago) Local GPT adds AI actions inside Obsidian: summarising, rewriting, translating and brainstorming from selected note text: with local Ollama models as the privacy-preserving default. Cloud providers are configured through the companion AI Providers plugin by the same author, which the README directs users to install; that plugin acts as a hub for OpenAI-compatible, Anthropic and Ollama endpoints with keys the user enters. Works on desktop and mobile Obsidian. MIT licensed, 671 stars, pushed roughly three months before review. Key handling: Your API key is stored locally on your device. Requests are sent directly to the AI provider. Scores: key handling 85/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 23/25: The key is held on your own device. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under MIT, so anyone can check how the key is handled. - Privacy 11/15: Local-first: it works without sending your data anywhere. - Maintenance 8/10: Commits within the last six months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/pfrankov/obsidian-local-gpt/blob/HEAD/README.md) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-05-02, about 5 month(s) ago. (https://github.com/pfrankov/obsidian-local-gpt/commits) ### Dyad - Page: https://www.byoaik.com/tools/dyad/ - Website: https://dyad.sh - Source: https://github.com/dyad-sh/dyad - Category: App Builders | Pricing: OPEN_SOURCE | Open source: yes (Mixed: Apache-2.0 core, Functional Source License for pro directory) | Trust score: 80/100 - Platforms: Desktop - Bring-your-own-key providers: OpenAI, Anthropic, Google Gemini, OpenRouter, Ollama, Custom / OpenAI-compatible - Key storage: LOCAL_DEVICE | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit 2 days ago) Dyad is a free, open-source desktop AI app builder that scaffolds and edits full web apps from prompts, running locally rather than in a hosted sandbox. The README states plainly that you bring your own keys and use your own AI API keys with no vendor lock-in, entering credentials for OpenAI, Anthropic, Google Gemini or OpenRouter in settings, or pointing it at a local model through Ollama or LM Studio. Ships as a cross-platform macOS and Windows app needing no account to start. 21,249 stars, pushed the day of review; the core is Apache-2.0 with a separate pro directory under the Functional Source License. Key handling: Your API key is stored locally on your device. Requests are sent directly to the AI provider. Scores: key handling 80/100 (assessable for any licence), verifiability 73/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 23/25: The key is held on your own device. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 15/20: Source is public, though the licence is unclear or non-standard. Key handling was located in the published source. - Privacy 8/15: Local-first: it works without sending your data anywhere. Analytics or error-reporting libraries are present; what they send was not established. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 7/10: Key handling was found in the published source by BYOAIK's scanner. What was checked (verification tier SOURCE_VERIFIED, derived from this evidence): - [STRONG via SOURCE_SCAN] The user supplies their own OpenAI key: the project references its API key variable. (https://github.com/dyad-sh/dyad/blob/HEAD/.env.example) - [STRONG via SOURCE_SCAN] The user supplies their own Anthropic key: the project references its API key variable. (https://github.com/dyad-sh/dyad/blob/HEAD/.env.example) - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/dyad-sh/dyad/blob/HEAD/.env.example) - [REPORTED via SOURCE_SCAN] References PostHog, so some analytics or error reporting is present. This scan cannot tell whether it is opt-in or what it sends. (https://github.com/dyad-sh/dyad/blob/HEAD/package.json) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-29, about 0 month(s) ago. (https://github.com/dyad-sh/dyad/commits) ### Whisper (Obsidian) - Page: https://www.byoaik.com/tools/whisper-obsidian/ - Website: https://github.com/nikdanilov/whisper-obsidian-plugin - Source: https://github.com/nikdanilov/whisper-obsidian-plugin - Category: Voice and Transcription Tools | Pricing: OPEN_SOURCE | Open source: yes (MIT) | Trust score: 80/100 - Platforms: Desktop, Mobile - Bring-your-own-key providers: OpenAI, Anthropic, Azure OpenAI, Groq, Ollama, Custom / OpenAI-compatible - Key storage: LOCAL_DEVICE | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Stale (last commit about 7 months ago) Whisper is an open-source (MIT) Obsidian community plugin by nikdanilov for speech-to-text. You record audio from the sidebar mic (or Alt+Q), upload audio files, or right-click an audio file in your vault, and it transcribes via the Whisper API and inserts the text into your notes. Transcription works with OpenAI, Groq, Azure, or any Whisper-compatible API; the user supplies their own API key and can override the endpoint URL in settings (the code exposes apiKey/openAiApiKey fields and a configurable apiUrl defaulting to https://api.openai.com/v1/audio/transcriptions). Optional LLM post-processing (fix grammar, remove filler words, format as markdown, extract action items) supports Claude and GPT plus any OpenAI-compatible endpoint such as Ollama or LM Studio, again via a user-provided key (anthropicApiKey/postProcessingApiKey + postProcessingUrl). Runs on Obsidian desktop and mobile. Roughly 48.8k downloads and well-maintained (latest v1.9.1, last commit April 2026, 66 of 71 issues closed). It is a client plugin only; all provider costs are billed to the user's own accounts. Key handling: Your API key is stored locally on your device. Requests are sent directly to the AI provider. Scores: key handling 85/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 23/25: The key is held on your own device. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under MIT, so anyone can check how the key is handled. - Privacy 11/15: Local-first: it works without sending your data anywhere. - Maintenance 8/10: Commits within the last six months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/nikdanilov/whisper-obsidian-plugin/blob/HEAD/README.md) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-04-07, about 6 month(s) ago. (https://github.com/nikdanilov/whisper-obsidian-plugin/commits) ### Headroom - Page: https://www.byoaik.com/tools/headroom/ - Website: https://headroomlabs-ai.github.io/headroom/ - Source: https://github.com/headroomlabs-ai/headroom - Category: Developer Infrastructure | Pricing: OPEN_SOURCE | Open source: yes (Apache-2.0) | Trust score: 80/100 - Platforms: CLI, Self-hosted, Local-only, Docker - Bring-your-own-key providers: OpenAI, Anthropic, Google Gemini, Azure OpenAI, AWS Bedrock, OpenRouter, Custom / OpenAI-compatible, Ollama - Key storage: ENVIRONMENT_VARIABLE | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit yesterday) Headroom sits between your agent and the model, compressing tool output, logs, files and RAG chunks so fewer tokens are spent for the same answer. It runs as a local proxy (headroom proxy --port 8787), a Python library, or an MCP server, and needs no code changes: you point the agent at localhost and it forwards upstream with the credential you supply. Backends include OpenAI, Anthropic, Gemini, Azure OpenAI, AWS Bedrock, Vertex AI, OpenRouter and LiteLLM. Worth being clear about what a compression proxy is: it necessarily reads your prompts in the clear and holds the upstream token while forwarding. The difference here is where that happens, which is your own machine rather than a vendor's, and savings reporting is a local CLI rather than a dashboard it uploads to. Apache-2.0. Key handling: Your API key is supplied via an environment variable or local config file. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 82/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 22/25: The key is supplied by an environment variable or local config file you control. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under Apache-2.0, so anyone can check how the key is handled. - Privacy 10/15: Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [CONFIRMED via SOURCE_SCAN] Ships a container definition, so it can be self-hosted on your own infrastructure. (https://github.com/headroomlabs-ai/headroom/blob/HEAD/docker-compose.yml) - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/headroomlabs-ai/headroom/blob/HEAD/pyproject.toml) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-29, about 0 month(s) ago. (https://github.com/headroomlabs-ai/headroom/commits) ### Haystack - Page: https://www.byoaik.com/tools/haystack/ - Website: https://haystack.deepset.ai/ - Source: https://github.com/deepset-ai/haystack - Category: Agent Frameworks | Pricing: OPEN_SOURCE | Open source: yes (Apache-2.0) | Trust score: 80/100 - Platforms: Library/SDK, Self-hosted, Docker - Bring-your-own-key providers: OpenAI, Anthropic, Google Gemini, Azure OpenAI, Mistral, Cohere, Groq, Ollama, Hugging Face, Custom / OpenAI-compatible - Key storage: SELF_HOSTED_ONLY | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit 2 days ago) Haystack is deepset's open-source, vendor-agnostic AI orchestration framework for building production-ready LLM applications in Python. Developers compose modular pipelines and agent workflows with explicit control over retrieval, routing, memory, and generation, covering RAG, semantic search, multimodal, and conversational systems. It is bring-your-own-credentials by design: generator components (e.g. OpenAIGenerator) read keys from environment variables by default but accept a key passed directly at init via Secret.from_token(""), and OpenAIGenerator exposes a configurable api_base_url so end users can point at OpenAI-compatible or self-hosted/local endpoints. It integrates a very broad provider set including OpenAI, Anthropic, Mistral, Cohere, Azure OpenAI, Google Gemini (Google Gen AI / Vertex), Hugging Face, Ollama for local models, AWS Bedrock, NVIDIA, and Groq. Key handling: You self-host the app, so your API key lives in your own deployment. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 82/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 21/25: The key lives in a deployment you run yourself. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under Apache-2.0, so anyone can check how the key is handled. - Privacy 11/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. Analytics or error-reporting libraries are present; what they send was not established. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [REPORTED via SOURCE_SCAN] References PostHog, so some analytics or error reporting is present. This scan cannot tell whether it is opt-in or what it sends. (https://github.com/deepset-ai/haystack/blob/HEAD/pyproject.toml) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-29, about 0 month(s) ago. (https://github.com/deepset-ai/haystack/commits) ### GraphRAG - Page: https://www.byoaik.com/tools/graphrag/ - Website: https://microsoft.github.io/graphrag/ - Source: https://github.com/microsoft/graphrag - Category: Knowledge Management | Pricing: OPEN_SOURCE | Open source: yes (MIT) | Trust score: 80/100 - Platforms: CLI, Self-hosted, Local-only, Docker - Bring-your-own-key providers: OpenAI, Azure OpenAI, Custom / OpenAI-compatible - Key storage: ENVIRONMENT_VARIABLE | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit 7 days ago) GraphRAG is Microsoft Research's structured retrieval pipeline: it extracts entities and relationships from a corpus into a graph, then answers global questions that plain retrieval cannot. Model access is configured with GRAPHRAG_API_KEY against OpenAI or Azure OpenAI endpoints you nominate. Be deliberate about scale before running it: indexing passes every document through a model, so cost and exposure both scale with the size of the corpus rather than with how much you ask. MIT. Key handling: Your API key is supplied via an environment variable or local config file. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 82/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 22/25: The key is supplied by an environment variable or local config file you control. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under MIT, so anyone can check how the key is handled. - Privacy 10/15: Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-23, about 0 month(s) ago. (https://github.com/microsoft/graphrag/commits) ### Supermemory - Page: https://www.byoaik.com/tools/supermemory/ - Website: https://supermemory.ai/ - Source: https://github.com/supermemoryai/supermemory - Category: Knowledge Management | Pricing: OPEN_SOURCE | Open source: yes (MIT) | Trust score: 80/100 - Platforms: Self-hosted, Docker, Web - Bring-your-own-key providers: OpenAI, Anthropic, Google Gemini, Azure OpenAI, Ollama, Custom / OpenAI-compatible - Key storage: SELF_HOSTED_ONLY | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit 6 days ago) Supermemory is a memory layer for AI applications, storing and retrieving context across sessions, and is self-hostable with support for Anthropic, Gemini, Ollama and OpenAI-compatible endpoints on keys you supply. The company also runs a hosted API, which is the arrangement most users will meet first; the distinction between that and the self-hosted build is the whole question for anything holding your accumulated context. MIT. Key handling: You self-host the app, so your API key lives in your own deployment. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 82/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 21/25: The key lives in a deployment you run yourself. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under MIT, so anyone can check how the key is handled. - Privacy 11/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. Analytics or error-reporting libraries are present; what they send was not established. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/supermemoryai/supermemory/blob/HEAD/README.md) - [REPORTED via SOURCE_SCAN] References Sentry, so some analytics or error reporting is present. This scan cannot tell whether it is opt-in or what it sends. (https://github.com/supermemoryai/supermemory/blob/HEAD/package.json) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-29, about 0 month(s) ago. (https://github.com/supermemoryai/supermemory/commits) ### Open WebUI - Page: https://www.byoaik.com/tools/open-webui/ - Website: https://openwebui.com/ - Source: https://github.com/open-webui/open-webui - Category: AI Chat Interfaces | Pricing: OPEN_SOURCE | Open source: yes (Open WebUI License (source-available, with branding-preservation clause)) | Trust score: 79/100 - Platforms: Web, Self-hosted, Mobile, Docker - Bring-your-own-key providers: OpenAI, Ollama, OpenRouter, Groq, Mistral, Anthropic, Google Gemini, Custom / OpenAI-compatible - Key storage: BROWSER_LOCAL_STORAGE | Key transmission: DIRECT_TO_PROVIDER | Risk: MEDIUM - Maintenance: Actively developed (last commit today) Open WebUI is an extensible, self-hosted web interface for chatting with large language models. Originally an Ollama frontend, it is now fully provider-agnostic: users can add their own API keys and configure arbitrary OpenAI-compatible base URLs via both Admin Settings > Connections and per-user User Settings > Connections (Direct Connections), with personal keys stored client-side in browser local storage rather than on the server. This lets end users plug in their own OpenAI, OpenRouter, Groq, Mistral, LMStudio, Anthropic (OpenAI-compatible), and Google Gemini (OpenAI-compatible) keys, alongside native Ollama/local model support. It ships extensive features including RAG with multiple vector databases, model management, and a PWA for mobile. Deployable via Docker, pip, or Kubernetes; ~142k GitHub stars, 163 releases (v0.9.6 in June 2026), and very active development. Source-available under the Open WebUI License (includes a branding-preservation clause), with an optional commercial enterprise plan. Key handling: Your API key is stored in your browser's local storage. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 78/100 (assessable for any licence), verifiability 73/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 13/25: The key is persisted in browser local storage, which any script on the page can read. - Request Routing 20/20: Requests go straight from you to the AI provider. The provider endpoint is configurable, so you can point it at the provider or your own gateway. - Transparency 15/20: Source is public, though the licence is unclear or non-standard. Key handling was located in the published source. - Privacy 14/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 7/10: Key handling was found in the published source by BYOAIK's scanner. What was checked (verification tier SOURCE_VERIFIED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [STRONG via SOURCE_SCAN] The user supplies their own OpenAI key: the project references its API key variable. (https://github.com/open-webui/open-webui/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/open-webui/open-webui/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] The provider endpoint is configurable, so requests can be pointed straight at the provider or at your own gateway. (https://github.com/open-webui/open-webui/blob/HEAD/.env.example) - [CONFIRMED via SOURCE_SCAN] Ships a container definition, so it can be self-hosted on your own infrastructure. (https://github.com/open-webui/open-webui/blob/HEAD/docker-compose.yaml) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-21, about 0 month(s) ago. (https://github.com/open-webui/open-webui/commits) ### LibreChat - Page: https://www.byoaik.com/tools/librechat/ - Website: https://www.librechat.ai/ - Source: https://github.com/LibreChat-AI/LibreChat - Category: AI Chat Interfaces | Pricing: OPEN_SOURCE | Open source: yes (MIT) | Trust score: 79/100 - Platforms: Web, Self-hosted, Docker - Bring-your-own-key providers: OpenAI, Anthropic, Google Gemini, Azure OpenAI, OpenRouter, Groq, Mistral, DeepSeek, Ollama, Custom / OpenAI-compatible - Key storage: SERVER_SIDE_ENCRYPTED | Key transmission: DIRECT_TO_PROVIDER | Risk: MEDIUM - Maintenance: Actively developed (last commit 8 days ago) LibreChat is an MIT-licensed, self-hosted AI chat platform that provides a polished ChatGPT-style web UI across many providers. It supports AI agents, Model Context Protocol (MCP), artifacts, a sandboxed code interpreter, RAG/file chat, conversation search, multimodal input, and secure multi-user auth. BYOK is a first-class feature: administrators set provider keys (e.g. ANTHROPIC_API_KEY, OPENAI_API_KEY, GOOGLE_KEY) to 'user_provided', which prompts each end user to enter their own API key in the WebUI under settings. Custom OpenAI-compatible endpoints likewise accept apiKey: user_provided and even a user-supplied baseURL, so users can plug in their own keys for essentially any compatible provider. Deployable via Docker, docker-compose, and cloud templates (Railway, Zeabur, Sealos). Key handling: Your API key is stored encrypted on the server instance you control. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 70/100 (assessable for any licence), verifiability 90/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 11/25: The key is stored server-side, encrypted, on an instance you control. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under MIT, so anyone can check how the key is handled. Key handling was located in the published source. - Privacy 14/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 7/10: Key handling was found in the published source by BYOAIK's scanner. What was checked (verification tier SOURCE_VERIFIED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/LibreChat-AI/LibreChat/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] The user supplies their own OpenAI key: the project references its API key variable. (https://github.com/LibreChat-AI/LibreChat/blob/HEAD/.env.example) - [STRONG via SOURCE_SCAN] The user supplies their own Anthropic key: the project references its API key variable. (https://github.com/LibreChat-AI/LibreChat/blob/HEAD/.env.example) - [STRONG via SOURCE_SCAN] The user supplies their own Google Gemini key: the project references its API key variable. (https://github.com/LibreChat-AI/LibreChat/blob/HEAD/.env.example) - [STRONG via SOURCE_SCAN] The user supplies their own Groq key: the project references its API key variable. (https://github.com/LibreChat-AI/LibreChat/blob/HEAD/.env.example) - [STRONG via SOURCE_SCAN] The user supplies their own Mistral key: the project references its API key variable. (https://github.com/LibreChat-AI/LibreChat/blob/HEAD/.env.example) - [STRONG via SOURCE_SCAN] The user supplies their own DeepSeek key: the project references its API key variable. (https://github.com/LibreChat-AI/LibreChat/blob/HEAD/.env.example) - [STRONG via SOURCE_SCAN] The user supplies their own Cohere key: the project references its API key variable. (https://github.com/LibreChat-AI/LibreChat/blob/HEAD/.env.example) - [STRONG via SOURCE_SCAN] The user supplies their own xAI Grok key: the project references its API key variable. (https://github.com/LibreChat-AI/LibreChat/blob/HEAD/.env.example) - [STRONG via SOURCE_SCAN] The user supplies their own Fireworks key: the project references its API key variable. (https://github.com/LibreChat-AI/LibreChat/blob/HEAD/.env.example) - [STRONG via SOURCE_SCAN] The user supplies their own Perplexity key: the project references its API key variable. (https://github.com/LibreChat-AI/LibreChat/blob/HEAD/.env.example) - [CONFIRMED via SOURCE_SCAN] Ships a container definition, so it can be self-hosted on your own infrastructure. (https://github.com/LibreChat-AI/LibreChat/blob/HEAD/docker-compose.yml) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-29, about 0 month(s) ago. (https://github.com/LibreChat-AI/LibreChat/commits) ### Cherry Studio - Page: https://www.byoaik.com/tools/cherry-studio/ - Website: https://www.cherry-ai.com/ - Source: https://github.com/CherryHQ/cherry-studio - Category: Desktop AI Assistants | Pricing: OPEN_SOURCE | Open source: yes (AGPL-3.0) | Trust score: 79/100 - Platforms: Desktop - Bring-your-own-key providers: OpenAI, Anthropic, Google Gemini, Azure OpenAI, DeepSeek, Perplexity, Ollama, Custom / OpenAI-compatible - Key storage: LOCAL_DEVICE | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit today) Cherry Studio is an open-source (AGPL-3.0) cross-platform desktop AI productivity client for Windows, macOS, and Linux. It runs on a bring-your-own-key model: users paste their own API keys under Settings -> Model Providers for cloud providers (OpenAI, Anthropic, Google Gemini, Azure OpenAI, DeepSeek, and others) or run local models via Ollama and LM Studio. A Custom Provider option lets users enter their own API key plus a configurable Base URL to connect any OpenAI-, Anthropic-, Gemini-, or Azure-compatible endpoint (e.g. vLLM, OpenRouter, self-hosted gateways). The app bundles smart chat, RAG/knowledge bases, autonomous agents, and a large library of preset assistants. The project is actively maintained (v1.9.11 released June 2026, 7,000+ commits) and very popular (~47.6k GitHub stars). Key handling: Your API key is stored locally on your device. Requests are sent directly to the AI provider. Scores: key handling 80/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 23/25: The key is held on your own device. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under AGPL-3.0, so anyone can check how the key is handled. - Privacy 8/15: Local-first: it works without sending your data anywhere. Analytics or error-reporting libraries are present; what they send was not established. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/CherryHQ/cherry-studio/blob/HEAD/README.md) - [REPORTED via SOURCE_SCAN] References Sentry, so some analytics or error reporting is present. This scan cannot tell whether it is opt-in or what it sends. (https://github.com/CherryHQ/cherry-studio/blob/HEAD/package.json) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-29, about 0 month(s) ago. (https://github.com/CherryHQ/cherry-studio/commits) ### Chatbox - Page: https://www.byoaik.com/tools/chatbox/ - Website: https://chatboxai.app/ - Source: https://github.com/chatboxai/chatbox - Category: Desktop AI Assistants | Pricing: FREEMIUM | Open source: yes (GPL-3.0) | Trust score: 79/100 - Platforms: Desktop, Mobile, Web - Bring-your-own-key providers: OpenAI, Anthropic, Google Gemini, Azure OpenAI, OpenRouter, Ollama, Custom / OpenAI-compatible - Key storage: LOCAL_DEVICE | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit today) Chatbox is a popular cross-platform AI client (desktop, mobile, and web) for chatting with LLMs. Its Community Edition is open-source under GPL-3.0. It explicitly supports a Bring-Your-Own-Key model: per its BYOK docs, "you register and obtain API Keys from AI service providers yourself, then configure them in Chatbox," with data going "directly to AI providers, not through intermediary servers." Users can plug in their own keys for OpenAI, Anthropic (Claude), Google Gemini, Azure OpenAI, OpenRouter, and local runtimes (Ollama / LM Studio), as well as configure custom OpenAI-compatible endpoints. A separate optional paid hosted "Chatbox AI Service" subscription is also offered, but the app itself is free and BYOK-first. Key handling: Your API key is stored locally on your device. Requests are sent directly to the AI provider. Scores: key handling 80/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 23/25: The key is held on your own device. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under GPL-3.0, so anyone can check how the key is handled. - Privacy 8/15: Local-first: it works without sending your data anywhere. Analytics or error-reporting libraries are present; what they send was not established. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/chatboxai/chatbox/blob/HEAD/README.md) - [REPORTED via SOURCE_SCAN] References Sentry, so some analytics or error reporting is present. This scan cannot tell whether it is opt-in or what it sends. (https://github.com/chatboxai/chatbox/blob/HEAD/package.json) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-24, about 0 month(s) ago. (https://github.com/chatboxai/chatbox/commits) ### Aider - Page: https://www.byoaik.com/tools/aider/ - Website: https://aider.chat/ - Source: https://github.com/Aider-AI/aider - Category: Coding Assistants | Pricing: OPEN_SOURCE | Open source: yes (Apache-2.0) | Trust score: 79/100 - Platforms: CLI, Self-hosted, Docker - Bring-your-own-key providers: OpenAI, Anthropic, Google Gemini, Azure OpenAI, OpenRouter, Groq, Cohere, DeepSeek, xAI Grok, Ollama, Custom / OpenAI-compatible - Key storage: ENVIRONMENT_VARIABLE | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Slowing down (last commit about 6 months ago) Aider is an open-source, terminal-based AI pair programmer that edits code directly in your local git repository and auto-commits each change with a descriptive message. It supports 100+ programming languages, repo-map context, voice-to-code, and automatic linting/testing. It is fully bring-your-own-key: users supply their own LLM credentials via command-line flags (e.g. `--api-key anthropic=`, `--openai-api-key`), environment variables, a `.env` file, or a `.aider.conf.yml` config. It connects to almost any LLM including cloud providers and local models (Ollama, LM Studio) and custom OpenAI-compatible endpoints via configurable base URLs. Key handling: Your API key is supplied via an environment variable or local config file. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 83/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 22/25: The key is supplied by an environment variable or local config file you control. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under Apache-2.0, so anyone can check how the key is handled. - Privacy 11/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. Analytics or error-reporting libraries are present; what they send was not established. - Maintenance 8/10: Commits within the last six months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [REPORTED via SOURCE_SCAN] References PostHog, so some analytics or error reporting is present. This scan cannot tell whether it is opt-in or what it sends. (https://github.com/Aider-AI/aider/blob/HEAD/requirements.txt) - [REPORTED via SOURCE_SCAN] References Mixpanel, so some analytics or error reporting is present. This scan cannot tell whether it is opt-in or what it sends. (https://github.com/Aider-AI/aider/blob/HEAD/requirements.txt) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-05-22, about 4 month(s) ago. (https://github.com/Aider-AI/aider/commits) ### aichat - Page: https://www.byoaik.com/tools/aichat/ - Website: https://github.com/sigoden/aichat - Source: https://github.com/sigoden/aichat - Category: Developer Infrastructure | Pricing: OPEN_SOURCE | Open source: yes (Apache-2.0 OR MIT) | Trust score: 79/100 - Platforms: CLI, Self-hosted, Docker - Bring-your-own-key providers: OpenAI, Anthropic, Google Gemini, Azure OpenAI, OpenRouter, Groq, Mistral, Cohere, Perplexity, DeepSeek, xAI Grok, Ollama, Custom / OpenAI-compatible - Key storage: ENVIRONMENT_VARIABLE | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Stale (last commit about 8 months ago) aichat is a Rust-based, all-in-one command-line tool for working with large language models. It bundles a Shell Assistant (natural-language to shell commands), an interactive Chat-REPL with completion and history, Retrieval-Augmented Generation (RAG), reusable roles/sessions, function calling with MCP support, and AI agents that combine instructions, tools, and documents. It also ships a built-in HTTP server exposing an OpenAI-compatible API plus an LLM Playground and Arena. Users bring their own provider API keys, configured in config.yaml (per-provider api_key: xxx placeholders) or via _API_KEY environment variables; recent versions add an api_key_helper for dynamic credential fetching from secret managers. It supports 20+ providers including OpenAI, Anthropic Claude, Google Gemini, Azure OpenAI, AWS Bedrock, Google Vertex AI, Ollama, Groq, Mistral, Cohere, Perplexity, DeepSeek, xAI Grok, OpenRouter, GitHub Models, and any OpenAI-compatible endpoint. Runs on macOS, Linux, Windows, and Android (Termux); installable via Cargo, Homebrew, Scoop, Pacman, or prebuilt binaries. Key handling: Your API key is supplied via an environment variable or local config file. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 88/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 22/25: The key is supplied by an environment variable or local config file you control. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under Apache-2.0 OR MIT, so anyone can check how the key is handled. - Privacy 14/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 5/10: Last commit was within the last year. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/sigoden/aichat/blob/HEAD/README.md) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-02-23, about 7 month(s) ago. (https://github.com/sigoden/aichat/commits) ### Activepieces - Page: https://www.byoaik.com/tools/activepieces/ - Website: https://www.activepieces.com/ - Source: https://github.com/activepieces/activepieces - Category: Automation Tools | Pricing: OPEN_SOURCE | Open source: yes (MIT (Community Edition; enterprise features under a separate commercial license)) | Trust score: 79/100 - Platforms: Web, Self-hosted, Docker - Bring-your-own-key providers: OpenAI, Anthropic, Google Gemini, Custom / OpenAI-compatible - Key storage: SELF_HOSTED_ONLY | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit today) Activepieces is an open-source workflow automation platform (an "open replacement for Zapier") with a visual builder, 280+ integrations, native AI steps, and one of the largest open-source MCP toolkits for AI agents. It is BYOK: in the Admin Console AI settings (and per-connection for AI pieces), users add their own provider base URL and API key for OpenAI, Anthropic, and Google Gemini, and pay only for the tokens they consume. Base-URL configuration also supports AI gateways (Vercel, Cloudflare) and OpenAI-compatible endpoints. Fully self-hostable via Docker Compose under an MIT license, with a hosted cloud option also available. Key handling: You self-host the app, so your API key lives in your own deployment. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 80/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 21/25: The key lives in a deployment you run yourself. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under MIT (Community Edition; enterprise features under a separate commercial license), so anyone can check how the key is handled. - Privacy 10/15: Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [CONFIRMED via SOURCE_SCAN] Ships a container definition, so it can be self-hosted on your own infrastructure. (https://github.com/activepieces/activepieces/blob/HEAD/docker-compose.yml) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-29, about 0 month(s) ago. (https://github.com/activepieces/activepieces/commits) ### Auto-Analyst - Page: https://www.byoaik.com/tools/auto-analyst/ - Website: https://github.com/FireBird-Technologies/Auto-Analyst - Source: https://github.com/FireBird-Technologies/Auto-Analyst - Category: Document Analysis | Pricing: OPEN_SOURCE | Open source: yes (MIT) | Trust score: 79/100 - Platforms: Web, Self-hosted, Docker - Bring-your-own-key providers: OpenAI, Anthropic, Google Gemini, Custom / OpenAI-compatible - Key storage: SELF_HOSTED_ONLY | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Maintained (last commit about 3 months ago) Auto-Analyst is a modular open-source system for automating data science work, from cleaning and statistical analysis through modelling to visualisation, driven by agents you point at your own provider key. It is self-hosted, which matters more than usual here: the data an analysis agent touches is typically the organisation's own, so keeping both the dataset and the credential inside your deployment is the point. MIT. Key handling: You self-host the app, so your API key lives in your own deployment. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 80/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 21/25: The key lives in a deployment you run yourself. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under MIT, so anyone can check how the key is handled. - Privacy 10/15: Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-08-20, about 1 month(s) ago. (https://github.com/FireBird-Technologies/Auto-Analyst/commits) ### SGLang - Page: https://www.byoaik.com/tools/sglang/ - Website: https://github.com/sgl-project/sglang - Source: https://github.com/sgl-project/sglang - Category: Self-Hosted AI Platforms | Pricing: OPEN_SOURCE | Open source: yes (Apache-2.0) | Trust score: 79/100 - Platforms: Self-hosted, Docker, Local-only - Bring-your-own-key providers: Custom / OpenAI-compatible - Key storage: SELF_HOSTED_ONLY | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit today) SGLang is a serving runtime for large language and vision models, built around a fast execution backend and a structured generation language. It serves an OpenAI-compatible HTTP endpoint from your own hardware, so any BYOK client can be pointed at it and no cloud key is involved at all. Widely used for self-hosted deployments where throughput and control both matter. Apache-2.0. Key handling: You self-host the app, so your API key lives in your own deployment. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 80/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 21/25: The key lives in a deployment you run yourself. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under Apache-2.0, so anyone can check how the key is handled. - Privacy 10/15: Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/sgl-project/sglang/blob/HEAD/README.md) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-29, about 0 month(s) ago. (https://github.com/sgl-project/sglang/commits) ### AI Engine (WordPress) - Page: https://www.byoaik.com/tools/ai-engine-wordpress/ - Website: https://wordpress.org/plugins/ai-engine/ - Source: https://github.com/jordymeow/ai-engine - Category: Marketing and SEO Tools | Pricing: OPEN_SOURCE | Open source: yes (GPL-2.0-or-later) | Trust score: 79/100 - Platforms: Web, Self-hosted, Docker - Bring-your-own-key providers: OpenAI, Anthropic, Google Gemini, Mistral, Perplexity, OpenRouter, xAI Grok, Azure OpenAI, Custom / OpenAI-compatible - Key storage: SELF_HOSTED_ONLY | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit 2 days ago) AI Engine is a WordPress plugin adding AI chatbots, content generation, image tools and an MCP server to any WordPress site. Setup requires the administrator to obtain an API key from their chosen provider and add it in the plugin settings; no AI credits ship with it. Stable connectors cover OpenAI, Anthropic, Google, xAI, Mistral, Perplexity, Azure OpenAI and OpenRouter, plus a custom connector for self-hosted Ollama, LM Studio, vLLM or LocalAI backends. GPL-2.0-or-later, over 100,000 active WordPress installs; the GitHub mirror carries few stars because WordPress.org is the real distribution channel. Key handling: You self-host the app, so your API key lives in your own deployment. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 80/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 21/25: The key lives in a deployment you run yourself. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under GPL-2.0-or-later, so anyone can check how the key is handled. - Privacy 10/15: Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-25, about 0 month(s) ago. (https://github.com/jordymeow/ai-engine/commits) ### Dify - Page: https://www.byoaik.com/tools/dify/ - Website: https://dify.ai/ - Source: https://github.com/langgenius/dify - Category: Agent Frameworks | Pricing: FREEMIUM | Open source: yes (Dify Open Source License (Apache 2.0-based with additional conditions)) | Trust score: 78/100 - Platforms: Web, Self-hosted, Docker - Bring-your-own-key providers: OpenAI, Anthropic, Google Gemini, Azure OpenAI, Mistral, DeepSeek, Cohere, Ollama, Custom / OpenAI-compatible - Key storage: SELF_HOSTED_ONLY | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit today) Dify is a production-ready, open-source platform for developing AI applications, agentic workflows, and RAG pipelines through a visual builder. Users configure their own LLM credentials under Settings -> Model Providers, where "Custom Providers use your own API keys for direct access to model providers like OpenAI, Anthropic, or Google," with full control and direct billing. It supports hundreds of proprietary and open-source models from dozens of inference providers, including OpenAI-API-compatible endpoints and local models via Ollama. The Community Edition is free and self-hostable via Docker Compose, Kubernetes, or Terraform; a hosted Dify Cloud and Enterprise edition are also offered. Key handling: You self-host the app, so your API key lives in your own deployment. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 87/100 (assessable for any licence), verifiability 53/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 21/25: The key lives in a deployment you run yourself. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 15/20: Source is public, though the licence is unclear or non-standard. - Privacy 14/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-29, about 0 month(s) ago. (https://github.com/langgenius/dify/commits) ### Warp - Page: https://www.byoaik.com/tools/warp/ - Website: https://www.warp.dev/ - Source: https://github.com/warpdotdev/warp - Category: Developer Infrastructure | Pricing: FREEMIUM | Open source: yes (AGPL-3.0 (client); MIT (warpui_core/warpui UI crates); cloud backend proprietary) | Trust score: 78/100 - Platforms: Desktop, CLI - Bring-your-own-key providers: OpenAI, Anthropic, Google Gemini, OpenRouter, Custom / OpenAI-compatible - Key storage: LOCAL_DEVICE | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit 2 days ago) Warp is a Rust-based, AI-native terminal that has evolved into an "agentic development environment." Its built-in AI agent can run commands, debug, and orchestrate multi-step work directly in the shell, with MCP support. Founded in 2020 by Zach Lloyd (ex-Google). Warp confirms BYOK: in Settings > AI users add their own OpenAI, Anthropic, or Google API key, or configure a custom inference endpoint compatible with the OpenAI Chat Completions API (examples given: OpenRouter, LiteLLM, z.ai, internal gateways, self-hosted). Keys are stored only on the user's device (OS keychain) and used in-flight by Warp's agent backend, never persisted on Warp servers; BYOK requests are billed by the provider directly and do not consume Warp credits. As of May 21, 2026, BYOK is available on every plan including Free, for individuals and orgs with 10 or fewer employees (larger orgs need Business/Enterprise). Note: the candidate's "closed-source" label is outdated. In April 2026 Warp open-sourced its terminal client and Oz cloud-agent orchestration platform (AGPL-3.0, with the warpui_core/warpui UI crates under MIT) at github.com/warpdotdev/warp (~62k stars); however the broader agent harness backend and cloud services like Warp Drive remain proprietary, so it is a hybrid open-core/freemium model. Platforms: macOS (mature), Linux (production), Windows (preview). Key handling: Your API key is stored locally on your device. Requests are sent directly to the AI provider. Scores: key handling 78/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 23/25: The key is held on your own device. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under AGPL-3.0 (client); MIT (warpui_core/warpui UI crates); cloud backend proprietary, so anyone can check how the key is handled. - Privacy 7/15: Nothing specific is known about its data handling beyond the key itself. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-29, about 0 month(s) ago. (https://github.com/warpdotdev/warp/commits) ### AutoGPT - Page: https://www.byoaik.com/tools/autogpt/ - Website: https://agpt.co/ - Source: https://github.com/Significant-Gravitas/AutoGPT - Category: Agent Frameworks | Pricing: OPEN_SOURCE | Open source: yes (PolyForm Shield (autogpt_platform) + MIT (classic/rest of repo)) | Trust score: 78/100 - Platforms: Web, Self-hosted, Docker - Bring-your-own-key providers: OpenAI, Anthropic, Groq, Ollama, OpenRouter - Key storage: SELF_HOSTED_ONLY | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit 3 days ago) AutoGPT is a pioneering autonomous-agent platform for building, deploying, and running continuous AI agents that automate multi-step workflows. It offers a low-code/visual block-based agent builder and a full server platform. When self-hosted (single Docker Compose command), users bring their own model API keys; the platform's LLM blocks support OpenAI, Anthropic, Groq, Ollama (local), and Open Router (which fronts Gemini, Mistral, Cohere, DeepSeek, Perplexity, etc.), plus AIML API, Llama API, and Vercel v0, each configured via user-supplied api_key credentials. A managed hosted cloud tier also exists where model usage is covered by a paid plan (no BYOK there). Key handling: You self-host the app, so your API key lives in your own deployment. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 87/100 (assessable for any licence), verifiability 53/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 21/25: The key lives in a deployment you run yourself. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 15/20: Source is public, though the licence is unclear or non-standard. - Privacy 14/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-27, about 0 month(s) ago. (https://github.com/Significant-Gravitas/AutoGPT/commits) ### Open Cowork - Page: https://www.byoaik.com/tools/open-cowork/ - Website: https://opencoworkai.github.io/open-cowork/ - Source: https://github.com/OpenCoworkAI/open-cowork - Category: Desktop AI Assistants | Pricing: OPEN_SOURCE | Open source: yes (MIT) | Trust score: 78/100 - Platforms: Desktop, Local-only - Bring-your-own-key providers: OpenAI, Anthropic, Google Gemini, DeepSeek, MiniMax, Moonshot AI (Kimi), Z.AI, Custom / OpenAI-compatible - Key storage: LOCAL_DEVICE | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit 19 days ago) Open Cowork is a desktop AI agent application with a one-click install, built around bringing your own API key rather than an account with the publisher. It works with Claude, GPT, Gemini, DeepSeek, GLM, MiniMax, Kimi and any OpenAI-compatible endpoint, which is a wider provider spread than most desktop clients and notably includes several of the Chinese providers this directory now tracks. MIT. Key handling: Your API key is stored locally on your device. Requests are sent directly to the AI provider. Scores: key handling 78/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 23/25: The key is held on your own device. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under MIT, so anyone can check how the key is handled. - Privacy 7/15: Nothing specific is known about its data handling beyond the key itself. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-28, about 0 month(s) ago. (https://github.com/OpenCoworkAI/open-cowork/commits) ### Microsoft Prompt Flow - Page: https://www.byoaik.com/tools/microsoft-prompt-flow/ - Website: https://microsoft.github.io/promptflow/ - Source: https://github.com/microsoft/promptflow - Category: Prompt Engineering Tools | Pricing: OPEN_SOURCE | Open source: yes (MIT) | Trust score: 78/100 - Platforms: Library/SDK, CLI, VS Code Extension, Self-hosted, Docker - Bring-your-own-key providers: OpenAI, Azure OpenAI, Custom / OpenAI-compatible - Key storage: ENVIRONMENT_VARIABLE | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit 20 days ago) Prompt flow is Microsoft's open-source toolkit for prototyping, testing, tuning and evaluating prompt flows, with a VS Code extension plus a Python SDK and CLI. You create local connection files holding your own credentials and flows reference those connections by name, so keys stay on your machine and are never proxied through a Microsoft service. Installed with pip. MIT licensed, 11,218 stars, last commit about three months before review. Key handling: Your API key is supplied via an environment variable or local config file. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 82/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 22/25: The key is supplied by an environment variable or local config file you control. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under MIT, so anyone can check how the key is handled. - Privacy 10/15: Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 8/10: Commits within the last six months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-05-14, about 5 month(s) ago. (https://github.com/microsoft/promptflow/commits) ### vim-ai - Page: https://www.byoaik.com/tools/vim-ai/ - Website: https://github.com/madox2/vim-ai - Source: https://github.com/madox2/vim-ai - Category: IDE Plugins | Pricing: OPEN_SOURCE | Open source: yes (MIT) | Trust score: 78/100 - Platforms: CLI, Local-only - Bring-your-own-key providers: OpenAI, Ollama, Custom / OpenAI-compatible - Key storage: ENVIRONMENT_VARIABLE | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit 25 days ago) vim-ai adds completion, in-place editing and chat to Vim and Neovim, working with OpenAI and any OpenAI-compatible endpoint including local servers. The key is read from a path you nominate rather than being pasted into your vimrc, which matters for a dotfile most people keep in public version control. Note the pace: the last commit was several months before review, so the maintenance badge on this listing reads accordingly. MIT. Key handling: Your API key is supplied via an environment variable or local config file. Requests are sent directly to the AI provider. Scores: key handling 77/100 (assessable for any licence), verifiability 90/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 22/25: The key is supplied by an environment variable or local config file you control. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under MIT, so anyone can check how the key is handled. Key handling was located in the published source. - Privacy 7/15: Nothing specific is known about its data handling beyond the key itself. - Maintenance 5/10: Last commit was within the last year. - Verification Confidence 7/10: Key handling was found in the published source by BYOAIK's scanner. What was checked (verification tier SOURCE_VERIFIED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [STRONG via SOURCE_SCAN] The user supplies their own OpenAI key: the project references its API key variable. (https://github.com/madox2/vim-ai/blob/HEAD/README.md) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-03-11, about 7 month(s) ago. (https://github.com/madox2/vim-ai/commits) ### OpenAI Codex CLI - Page: https://www.byoaik.com/tools/openai-codex-cli/ - Website: https://developers.openai.com/codex/cli/ - Source: https://github.com/openai/codex - Category: Coding Assistants | Pricing: OPEN_SOURCE | Open source: yes (Apache-2.0) | Trust score: 77/100 - Platforms: CLI - Bring-your-own-key providers: OpenAI, Azure OpenAI, OpenRouter, Mistral, DeepSeek, Ollama, Custom / OpenAI-compatible - Key storage: ENVIRONMENT_VARIABLE | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit today) OpenAI Codex CLI is a lightweight, open-source (Apache-2.0) coding agent written in Rust that runs in the terminal on macOS, Linux, and Windows. It reads and edits files in your repo, runs shell commands inside a sandbox, and iterates on tasks. While OpenAI promotes ChatGPT-plan sign-in, an end user can bring their own key: authenticate with your own OPENAI_API_KEY (e.g. `printenv OPENAI_API_KEY | codex login --with-api-key`), and configure custom model providers in ~/.codex/config.toml. The [model_providers.] table accepts base_url, env_key (your own API-key env var), and wire_api ("responses" or "chat"), so you can point Codex at any OpenAI-compatible endpoint. Official docs show examples for Azure OpenAI, Ollama, LM Studio, and Mistral; community guides cover OpenRouter, DeepSeek, and other gateways. Key handling: Your API key is supplied via an environment variable or local config file. Requests are sent directly to the AI provider. Scores: key handling 77/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 22/25: The key is supplied by an environment variable or local config file you control. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under Apache-2.0, so anyone can check how the key is handled. - Privacy 7/15: Nothing specific is known about its data handling beyond the key itself. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-29, about 0 month(s) ago. (https://github.com/openai/codex/commits) ### Zero (Mail0) - Page: https://www.byoaik.com/tools/zero-mail0/ - Website: https://0.email - Source: https://github.com/Mail-0/Zero - Category: Productivity Tools | Pricing: OPEN_SOURCE | Open source: yes (MIT) | Trust score: 77/100 - Platforms: Web, Self-hosted, Docker - Bring-your-own-key providers: OpenAI - Key storage: SELF_HOSTED_ONLY | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit 13 days ago) Zero is an open-source email client offering a unified inbox across providers together with AI features for composing, summarising and triaging mail. The repository's .env.example lists OPENAI_API_KEY along with model selection variables, so an operator running their own instance supplies their own key rather than routing through a Zero-owned key or subscription. Self-hosted as a standard Next.js deployment. MIT licensed, 10,766 stars, last pushed about two and a half months before review. The project states it does not track, collect or sell email data. Key handling: You self-host the app, so your API key lives in your own deployment. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 80/100 (assessable for any licence), verifiability 90/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 21/25: The key lives in a deployment you run yourself. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under MIT, so anyone can check how the key is handled. Key handling was located in the published source. - Privacy 10/15: Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 2/10: Last commit was about 13 months ago. - Verification Confidence 7/10: Key handling was found in the published source by BYOAIK's scanner. What was checked (verification tier SOURCE_VERIFIED, derived from this evidence): - [STRONG via SOURCE_SCAN] The user supplies their own OpenAI key: the project references its API key variable. (https://github.com/Mail-0/Zero/blob/HEAD/.env.example) - [STRONG via SOURCE_SCAN] The user supplies their own Perplexity key: the project references its API key variable. (https://github.com/Mail-0/Zero/blob/HEAD/.env.example) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2025-08-31, about 13 month(s) ago. (https://github.com/Mail-0/Zero/commits) ### Kilo Code - Page: https://www.byoaik.com/tools/kilo-code/ - Website: https://kilo.ai/code - Source: https://github.com/Kilo-Org/kilocode - Category: Coding Assistants | Pricing: OPEN_SOURCE | Open source: yes (MIT) | Trust score: 77/100 - Platforms: VS Code Extension, JetBrains Plugin, CLI - Bring-your-own-key providers: OpenRouter, Anthropic, OpenAI, Google Gemini, Azure OpenAI, Groq, Mistral, DeepSeek, Ollama, Custom / OpenAI-compatible - Key storage: ENVIRONMENT_VARIABLE | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit yesterday) Kilo Code is an open-source agentic coding assistant that runs as a VS Code extension, JetBrains plugin, and CLI (with cloud and code-review offerings). Users can bring their own API keys for any provider at zero markup, configured in the extension's Settings under a Providers tab, or run models locally via Ollama / LM Studio. It also offers an optional hosted credits marketplace (Kilo Pass / pay-as-you-go) routing to 500+ models, but markets that BYOK lets you pay providers their exact list price with no commission or hidden fees. Key handling: Your API key is supplied via an environment variable or local config file. Requests are sent directly to the AI provider. Scores: key handling 77/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 22/25: The key is supplied by an environment variable or local config file you control. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under MIT, so anyone can check how the key is handled. - Privacy 7/15: Nothing specific is known about its data handling beyond the key itself. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-29, about 0 month(s) ago. (https://github.com/Kilo-Org/kilocode/commits) ### llm.nvim - Page: https://www.byoaik.com/tools/llm-nvim/ - Website: https://github.com/huggingface/llm.nvim - Source: https://github.com/huggingface/llm.nvim - Category: IDE Plugins | Pricing: OPEN_SOURCE | Open source: yes (Apache-2.0) | Trust score: 77/100 - Platforms: CLI, Local-only - Bring-your-own-key providers: Hugging Face, Ollama, Custom / OpenAI-compatible - Key storage: ENVIRONMENT_VARIABLE | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit 2 days ago) llm.nvim is Hugging Face's Neovim plugin for LLM-backed code completion, using llm-ls as its backend. It can talk to the Hugging Face Inference API or any HTTP endpoint you nominate, with the backend URL overridable through the LLM_NVIM_URL environment variable, which means it works equally against a local Ollama or vLLM server and against a hosted provider on your own token. Apache-2.0. Key handling: Your API key is supplied via an environment variable or local config file. Requests are sent directly to the AI provider. Scores: key handling 77/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 22/25: The key is supplied by an environment variable or local config file you control. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under Apache-2.0, so anyone can check how the key is handled. - Privacy 7/15: Nothing specific is known about its data handling beyond the key itself. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/huggingface/llm.nvim/blob/HEAD/README.md) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-23, about 0 month(s) ago. (https://github.com/huggingface/llm.nvim/commits) ### LangChain - Page: https://www.byoaik.com/tools/langchain/ - Website: https://www.langchain.com - Source: https://github.com/langchain-ai/langchain - Category: Agent Frameworks | Pricing: OPEN_SOURCE | Open source: yes (MIT) | Trust score: 77/100 - Platforms: Library/SDK, Self-hosted - Bring-your-own-key providers: OpenAI, Anthropic, Google Gemini, OpenRouter, Ollama, Custom / OpenAI-compatible, AWS Bedrock, Azure OpenAI, Mistral, Cohere, Groq, DeepSeek - Key storage: ENVIRONMENT_VARIABLE | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit today) LangChain provides the building blocks (chains, agents, tools, memory) for building coding assistants, RAG systems and autonomous agents on top of any LLM. To use Claude, for example, the user sets the ANTHROPIC_API_KEY environment variable or passes api_key directly to ChatAnthropic, and the same pattern applies for OpenAI, Gemini, Bedrock and dozens of other integrations. MIT licensed and the most-starred project in this space, with same-day commit activity. Key handling: Your API key is supplied via an environment variable or local config file. Requests are sent directly to the AI provider. Scores: key handling 77/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 22/25: The key is supplied by an environment variable or local config file you control. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under MIT, so anyone can check how the key is handled. - Privacy 7/15: Nothing specific is known about its data handling beyond the key itself. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-29, about 0 month(s) ago. (https://github.com/langchain-ai/langchain/commits) ### Continue - Page: https://www.byoaik.com/tools/continue/ - Website: https://continue.dev - Source: https://github.com/continuedev/continue - Category: Coding Assistants | Pricing: OPEN_SOURCE | Open source: yes (Apache-2.0) | Trust score: 77/100 - Platforms: VS Code Extension, JetBrains Plugin, CLI - Bring-your-own-key providers: OpenAI, Anthropic, Google Gemini, OpenRouter, Ollama, Custom / OpenAI-compatible, Azure OpenAI, Mistral, Groq, DeepSeek, AWS Bedrock - Key storage: ENVIRONMENT_VARIABLE | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Slowing down (last commit about 4 months ago) Continue adds chat, autocomplete, inline edit, and codebase-aware agent mode to VS Code and JetBrains IDEs, with a fully model-agnostic backend. Users add their key in config.yaml (for example 'provider: anthropic' with 'apiKey: ') for Anthropic, OpenAI, Gemini and dozens of other providers, or point at a local Ollama or LM Studio endpoint with no key at all. Apache-2.0 licensed with 35,000+ stars and same-day commit activity. No account or subscription is required to use your own key. Key handling: Your API key is supplied via an environment variable or local config file. Requests are sent directly to the AI provider. Scores: key handling 77/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 22/25: The key is supplied by an environment variable or local config file you control. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under Apache-2.0, so anyone can check how the key is handled. - Privacy 7/15: Nothing specific is known about its data handling beyond the key itself. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-07-21, about 2 month(s) ago. (https://github.com/continuedev/continue/commits) ### MarkItDown - Page: https://www.byoaik.com/tools/markitdown/ - Website: https://github.com/microsoft/markitdown - Source: https://github.com/microsoft/markitdown - Category: Document Analysis | Pricing: OPEN_SOURCE | Open source: yes (MIT) | Trust score: 77/100 - Platforms: CLI, Local-only - Bring-your-own-key providers: OpenAI, Azure OpenAI, Custom / OpenAI-compatible - Key storage: ENVIRONMENT_VARIABLE | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit 2 days ago) MarkItDown converts Office documents, PDFs, images, audio and more into markdown for LLM pipelines, preserving headings, lists, tables and links. Conversion itself needs no model. Where AI does come in, for image descriptions and the OCR plugin, the README's pattern is that you pass your own llm_client and llm_model, so the credential and the choice of provider stay with you rather than being baked in. MIT, and by a distance the most-starred project in this directory. Key handling: Your API key is supplied via an environment variable or local config file. Requests are sent directly to the AI provider. Scores: key handling 77/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 22/25: The key is supplied by an environment variable or local config file you control. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under MIT, so anyone can check how the key is handled. - Privacy 7/15: Nothing specific is known about its data handling beyond the key itself. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [CONFIRMED via SOURCE_SCAN] Ships a container definition, so it can be self-hosted on your own infrastructure. (https://github.com/microsoft/markitdown/blob/HEAD/Dockerfile) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-21, about 0 month(s) ago. (https://github.com/microsoft/markitdown/commits) ### Agent Browser - Page: https://www.byoaik.com/tools/agent-browser/ - Website: https://github.com/vercel-labs/agent-browser - Source: https://github.com/vercel-labs/agent-browser - Category: Automation Tools | Pricing: OPEN_SOURCE | Open source: yes (Apache-2.0) | Trust score: 77/100 - Platforms: CLI, Local-only - Bring-your-own-key providers: OpenAI, Anthropic, Custom / OpenAI-compatible - Key storage: ENVIRONMENT_VARIABLE | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit yesterday) Agent Browser is a Rust command-line tool from Vercel Labs that hands a coding agent direct control of a browser, designed to slot into an existing agent setup rather than replace it. It runs locally and uses whichever model and credential your agent is already configured with, so it adds browser capability without adding a party to the request path. Apache-2.0. Key handling: Your API key is supplied via an environment variable or local config file. Requests are sent directly to the AI provider. Scores: key handling 77/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 22/25: The key is supplied by an environment variable or local config file you control. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under Apache-2.0, so anyone can check how the key is handled. - Privacy 7/15: Nothing specific is known about its data handling beyond the key itself. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-22, about 0 month(s) ago. (https://github.com/vercel-labs/agent-browser/commits) ### opencode.nvim - Page: https://www.byoaik.com/tools/opencode-nvim/ - Website: https://github.com/nickjvandyke/opencode.nvim - Source: https://github.com/nickjvandyke/opencode.nvim - Category: IDE Plugins | Pricing: OPEN_SOURCE | Open source: yes (MIT) | Trust score: 77/100 - Platforms: CLI, Local-only - Bring-your-own-key providers: OpenAI, Anthropic, Google Gemini, OpenRouter, Ollama, Custom / OpenAI-compatible - Key storage: ENVIRONMENT_VARIABLE | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit 9 days ago) opencode.nvim wires the OpenCode agent into Neovim so prompts, selections and buffer context go to an agent already running on your machine rather than to a separate service. It adds no credential of its own: the provider and key are whatever OpenCode is configured with, which is one fewer place a key has to be stored. MIT. Key handling: Your API key is supplied via an environment variable or local config file. Requests are sent directly to the AI provider. Scores: key handling 77/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 22/25: The key is supplied by an environment variable or local config file you control. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under MIT, so anyone can check how the key is handled. - Privacy 7/15: Nothing specific is known about its data handling beyond the key itself. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-24, about 0 month(s) ago. (https://github.com/nickjvandyke/opencode.nvim/commits) ### ProxyAI - Page: https://www.byoaik.com/tools/proxyai/ - Website: https://www.tryproxy.io/ - Source: https://github.com/carlrobertoh/ProxyAI - Category: IDE Plugins | Pricing: OPEN_SOURCE | Open source: yes (Apache-2.0) | Trust score: 76/100 - Platforms: JetBrains Plugin - Bring-your-own-key providers: OpenAI, Anthropic, Groq, Together AI, Ollama, Custom / OpenAI-compatible, Azure OpenAI - Key storage: LOCAL_DEVICE | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Slowing down (last commit about 5 months ago) ProxyAI is an open-source AI coding assistant for JetBrains IDEs including IntelliJ, PyCharm and WebStorm, previously called CodeGPT. Users paste their own key into plugin settings for OpenAI, or connect Anthropic, Azure OpenAI, Groq, Together, Ollama and other OpenAI-compatible providers, paying provider rates rather than the optional ProxyAI cloud tier. Offers chat, inline completion and codebase-aware features. Written in Kotlin, Apache-2.0, 1,927 stars, pushed about two and a half months before review. Key handling: Your API key is stored locally on your device. Requests are sent directly to the AI provider. Scores: key handling 78/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 23/25: The key is held on your own device. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under Apache-2.0, so anyone can check how the key is handled. - Privacy 7/15: Nothing specific is known about its data handling beyond the key itself. - Maintenance 8/10: Commits within the last six months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-05-28, about 4 month(s) ago. (https://github.com/carlrobertoh/ProxyAI/commits) ### prompttools - Page: https://www.byoaik.com/tools/prompttools/ - Website: https://github.com/hegelai/prompttools - Source: https://github.com/hegelai/prompttools - Category: Prompt Engineering Tools | Pricing: OPEN_SOURCE | Open source: yes (Apache-2.0) | Trust score: 76/100 - Platforms: Library/SDK, CLI, Self-hosted, Docker - Bring-your-own-key providers: OpenAI, Anthropic, Hugging Face, Ollama - Key storage: ENVIRONMENT_VARIABLE | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Stale (last commit about 9 months ago) prompttools provides notebook and CLI experiment harnesses for testing and benchmarking prompts across providers and vector databases. It runs entirely on your machine: the project states that any call to LLM APIs is executed directly from your machine without forwarding, and examples set the provider key directly or through environment variables. Distributed as a Python package. Apache-2.0, 3,047 stars, last commit about six months before review. Key handling: Your API key is supplied via an environment variable or local config file. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 83/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 22/25: The key is supplied by an environment variable or local config file you control. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under Apache-2.0, so anyone can check how the key is handled. - Privacy 11/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. Analytics or error-reporting libraries are present; what they send was not established. - Maintenance 5/10: Last commit was within the last year. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/hegelai/prompttools/blob/HEAD/README.md) - [REPORTED via SOURCE_SCAN] References Sentry, so some analytics or error reporting is present. This scan cannot tell whether it is opt-in or what it sends. (https://github.com/hegelai/prompttools/blob/HEAD/requirements.txt) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-02-11, about 8 month(s) ago. (https://github.com/hegelai/prompttools/commits) ### Summarize - Page: https://www.byoaik.com/tools/summarize/ - Website: https://github.com/steipete/summarize - Source: https://github.com/steipete/summarize - Category: Browser Extensions | Pricing: OPEN_SOURCE | Open source: yes (MIT) | Trust score: 75/100 - Platforms: Browser Extension, CLI, Self-hosted, Docker - Bring-your-own-key providers: OpenAI, Anthropic, Google Gemini, xAI Grok, OpenRouter, Ollama, Custom / OpenAI-compatible - Key storage: BROWSER_LOCAL_STORAGE | Key transmission: DIRECT_TO_PROVIDER | Risk: MEDIUM - Maintenance: Actively developed (last commit 3 days ago) Open-source (MIT) summarization tool by Peter Steinberger that extracts and summarizes content from web pages, PDFs, images, audio/video, YouTube, podcasts, and RSS. It ships as a Chrome Side Panel extension (with chat and history), a Firefox Sidebar, a CLI, and an optional local daemon for streaming Markdown output. It performs transcript-first media processing, video slide extraction with OCR, and timestamped cards. Users bring their own LLM provider API key: configuration is via environment variables (OPENAI_API_KEY, ANTHROPIC_API_KEY, GEMINI_API_KEY / GOOGLE_GENERATIVE_AI_API_KEY, XAI_API_KEY, OPENROUTER_API_KEY, plus OpenAI-compatible endpoints like Z.AI, NVIDIA, MiniMax) or a local config file at ~/.summarize/config.json. Ollama and other OpenAI-compatible local endpoints are supported for fully local use; if no key is set it can fall back to on-device Gemini Nano. Note: Groq/AssemblyAI/ElevenLabs/FAL keys are used only for audio transcription, not for the summarization LLM. Verified against the live GitHub API on 2026-06-20: 6,226 stars, MIT license, not archived, not a fork, latest release v0.20.0 published 2026-06-19, last pushed 2026-06-19. Key handling: Your API key is stored in your browser's local storage. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 73/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 13/25: The key is persisted in browser local storage, which any script on the page can read. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under MIT, so anyone can check how the key is handled. - Privacy 14/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/steipete/summarize/blob/HEAD/README.md) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-26, about 0 month(s) ago. (https://github.com/steipete/summarize/commits) ### Langfuse - Page: https://www.byoaik.com/tools/langfuse/ - Website: https://langfuse.com - Source: https://github.com/langfuse/langfuse - Category: Developer Infrastructure | Pricing: OPEN_SOURCE | Open source: yes (Mixed: MIT core, separate enterprise licence for ee/) | Trust score: 75/100 - Platforms: Self-hosted, Docker, Web, Library/SDK - Bring-your-own-key providers: OpenAI, Anthropic, Azure OpenAI, Ollama, Custom / OpenAI-compatible - Key storage: SELF_HOSTED_ONLY | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit today) Langfuse is an open-source LLM engineering platform combining prompt management, a prompt playground, tracing, evaluations and datasets. The self-hosting documentation describes the LLM connection for the playground and LLM-as-judge evaluations as optional and bring-your-own: you connect your own key for OpenAI, Anthropic, Azure OpenAI or any OpenAI-compatible endpoint including Ollama, and nothing is routed through a Langfuse-owned key. Deploys through Docker Compose, Kubernetes or Helm. Core code is MIT with enterprise features under a separate licence in the ee/ directory. 32,899 stars. Key handling: You self-host the app, so your API key lives in your own deployment. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 82/100 (assessable for any licence), verifiability 53/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 21/25: The key lives in a deployment you run yourself. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 15/20: Source is public, though the licence is unclear or non-standard. - Privacy 11/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. Analytics or error-reporting libraries are present; what they send was not established. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/langfuse/langfuse/blob/HEAD/README.md) - [REPORTED via SOURCE_SCAN] References PostHog, so some analytics or error reporting is present. This scan cannot tell whether it is opt-in or what it sends. (https://github.com/langfuse/langfuse/blob/HEAD/README.md) - [CONFIRMED via SOURCE_SCAN] Ships a container definition, so it can be self-hosted on your own infrastructure. (https://github.com/langfuse/langfuse/blob/HEAD/docker-compose.yml) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-29, about 0 month(s) ago. (https://github.com/langfuse/langfuse/commits) ### KISS Translator - Page: https://www.byoaik.com/tools/kiss-translator/ - Website: https://github.com/fishjar/kiss-translator - Source: https://github.com/fishjar/kiss-translator - Category: Productivity Tools | Pricing: OPEN_SOURCE | Open source: yes (GPL-3.0) | Trust score: 75/100 - Platforms: Browser Extension, Web, Self-hosted, Docker - Bring-your-own-key providers: OpenAI, Anthropic, Google Gemini, OpenRouter, DeepSeek, Ollama, Custom / OpenAI-compatible - Key storage: BROWSER_LOCAL_STORAGE | Key transmission: DIRECT_TO_PROVIDER | Risk: MEDIUM - Maintenance: Actively developed (last commit today) KISS Translator is a simple, open-source bilingual translation browser extension and userscript (Greasemonkey/Tampermonkey) for translating webpages, selected text, input boxes, and YouTube/video subtitles side-by-side. It supports a wide range of translation backends, including AI/LLM services where the end user supplies their own API credentials: OpenAI, Google Gemini, Anthropic Claude, Ollama, DeepSeek, and OpenRouter, alongside conventional engines like Google, Microsoft, DeepL/DeepLX, Azure AI, and Cloudflare AI. Users configure their own API keys and can point AI services at custom OpenAI-compatible base URLs/endpoints; a Hook mechanism lets it work with virtually any translation interface. It runs across Chrome, Edge, Firefox, Safari (Mac/iOS), Kiwi (Android), Orion, and Thunderbird, and offers cross-device sync via WebDAV or a self-hostable KISS-Worker. Released under GPL-3.0, actively maintained (v2.0.25, June 2026), ~10.9k GitHub stars. Key handling: Your API key is stored in your browser's local storage. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 73/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 13/25: The key is persisted in browser local storage, which any script on the page can read. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under GPL-3.0, so anyone can check how the key is handled. - Privacy 14/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/fishjar/kiss-translator/blob/HEAD/README.md) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-29, about 0 month(s) ago. (https://github.com/fishjar/kiss-translator/commits) ### Karakeep - Page: https://www.byoaik.com/tools/karakeep/ - Website: https://karakeep.app/ - Source: https://github.com/karakeep-app/karakeep - Category: Productivity Tools | Pricing: OPEN_SOURCE | Open source: yes (AGPL-3.0) | Trust score: 75/100 - Platforms: Self-hosted, Web, Mobile, Browser Extension, Docker - Bring-your-own-key providers: OpenAI, Azure OpenAI, Ollama, Custom / OpenAI-compatible - Key storage: BROWSER_LOCAL_STORAGE | Key transmission: DIRECT_TO_PROVIDER | Risk: MEDIUM - Maintenance: Actively developed (last commit today) Karakeep (formerly Hoarder) is an open-source, self-hostable "bookmark everything" app: save links, notes, images, and PDFs with automatic AI tagging, summarization, OCR, and Meilisearch-powered full-text search. Self-hosters bring their own AI provider by setting OPENAI_API_KEY (with optional OPENAI_BASE_URL for any OpenAI-compatible endpoint such as Azure OpenAI) or pointing OLLAMA_BASE_URL at a local Ollama instance, and can pick their own text/image/embedding models via INFERENCE_TEXT_MODEL, INFERENCE_IMAGE_MODEL, and EMBEDDING_TEXT_MODEL. Ships as a Docker/Kubernetes web app plus browser extensions (Chrome/Firefox/Safari) and iOS/Android apps. A hosted Karakeep Cloud (free tier, $4/mo Pro) also exists, but BYOK applies to the self-hosted deployment. Key handling: Your API key is stored in your browser's local storage. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 73/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 13/25: The key is persisted in browser local storage, which any script on the page can read. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under AGPL-3.0, so anyone can check how the key is handled. - Privacy 14/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/karakeep-app/karakeep/blob/HEAD/README.md) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-27, about 0 month(s) ago. (https://github.com/karakeep-app/karakeep/commits) ### Twenty CRM - Page: https://www.byoaik.com/tools/twenty-crm/ - Website: https://twenty.com/ - Source: https://github.com/twentyhq/twenty - Category: Enterprise AI Tools | Pricing: OPEN_SOURCE | Open source: yes (AGPLv3 (with commercial/enterprise license for enterprise-flagged files)) | Trust score: 75/100 - Platforms: Web, Self-hosted, Docker - Bring-your-own-key providers: OpenAI, Anthropic, Google Gemini, xAI Grok, Groq, Mistral, OpenRouter, Ollama, Custom / OpenAI-compatible - Key storage: SELF_HOSTED_ONLY | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit today) Twenty is an open-source CRM positioned as "the open alternative to Salesforce, designed for AI." Built with TypeScript, NestJS, React, and PostgreSQL, it offers customizable objects, workflow automation, and AI agents/chats for tasks like prospect research, record updates, and summaries. On self-hosted deployments, AI is fully BYOK: the server's .env.example exposes native API-key slots for OpenAI, Anthropic (Claude), Google, xAI (Grok), Groq, and Mistral, plus an AI_PROVIDERS JSON variable supporting custom "openai-compatible" providers with a configurable baseUrl and apiKey (covering OpenRouter, local/Ollama gateways, and any OpenAI-compatible endpoint). Default model configs reference current OpenAI GPT-5.x and Anthropic Claude Sonnet/Haiku 4.x models. The repo is very actively maintained (last push 2026-06-20, ~50.7k stars). Licensed under AGPLv3 with a separate commercial license for enterprise-flagged files. A paid hosted cloud is also offered. Key handling: You self-host the app, so your API key lives in your own deployment. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 82/100 (assessable for any licence), verifiability 53/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 21/25: The key lives in a deployment you run yourself. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 15/20: Source is public, though the licence is unclear or non-standard. - Privacy 11/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. Analytics or error-reporting libraries are present; what they send was not established. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [REPORTED via SOURCE_SCAN] References Sentry, so some analytics or error reporting is present. This scan cannot tell whether it is opt-in or what it sends. (https://github.com/twentyhq/twenty/blob/HEAD/package.json) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-29, about 0 month(s) ago. (https://github.com/twentyhq/twenty/commits) ### claudecode.nvim - Page: https://www.byoaik.com/tools/claudecode-nvim/ - Website: https://github.com/coder/claudecode.nvim - Source: https://github.com/coder/claudecode.nvim - Category: IDE Plugins | Pricing: OPEN_SOURCE | Open source: yes (MIT) | Trust score: 75/100 - Platforms: CLI, Local-only - Bring-your-own-key providers: Anthropic - Key storage: ENVIRONMENT_VARIABLE | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit 6 days ago) claudecode.nvim implements Claude Code's websocket protocol natively in Lua, giving Neovim the selection tracking and diff review that the official editor extensions have. Authentication stays with the Claude Code CLI you have already set up, so the plugin never holds a credential itself. Maintained by Coder. MIT. Key handling: Your API key is supplied via an environment variable or local config file. Requests are sent directly to the AI provider. Scores: key handling 77/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 22/25: The key is supplied by an environment variable or local config file you control. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under MIT, so anyone can check how the key is handled. - Privacy 7/15: Nothing specific is known about its data handling beyond the key itself. - Maintenance 8/10: Commits within the last six months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-06-25, about 3 month(s) ago. (https://github.com/coder/claudecode.nvim/commits) ### LobeChat - Page: https://www.byoaik.com/tools/lobechat/ - Website: https://lobehub.com/ - Source: https://github.com/lobehub/lobehub - Category: AI Chat Interfaces | Pricing: OPEN_SOURCE | Open source: yes (LobeHub Community / Apache-2.0-based) | Trust score: 74/100 - Platforms: Web, Desktop, Mobile, Self-hosted, Docker - Bring-your-own-key providers: OpenAI, Anthropic, Google Gemini, Azure OpenAI, OpenRouter, Groq, Mistral, Cohere, Perplexity, DeepSeek, xAI Grok, Ollama, Together AI, Fireworks, Custom / OpenAI-compatible - Key storage: SERVER_SIDE_ENCRYPTED | Key transmission: DIRECT_TO_PROVIDER | Risk: MEDIUM - Maintenance: Actively developed (last commit today) LobeChat (LobeHub) is an open-source, modern-design AI chat framework available as a web PWA, desktop app, and self-hosted Docker deployment. Users add their own API keys for 40+ AI providers under Settings > Model Provider (keys are encrypted at rest via KEY_VAULTS_SECRET), or configure them through environment variables. It supports custom OpenAI-compatible endpoints and proxy URLs, local models via Ollama, a knowledge base with RAG, plugins/artifacts, TTS/STT, and vision. The project is very actively maintained (v2.2.7 released June 2026) with ~79k GitHub stars. Key handling: Your API key is stored encrypted on the server instance you control. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 70/100 (assessable for any licence), verifiability 73/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 11/25: The key is stored server-side, encrypted, on an instance you control. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 15/20: Source is public, though the licence is unclear or non-standard. Key handling was located in the published source. - Privacy 14/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 7/10: Key handling was found in the published source by BYOAIK's scanner. What was checked (verification tier SOURCE_VERIFIED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [STRONG via SOURCE_SCAN] The user supplies their own OpenAI key: the project references its API key variable. (https://github.com/lobehub/lobehub/blob/HEAD/README.md) - [STRONG via SOURCE_SCAN] The user supplies their own Anthropic key: the project references its API key variable. (https://github.com/lobehub/lobehub/blob/HEAD/.env.example) - [STRONG via SOURCE_SCAN] The user supplies their own OpenRouter key: the project references its API key variable. (https://github.com/lobehub/lobehub/blob/HEAD/.env.example) - [STRONG via SOURCE_SCAN] The user supplies their own Groq key: the project references its API key variable. (https://github.com/lobehub/lobehub/blob/HEAD/.env.example) - [STRONG via SOURCE_SCAN] The user supplies their own Mistral key: the project references its API key variable. (https://github.com/lobehub/lobehub/blob/HEAD/.env.example) - [STRONG via SOURCE_SCAN] The user supplies their own DeepSeek key: the project references its API key variable. (https://github.com/lobehub/lobehub/blob/HEAD/.env.example) - [STRONG via SOURCE_SCAN] The user supplies their own Perplexity key: the project references its API key variable. (https://github.com/lobehub/lobehub/blob/HEAD/.env.example) - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/lobehub/lobehub/blob/HEAD/.env.example) - [STRONG via SOURCE_SCAN] The user supplies their own Cohere key: the project references its API key variable. (https://github.com/lobehub/lobehub/blob/HEAD/Dockerfile) - [STRONG via SOURCE_SCAN] The user supplies their own xAI Grok key: the project references its API key variable. (https://github.com/lobehub/lobehub/blob/HEAD/Dockerfile) - [STRONG via SOURCE_SCAN] The user supplies their own Hugging Face key: the project references its API key variable. (https://github.com/lobehub/lobehub/blob/HEAD/Dockerfile) - [CONFIRMED via SOURCE_SCAN] Ships a container definition, so it can be self-hosted on your own infrastructure. (https://github.com/lobehub/lobehub/blob/HEAD/Dockerfile) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-29, about 0 month(s) ago. (https://github.com/lobehub/lobehub/commits) ### TypingMind - Page: https://www.byoaik.com/tools/typingmind/ - Website: https://www.typingmind.com/ - Source: https://github.com/TypingMind/typingmind - Category: AI Chat Interfaces | Pricing: ONE_TIME_PURCHASE | Open source: no | Trust score: 74/100 - Platforms: Web, Desktop, Self-hosted, Docker - Bring-your-own-key providers: OpenAI, Anthropic, Google Gemini, Azure OpenAI, OpenRouter, Groq, Mistral, Perplexity, DeepSeek, xAI Grok, Ollama, Hugging Face, Fireworks, Custom / OpenAI-compatible - Key storage: LOCAL_DEVICE | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit 29 days ago) TypingMind is a closed-source, one-time-license web/desktop chat frontend for LLMs. Users bring their own provider API keys (OpenAI, Anthropic, Google Gemini, OpenRouter, and many more) which are stored locally in the browser and never sent to TypingMind's servers; the app calls the provider API directly and the provider bills the user per token. It adds power-user features on top: AI agents/characters, plugins, multi-model side-by-side chats, projects/folders, web search, vision, document uploads, voice input, and artifacts. Any OpenAI-compatible endpoint or proxy can be added via a custom base URL, including local runtimes like Ollama, LM Studio, and LocalAI. It runs in any browser, installs as a PWA / Mac app, and licensed users get a static self-host package. Pricing is one-time: Standard $39, Extended $79, Premium $99 (discounted from $198), plus a $395 bulk license for up to 10 users; a separate TypingMind for Teams product offers a hosted admin portal at custom pricing. Key handling: Your API key is stored locally on your device. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 90/100 (assessable for any licence), verifiability 33/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 23/25: The key is held on your own device. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 9/20: A public repository exists but the project is not open source. - Privacy 14/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/TypingMind/typingmind/blob/HEAD/README.md) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-03, about 1 month(s) ago. (https://github.com/TypingMind/typingmind/commits) ### ChatALL - Page: https://www.byoaik.com/tools/chatall/ - Website: https://chatall.ai - Source: https://github.com/ai-shifu/ChatALL - Category: AI Chat Interfaces | Pricing: OPEN_SOURCE | Open source: yes (Apache-2.0) | Trust score: 74/100 - Platforms: Desktop - Bring-your-own-key providers: OpenAI, Anthropic, Google Gemini, Custom / OpenAI-compatible, Ollama - Key storage: LOCAL_DEVICE | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Dormant (last commit about 13 months ago) ChatALL is a desktop client for Windows, macOS and Linux that sends one prompt to many AI services concurrently and shows the answers side by side, so you can compare models before committing. It connects directly to each provider using credentials you supply; the project describes itself as 'a client, not a proxy', meaning conversations go straight from your machine to each vendor with your own account or API token. Supports OpenAI, Anthropic, Google and many others plus local and self-hosted backends. Apache-2.0 licensed with 16,000+ stars. Key handling: Your API key is stored locally on your device. Requests are sent directly to the AI provider. Scores: key handling 85/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 23/25: The key is held on your own device. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under Apache-2.0, so anyone can check how the key is handled. - Privacy 11/15: Local-first: it works without sending your data anywhere. - Maintenance 2/10: Last commit was about 12 months ago. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [CONFIRMED via SOURCE_SCAN] Most recent commit 2025-09-25, about 12 month(s) ago. (https://github.com/ai-shifu/ChatALL/commits) ### Budibase - Page: https://www.byoaik.com/tools/budibase/ - Website: https://budibase.com - Source: https://github.com/budibase/budibase - Category: App Builders | Pricing: OPEN_SOURCE | Open source: yes (Mixed: AGPL-3.0/GPL-3.0/MPL-2.0 components, BSL on some AI features) | Trust score: 74/100 - Platforms: Web, Self-hosted, Docker - Bring-your-own-key providers: OpenAI, Anthropic, Google Gemini, Mistral, OpenRouter, Groq - Key storage: SELF_HOSTED_ONLY | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit 3 days ago) Budibase is an open-source low-code platform for building internal tools, workflows and AI agents, fully self-hostable. Under workspace connection settings you add credentials for the provider you want, OpenAI, Anthropic, Google, Mistral, OpenRouter or Groq, before any AI feature or agent can run, so every call uses the key you supplied rather than a pooled one. Self-hostable through Docker or Kubernetes. Licensing is split across the builder, server and client components, with some agent features under a Business Source License. 28,202 stars, last commit within a day of review. Key handling: You self-host the app, so your API key lives in your own deployment. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 80/100 (assessable for any licence), verifiability 53/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 21/25: The key lives in a deployment you run yourself. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 15/20: Source is public, though the licence is unclear or non-standard. - Privacy 10/15: Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-29, about 0 month(s) ago. (https://github.com/budibase/budibase/commits) ### ChatTutor - Page: https://www.byoaik.com/tools/chattutor/ - Website: https://github.com/HugeCatLab/ChatTutor - Source: https://github.com/HugeCatLab/ChatTutor - Category: Education Tools | Pricing: OPEN_SOURCE | Open source: yes (AGPL-3.0) | Trust score: 74/100 - Platforms: Web, Self-hosted, Docker - Bring-your-own-key providers: OpenAI, Anthropic, DeepSeek - Key storage: SELF_HOSTED_ONLY | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Stale (last commit about 10 months ago) ChatTutor is an open-source AI tutoring application pairing conversational AI with teaching tools including a maths and drawing canvas and mind maps, for one-to-one study sessions. Self-hosted deployments set a model API key environment variable and a provider selector for OpenAI, Anthropic or DeepSeek, and the hosted instance asks users to set their own key and models in settings. Self-hostable web app. AGPL-3.0, 1,306 stars, last commit about seven months before review. Key handling: You self-host the app, so your API key lives in your own deployment. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 80/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 21/25: The key lives in a deployment you run yourself. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under AGPL-3.0, so anyone can check how the key is handled. - Privacy 10/15: Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 5/10: Last commit was within the last year. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-01-09, about 9 month(s) ago. (https://github.com/HugeCatLab/ChatTutor/commits) ### Automatisch - Page: https://www.byoaik.com/tools/automatisch/ - Website: https://automatisch.io - Source: https://github.com/automatisch/automatisch - Category: Automation Tools | Pricing: OPEN_SOURCE | Open source: yes (AGPL-3.0) | Trust score: 74/100 - Platforms: Web, Self-hosted, Docker - Bring-your-own-key providers: OpenAI - Key storage: SELF_HOSTED_ONLY | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Stale (last commit about 10 months ago) Automatisch is a self-hosted, open-source workflow automation platform in the mould of Zapier, with OpenAI available as one of its integrations. To use the OpenAI step you create your own secret key and paste it into the connection's API key field inside Automatisch. There are no bundled AI credits. Runs as a self-hosted Node and Docker application with a visual workflow builder. AGPL-3.0, 13,932 stars, last commit about seven months before review. Key handling: You self-host the app, so your API key lives in your own deployment. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 80/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 21/25: The key lives in a deployment you run yourself. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under AGPL-3.0, so anyone can check how the key is handled. - Privacy 10/15: Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 5/10: Last commit was within the last year. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [CONFIRMED via SOURCE_SCAN] Ships a container definition, so it can be self-hosted on your own infrastructure. (https://github.com/automatisch/automatisch/blob/HEAD/docker-compose.yml) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-01-15, about 8 month(s) ago. (https://github.com/automatisch/automatisch/commits) ### stagewise - Page: https://www.byoaik.com/tools/stagewise/ - Website: https://stagewise.io/ - Source: https://github.com/stagewise-io/stagewise - Category: Coding Assistants | Pricing: OPEN_SOURCE | Open source: yes (AGPL-3.0) | Trust score: 74/100 - Platforms: CLI, Local-only, VS Code Extension - Bring-your-own-key providers: OpenAI, Anthropic, Google Gemini, Custom / OpenAI-compatible - Key storage: ENVIRONMENT_VARIABLE | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit 4 days ago) stagewise connects a coding agent to the running frontend, so instead of describing a UI problem you select the element and the agent sees the actual DOM, styles and React context around it. It runs locally beside the dev server on the model provider you configure with your own key. AGPL-3.0, which is worth noting if you intend to embed it rather than use it. Key handling: Your API key is supplied via an environment variable or local config file. Requests are sent directly to the AI provider. Scores: key handling 72/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 22/25: The key is supplied by an environment variable or local config file you control. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under AGPL-3.0, so anyone can check how the key is handled. - Privacy 4/15: Analytics or error-reporting libraries are present; what they send was not established. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [REPORTED via SOURCE_SCAN] References PostHog, so some analytics or error reporting is present. This scan cannot tell whether it is opt-in or what it sends. (https://github.com/stagewise-io/stagewise/blob/HEAD/.env.example) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-23, about 0 month(s) ago. (https://github.com/stagewise-io/stagewise/commits) ### AFFiNE - Page: https://www.byoaik.com/tools/affine/ - Website: https://affine.pro - Source: https://github.com/toeverything/AFFiNE - Category: Writing Tools | Pricing: OPEN_SOURCE | Open source: yes (Mixed: see repository LICENSE, no single SPDX tag) | Trust score: 73/100 - Platforms: Desktop, Web, Self-hosted, Docker - Bring-your-own-key providers: OpenAI, Anthropic, Google Gemini - Key storage: LOCAL_DEVICE | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit today) AFFiNE is an open-source, privacy-first workspace combining documents, whiteboard and database views, positioned as a Notion and Miro alternative. On a self-hosted instance you enable copilot with byok in config.json, then add a key under Settings, Integrations, AI BYOK, choosing a provider and entering its key; AFFiNE documents BYOK routes for OpenAI, Anthropic, Gemini and FAL. Note this path exists only on self-hosted deployments. The hosted AFFiNE Cloud AI add-on is a separate paid subscription. Desktop, web and Docker. 71,445 stars with daily commits; the repository reports no single SPDX licence. Key handling: Your API key is stored locally on your device. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 78/100 (assessable for any licence), verifiability 53/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 23/25: The key is held on your own device. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 15/20: Source is public, though the licence is unclear or non-standard. - Privacy 7/15: Can be self-hosted, so the data path stays inside infrastructure you control. Analytics or error-reporting libraries are present; what they send was not established. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via SOURCE_SCAN] References Sentry, so some analytics or error reporting is present. This scan cannot tell whether it is opt-in or what it sends. (https://github.com/toeverything/AFFiNE/blob/HEAD/package.json) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-29, about 0 month(s) ago. (https://github.com/toeverything/AFFiNE/commits) ### Flowise - Page: https://www.byoaik.com/tools/flowise/ - Website: https://flowiseai.com/ - Source: https://github.com/FlowiseAI/Flowise - Category: Agent Frameworks | Pricing: OPEN_SOURCE | Open source: yes (Apache-2.0) | Trust score: 73/100 - Platforms: Web, Self-hosted, Library/SDK, Docker - Bring-your-own-key providers: OpenAI, Anthropic, Google Gemini, Azure OpenAI, Mistral, Cohere, Groq, Together AI, Fireworks, Hugging Face, Ollama, Custom / OpenAI-compatible - Key storage: SELF_HOSTED_ONLY | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Maintained (last commit about 3 months ago) Flowise is an open-source (Apache 2.0), self-hostable low-code platform for visually building LLM apps, chatflows, and multi-agent systems on top of LangChain/LlamaIndex. Users drag model nodes onto a canvas and connect their own provider credentials: the ChatOpenAI/Azure/Anthropic/etc. nodes use a "Connect Credential" flow where you paste your own API key, and the ChatLocalAI node exposes a configurable Base Path that points at any OpenAI-compatible endpoint (e.g. http://localhost:8080/v1 or Ollama). It ships nodes for 15+ chat-model providers plus a JS/REST SDK for embedding flows. A managed Flowise Cloud option also exists, but the self-hosted edition is fully BYOK and free. Key handling: You self-host the app, so your API key lives in your own deployment. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 87/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 21/25: The key lives in a deployment you run yourself. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under Apache-2.0, so anyone can check how the key is handled. - Privacy 14/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 0/10: The repository is archived and no longer maintained. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [CONFIRMED via SOURCE_SCAN] Ships a container definition, so it can be self-hosted on your own infrastructure. (https://github.com/FlowiseAI/Flowise/blob/HEAD/Dockerfile) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-08-13, about 2 month(s) ago. (https://github.com/FlowiseAI/Flowise/commits) ### Copilot for Xcode - Page: https://www.byoaik.com/tools/copilot-for-xcode/ - Website: https://github.com/intitni/CopilotForXcode - Source: https://github.com/intitni/CopilotForXcode - Category: IDE Plugins | Pricing: OPEN_SOURCE | Open source: yes (MIT) | Trust score: 73/100 - Platforms: Desktop - Bring-your-own-key providers: OpenAI, Custom / OpenAI-compatible - Key storage: LOCAL_DEVICE | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit 15 days ago) An open-source macOS Xcode Source Editor Extension, unaffiliated with GitHub's official Copilot for Xcode, adding suggestion and chat panels backed by GitHub Copilot, Codeium or ChatGPT. For the ChatGPT path the user configures their own OpenAI key under Service, Chat Model in the host app rather than routing through a paid proxy; the Copilot and Codeium modes use those vendors' own authentication instead. MIT licensed, distributed free through GitHub releases. 8,234 stars, pushed about four months before review. Key handling: Your API key is stored locally on your device. Requests are sent directly to the AI provider. Scores: key handling 78/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 23/25: The key is held on your own device. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under MIT, so anyone can check how the key is handled. - Privacy 7/15: Nothing specific is known about its data handling beyond the key itself. - Maintenance 5/10: Last commit was within the last year. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-02-05, about 8 month(s) ago. (https://github.com/intitni/CopilotForXcode/commits) ### Nanobrowser - Page: https://www.byoaik.com/tools/nanobrowser/ - Website: https://www.nanobrowser.ai/ - Source: https://github.com/nanobrowser/nanobrowser - Category: Browser Extensions | Pricing: OPEN_SOURCE | Open source: yes (Apache-2.0) | Trust score: 72/100 - Platforms: Browser Extension, Self-hosted, Docker - Bring-your-own-key providers: OpenAI, Anthropic, Google Gemini, Groq, Ollama, Custom / OpenAI-compatible - Key storage: BROWSER_LOCAL_STORAGE | Key transmission: DIRECT_TO_PROVIDER | Risk: MEDIUM - Maintenance: Actively developed (last commit today) Nanobrowser is an open-source (Apache-2.0) Chrome/Edge browser extension that performs AI-powered web automation through a multi-agent system (planner, navigator, validator agents) operating directly in the browser via an interactive side panel. It positions itself as a free, self-hosted alternative to OpenAI Operator. Users bring their own LLM API keys; everything runs locally and credentials never leave the browser or touch any cloud service. The project is actively maintained (latest release v0.1.13, Nov 2025) with broad provider support and is downloadable from the Chrome Web Store. Key handling: Your API key is stored in your browser's local storage. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 68/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 13/25: The key is persisted in browser local storage, which any script on the page can read. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under Apache-2.0, so anyone can check how the key is handled. - Privacy 11/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. Analytics or error-reporting libraries are present; what they send was not established. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/nanobrowser/nanobrowser/blob/HEAD/README.md) - [REPORTED via SOURCE_SCAN] References PostHog, so some analytics or error reporting is present. This scan cannot tell whether it is opt-in or what it sends. (https://github.com/nanobrowser/nanobrowser/blob/HEAD/.env.example) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-08-18, about 1 month(s) ago. (https://github.com/nanobrowser/nanobrowser/commits) ### Pot (pot-desktop) - Page: https://www.byoaik.com/tools/pot-pot-desktop/ - Website: https://pot-app.com/ - Source: https://github.com/pot-app/pot-desktop - Category: Productivity Tools | Pricing: OPEN_SOURCE | Open source: yes (GPL-3.0) | Trust score: 72/100 - Platforms: Desktop - Bring-your-own-key providers: OpenAI, Google Gemini, Azure OpenAI, Ollama, DeepSeek, OpenRouter, Custom / OpenAI-compatible - Key storage: LOCAL_DEVICE | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Slowing down (last commit about 4 months ago) Pot is an open-source (GPL-3.0) cross-platform desktop application for text translation, OCR/screenshot translation, text-to-speech, and vocabulary collection. It runs on Windows, macOS, and Linux. Users bring their own credentials: each cloud service (OpenAI, Gemini Pro, DeepL, Zhipu/ChatGLM, Baidu, Tencent, Volcengine, Youdao, Yandex, and more) is configured in Service Settings with the user's own API key, and v3.0+ allows multiple instances of the same service with different keys/configs. The built-in OpenAI service reads a user-supplied requestPath (custom domain/base URL), apiKey, and model, and supports both OpenAI and Azure modes, so it can target any OpenAI-compatible endpoint (e.g. DeepSeek, OpenRouter, local servers). Ollama is supported for fully local/offline models. An extensible .potext plugin system lets users add additional translate/OCR/TTS providers with their own API key and parameters. Actively maintained (latest pushes June 2026; 3.0.7 release), ~18.8k GitHub stars. Key handling: Your API key is stored locally on your device. Requests are sent directly to the AI provider. Scores: key handling 85/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 23/25: The key is held on your own device. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under GPL-3.0, so anyone can check how the key is handled. - Privacy 11/15: Local-first: it works without sending your data anywhere. - Maintenance 0/10: The repository is archived and no longer maintained. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/pot-app/pot-desktop/blob/HEAD/README.md) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-07-04, about 3 month(s) ago. (https://github.com/pot-app/pot-desktop/commits) ### org-ai - Page: https://www.byoaik.com/tools/org-ai/ - Website: https://github.com/rksm/org-ai - Source: https://github.com/rksm/org-ai - Category: Writing Tools | Pricing: OPEN_SOURCE | Open source: yes (GPL-3.0) | Trust score: 72/100 - Platforms: Desktop - Bring-your-own-key providers: OpenAI, Anthropic, Perplexity, Azure OpenAI, Custom / OpenAI-compatible - Key storage: ENVIRONMENT_VARIABLE | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Stale (last commit about 10 months ago) org-ai is an Emacs package embedding AI text and image generation directly into org-mode buffers through special begin_ai blocks. The key is set with org-ai-openai-api-token or stored through auth-source and authinfo.gpg, and users can switch per block or globally between OpenAI, Azure OpenAI, Anthropic Claude, Perplexity, Stable Diffusion via a local webui, and local models through text-generation-webui. GPL-3.0, distributed via MELPA, 821 stars, pushed about seven months before review. The slowest-moving entry in this batch but still inside the freshness window. Key handling: Your API key is supplied via an environment variable or local config file. Requests are sent directly to the AI provider. Scores: key handling 77/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 22/25: The key is supplied by an environment variable or local config file you control. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under GPL-3.0, so anyone can check how the key is handled. - Privacy 7/15: Nothing specific is known about its data handling beyond the key itself. - Maintenance 5/10: Last commit was within the last year. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-01-07, about 9 month(s) ago. (https://github.com/rksm/org-ai/commits) ### Arize Phoenix - Page: https://www.byoaik.com/tools/arize-phoenix/ - Website: https://phoenix.arize.com/ - Source: https://github.com/Arize-ai/phoenix - Category: Developer Infrastructure | Pricing: FREE | Open source: no | Trust score: 72/100 - Platforms: Self-hosted, Docker, Web - Bring-your-own-key providers: OpenAI, Anthropic, Google Gemini, OpenRouter, Ollama, Custom / OpenAI-compatible - Key storage: SELF_HOSTED_ONLY | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit yesterday) Phoenix gives OpenTelemetry-based tracing, evaluation and experiment tracking for LLM applications, and runs entirely on your own infrastructure, which keeps traced prompts and outputs in your hands. Read the licence before adopting it: GitHub reports the repository as unlicensed, and the file is the Elastic License 2.0, which is source-available rather than open source and specifically restricts offering the software to others as a managed service. For internal use that restriction rarely bites, but it is not the same grant as the Apache and MIT projects alongside it. Key handling: You self-host the app, so your API key lives in your own deployment. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 87/100 (assessable for any licence), verifiability 33/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 21/25: The key lives in a deployment you run yourself. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 9/20: A public repository exists but the project is not open source. - Privacy 14/15: Local-first: it works without sending your data anywhere. Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/Arize-ai/phoenix/blob/HEAD/README.md) - [CONFIRMED via SOURCE_SCAN] Ships a container definition, so it can be self-hosted on your own infrastructure. (https://github.com/Arize-ai/phoenix/blob/HEAD/docker-compose.yml) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-29, about 0 month(s) ago. (https://github.com/Arize-ai/phoenix/commits) ### Open Generative AI - Page: https://www.byoaik.com/tools/open-generative-ai/ - Website: https://github.com/Anil-matcha/Open-Generative-AI - Source: https://github.com/Anil-matcha/Open-Generative-AI - Category: Image Generation Tools | Pricing: OPEN_SOURCE | Open source: yes (MIT) | Trust score: 71/100 - Platforms: Web, Self-hosted, Local-only, Docker - Bring-your-own-key providers: Custom / OpenAI-compatible - Key storage: BROWSER_LOCAL_STORAGE | Key transmission: DIRECT_TO_PROVIDER | Risk: MEDIUM - Maintenance: Actively developed (last commit today) Open Generative AI is a self-hosted studio for image and video generation spanning a large catalogue of models. Your API key is entered in the browser and kept in localStorage, and the README states it is never sent to any server other than the upstream model service; the source names the pattern BYOK directly. Local models can be selected instead, which the README notes needs no API key at all. History persists in the browser rather than on a server. MIT licensed and very widely starred. Key handling: Your API key is stored in your browser's local storage. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 67/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 13/25: The key is persisted in browser local storage, which any script on the page can read. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under MIT, so anyone can check how the key is handled. - Privacy 10/15: Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/Anil-matcha/Open-Generative-AI/blob/HEAD/README.md) - [CONFIRMED via SOURCE_SCAN] Ships a container definition, so it can be self-hosted on your own infrastructure. (https://github.com/Anil-matcha/Open-Generative-AI/blob/HEAD/docker-compose.yml) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-28, about 0 month(s) ago. (https://github.com/Anil-matcha/Open-Generative-AI/commits) ### Chatwoot - Page: https://www.byoaik.com/tools/chatwoot/ - Website: https://www.chatwoot.com - Source: https://github.com/chatwoot/chatwoot - Category: Enterprise AI Tools | Pricing: OPEN_SOURCE | Open source: yes (Mixed: MIT core, separate commercial licence for enterprise directory) | Trust score: 71/100 - Platforms: Web, Self-hosted, Docker - Bring-your-own-key providers: OpenAI, Custom / OpenAI-compatible - Key storage: SELF_HOSTED_ONLY | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit today) Chatwoot is an open-source customer engagement platform covering live chat, email and omnichannel helpdesk, with an AI agent feature called Captain. On self-hosted installs an administrator enters their own OpenAI key, or points it at an OpenAI-compatible self-hosted endpoint, in the Captain settings or through an environment variable. The core is MIT licensed with an enterprise directory under separate commercial terms; self-hostable via Docker. 35,894 stars, last commit within a day of review. Key handling: You self-host the app, so your API key lives in your own deployment. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 75/100 (assessable for any licence), verifiability 53/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 21/25: The key lives in a deployment you run yourself. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 15/20: Source is public, though the licence is unclear or non-standard. - Privacy 7/15: Can be self-hosted, so the data path stays inside infrastructure you control. Analytics or error-reporting libraries are present; what they send was not established. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [CONFIRMED via SOURCE_SCAN] Ships a container definition, so it can be self-hosted on your own infrastructure. (https://github.com/chatwoot/chatwoot/blob/HEAD/docker-compose.yaml) - [REPORTED via SOURCE_SCAN] References Sentry, so some analytics or error reporting is present. This scan cannot tell whether it is opt-in or what it sends. (https://github.com/chatwoot/chatwoot/blob/HEAD/package.json) - [REPORTED via SOURCE_SCAN] References Amplitude, so some analytics or error reporting is present. This scan cannot tell whether it is opt-in or what it sends. (https://github.com/chatwoot/chatwoot/blob/HEAD/package.json) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-29, about 0 month(s) ago. (https://github.com/chatwoot/chatwoot/commits) ### pdfGPT - Page: https://www.byoaik.com/tools/pdfgpt/ - Website: https://github.com/bhaskatripathi/pdfGPT - Source: https://github.com/bhaskatripathi/pdfGPT - Category: Document Analysis | Pricing: OPEN_SOURCE | Open source: yes (MIT) | Trust score: 71/100 - Platforms: Self-hosted, Docker, Web - Bring-your-own-key providers: OpenAI - Key storage: SELF_HOSTED_ONLY | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Dormant (last commit about 20 months ago) pdfGPT turns a PDF, uploaded or fetched by URL, into a chatbot: it chunks the text, embeds the chunks with a Universal Sentence Encoder rather than a third-party vector service, retrieves nearest matches and asks an OpenAI model to answer with the page number cited. The interface begins with an 'Enter API Key' step, so each user supplies their own OpenAI key rather than the app pooling one. Ships as a self-hostable app with a Docker path. MIT licensed, 7,168 stars, last pushed about five months before review. Key handling: You self-host the app, so your API key lives in your own deployment. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 80/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 21/25: The key lives in a deployment you run yourself. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under MIT, so anyone can check how the key is handled. - Privacy 10/15: Can be self-hosted, so the data path stays inside infrastructure you control. - Maintenance 2/10: Last commit was about 19 months ago. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [CONFIRMED via SOURCE_SCAN] Ships a container definition, so it can be self-hosted on your own infrastructure. (https://github.com/bhaskatripathi/pdfGPT/blob/HEAD/docker-compose.yaml) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2025-03-03, about 19 month(s) ago. (https://github.com/bhaskatripathi/pdfGPT/commits) ### InSidebar AI - Page: https://www.byoaik.com/tools/insidebar-ai/ - Website: https://github.com/xiaolai/insidebar-ai - Source: https://github.com/xiaolai/insidebar-ai - Category: Browser Extensions | Pricing: OPEN_SOURCE | Open source: yes (MIT) | Trust score: 70/100 - Platforms: Browser Extension - Bring-your-own-key providers: OpenAI, Anthropic, Google Gemini, Custom / OpenAI-compatible - Key storage: BROWSER_LOCAL_STORAGE | Key transmission: DIRECT_TO_PROVIDER | Risk: MEDIUM - Maintenance: Actively developed (last commit 9 days ago) InSidebar AI puts an assistant in the browser side panel alongside the page rather than in a separate tab. Requests go from your browser straight to the provider on a key held in browser storage, so no intermediate server sees the traffic. That also means the key sits in browser storage rather than an OS keychain, which is weaker than an environment variable and worth knowing. MIT. Key handling: Your API key is stored in your browser's local storage. Requests are sent directly to the AI provider. Scores: key handling 65/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 13/25: The key is persisted in browser local storage, which any script on the page can read. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under MIT, so anyone can check how the key is handled. - Privacy 9/15: The project states it sends no telemetry. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/xiaolai/insidebar-ai/blob/HEAD/README.md) - [REPORTED via SOURCE_SCAN] The project states it sends no telemetry. (https://github.com/xiaolai/insidebar-ai/blob/HEAD/README.md) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-21, about 0 month(s) ago. (https://github.com/xiaolai/insidebar-ai/commits) ### NextAI Translator - Page: https://www.byoaik.com/tools/nextai-translator/ - Website: https://github.com/nextai-translator/nextai-translator - Source: https://github.com/nextai-translator/nextai-translator - Category: Productivity Tools | Pricing: OPEN_SOURCE | Open source: yes (AGPL-3.0) | Trust score: 69/100 - Platforms: Browser Extension, Desktop - Bring-your-own-key providers: OpenAI, Anthropic, Google Gemini, Azure OpenAI, Groq, Cohere, DeepSeek, Ollama, Custom / OpenAI-compatible - Key storage: BROWSER_LOCAL_STORAGE | Key transmission: DIRECT_TO_PROVIDER | Risk: MEDIUM - Maintenance: Actively developed (last commit 6 days ago) NextAI Translator (formerly OpenAI Translator, renamed after an OpenAI trademark warning) is an open-source, cross-platform translation tool that started as a Bob/Chrome plugin and grew into a browser extension (Chrome, Firefox) plus desktop apps for Windows, macOS, and Linux. It offers three modes (translate, polish, summarize) across 55 languages, with streaming responses, text-to-speech, screenshot/OCR translation, vocabulary book, and clip-extension integrations. Users bring their own provider API keys, configured in the settings UI. The repo ships dedicated engine implementations for a wide range of providers and confirmed OpenAI-compatible custom-endpoint support (used for local servers like Ollama and LM Studio, and providers such as MiniMax). It is AGPL-3.0 licensed, has ~24.9k GitHub stars, and is actively maintained (v0.6.20 released June 2026). Key handling: Your API key is stored in your browser's local storage. Requests are sent directly to the AI provider. Scores: key handling 63/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 13/25: The key is persisted in browser local storage, which any script on the page can read. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under AGPL-3.0, so anyone can check how the key is handled. - Privacy 8/15: Local-first: it works without sending your data anywhere. Analytics or error-reporting libraries are present; what they send was not established. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [REPORTED via SOURCE_SCAN] References Sentry, so some analytics or error reporting is present. This scan cannot tell whether it is opt-in or what it sends. (https://github.com/nextai-translator/nextai-translator/blob/HEAD/package.json) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-03, about 1 month(s) ago. (https://github.com/nextai-translator/nextai-translator/commits) ### Page Assist - Page: https://www.byoaik.com/tools/page-assist/ - Website: https://docs.pageassist.xyz/ - Source: https://github.com/n4ze3m/page-assist - Category: Browser Extensions | Pricing: OPEN_SOURCE | Open source: yes (MIT) | Trust score: 68/100 - Platforms: Browser Extension, Web - Bring-your-own-key providers: OpenAI, Anthropic, Google Gemini, OpenRouter, Groq, Mistral, DeepSeek, xAI Grok, Ollama, Hugging Face, Together AI, Fireworks, Custom / OpenAI-compatible - Key storage: BROWSER_LOCAL_STORAGE | Key transmission: DIRECT_TO_PROVIDER | Risk: MEDIUM - Maintenance: Actively developed (last commit today) Page Assist is an open-source (MIT) browser extension that adds a sidebar and full web UI for interacting with AI models from any webpage, including web-page-aware chat and document/RAG-style retrieval. It targets local models first (Ollama, LM Studio, llama.cpp, Llamafile, vLLM, Chrome AI/Gemini Nano) but also supports any OpenAI-compatible endpoint. Users add providers via Settings to an "OpenAI Compatible API" tab, choosing from a large dropdown (OpenAI, Anthropic/Claude, Google AI, Groq, Mistral, DeepSeek, OpenRouter, Together, Fireworks, xAI, Hugging Face, and many more) or "Custom" to enter an arbitrary API URL plus their own API key. All data is stored locally in browser storage with no telemetry. Available on Chrome, Brave, Edge, Firefox, Vivaldi, Opera, Arc and other Chromium/Firefox browsers. Key handling: Your API key is stored in your browser's local storage. Requests are sent directly to the AI provider. Scores: key handling 62/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 13/25: The key is persisted in browser local storage, which any script on the page can read. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under MIT, so anyone can check how the key is handled. - Privacy 7/15: Nothing specific is known about its data handling beyond the key itself. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/n4ze3m/page-assist/blob/HEAD/README.md) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-27, about 0 month(s) ago. (https://github.com/n4ze3m/page-assist/commits) ### ChatGPT Box - Page: https://www.byoaik.com/tools/chatgpt-box/ - Website: https://github.com/ChatGPTBox-dev/chatGPTBox - Source: https://github.com/ChatGPTBox-dev/chatGPTBox - Category: Browser Extensions | Pricing: OPEN_SOURCE | Open source: yes (MIT) | Trust score: 68/100 - Platforms: Browser Extension - Bring-your-own-key providers: OpenAI, Anthropic, Azure OpenAI, OpenRouter, Ollama, Custom / OpenAI-compatible - Key storage: BROWSER_LOCAL_STORAGE | Key transmission: DIRECT_TO_PROVIDER | Risk: MEDIUM - Maintenance: Actively developed (last commit 6 days ago) ChatGPT Box is an open-source (MIT) browser extension that embeds AI assistants directly into web pages. It offers an on-page chat dialog, an independent conversation page, right-click summarization of any page, and a selection tool for translation, summarization, and text polishing, plus code-page helpers on sites like GitHub. It supports many backends through a unified OpenAI-compatible provider registry: OpenAI (GPT-3.5/4 and web API), Anthropic Claude, Azure, OpenRouter, Moonshot, Ollama, and self-hosted/custom OpenAI-compatible endpoints (including via LiteLLM/One-API). Users supply their own API keys and can configure a custom API base URL. Available for Chrome, Edge, Firefox, Safari, and Android. Key handling: Your API key is stored in your browser's local storage. Requests are sent directly to the AI provider. Scores: key handling 62/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 13/25: The key is persisted in browser local storage, which any script on the page can read. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under MIT, so anyone can check how the key is handled. - Privacy 7/15: Nothing specific is known about its data handling beyond the key itself. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/ChatGPTBox-dev/chatGPTBox/blob/HEAD/README.md) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-28, about 0 month(s) ago. (https://github.com/ChatGPTBox-dev/chatGPTBox/commits) ### Screenpipe - Page: https://www.byoaik.com/tools/screenpipe/ - Website: https://screenpipe.com/?utm_source=byoaik&utm_medium=directory&utm_campaign=evergreen - Source: https://github.com/screenpipe/screenpipe - Category: Desktop AI Assistants | Pricing: SUBSCRIPTION | Open source: no | Trust score: 68/100 - Platforms: Desktop, Local-only - Bring-your-own-key providers: OpenAI, Anthropic, Ollama, Custom / OpenAI-compatible - Key storage: LOCAL_DEVICE | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Actively developed (last commit today) Screenpipe captures screen and audio continuously and turns it into a searchable memory your agents can query, positioned against Rewind, Microsoft Recall and Otter. It is the largest data surface in this directory by a distance, so the key-handling question is unusually consequential: the README states data lives on your machine only, transcription runs on local Whisper by default with Deepgram available as a cloud option, and the AI can be Ollama or any OpenAI-compatible server, in which case no cloud key is involved at all. Two things temper that. Analytics telemetry is on by default and is turned off under Settings, Privacy, Analytics. And the paid Pro tier adds cloud sync and cloud AI, which inverts the property that makes the local setup worth choosing. Not open source: the licence is a Screenpipe Commercial License, source-available for personal non-commercial use, with commercial use requiring a paid licence. The signed desktop app is a subscription from $25 a month. Key handling: Your API key is stored locally on your device. Requests are sent directly to the AI provider. Scores: key handling 80/100 (assessable for any licence), verifiability 33/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 23/25: The key is held on your own device. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 9/20: A public repository exists but the project is not open source. - Privacy 8/15: Local-first: it works without sending your data anywhere. Analytics or error-reporting libraries are present; what they send was not established. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/screenpipe/screenpipe/blob/HEAD/README.md) - [REPORTED via SOURCE_SCAN] References PostHog, so some analytics or error reporting is present. This scan cannot tell whether it is opt-in or what it sends. (https://github.com/screenpipe/screenpipe/blob/HEAD/README.md) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-29, about 0 month(s) ago. (https://github.com/screenpipe/screenpipe/commits) ### WebBrain - Page: https://www.byoaik.com/tools/webbrain/ - Website: https://github.com/webbrain-one/webbrain - Source: https://github.com/webbrain-one/webbrain - Category: Browser Extensions | Pricing: OPEN_SOURCE | Open source: yes (GPL-3.0) | Trust score: 68/100 - Platforms: Browser Extension, Local-only - Bring-your-own-key providers: OpenAI, Anthropic, Ollama, Custom / OpenAI-compatible - Key storage: BROWSER_LOCAL_STORAGE | Key transmission: DIRECT_TO_PROVIDER | Risk: MEDIUM - Maintenance: Actively developed (last commit today) WebBrain is a browser agent for Chrome and Firefox that reads and acts on the page you are on. Its default model needs no key or local setup, and it can be pointed at any OpenAI-compatible server instead, which is the configuration worth choosing if you would rather your browsing context went to a machine you run. Licensed GPL-3.0: the repository's LICENSE opens with a project notice and then reproduces the GPL in full, which is why GitHub reports it as unlicensed. Key handling: Your API key is stored in your browser's local storage. Requests are sent directly to the AI provider. Scores: key handling 62/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 13/25: The key is persisted in browser local storage, which any script on the page can read. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under GPL-3.0, so anyone can check how the key is handled. - Privacy 7/15: Nothing specific is known about its data handling beyond the key itself. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/webbrain-one/webbrain/blob/HEAD/README.md) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-29, about 0 month(s) ago. (https://github.com/webbrain-one/webbrain/commits) ### ChromeClaw - Page: https://www.byoaik.com/tools/chromeclaw/ - Website: https://github.com/algopian/chromeclaw - Source: https://github.com/algopian/chromeclaw - Category: Browser Extensions | Pricing: OPEN_SOURCE | Open source: yes (MIT) | Trust score: 68/100 - Platforms: Browser Extension - Bring-your-own-key providers: OpenAI, Anthropic, Google Gemini, OpenRouter, Custom / OpenAI-compatible - Key storage: BROWSER_LOCAL_STORAGE | Key transmission: DIRECT_TO_PROVIDER | Risk: MEDIUM - Maintenance: Slowing down (last commit about 4 months ago) ChromeClaw brings an agent assistant into Chrome and Firefox as an extension, supporting OpenAI, Anthropic, Google and OpenRouter plus any OpenAI-compatible endpoint on keys you supply. It is a much smaller independent project than the OpenClaw gateway it takes its name from, and should be judged on its own adoption rather than that association. MIT. Key handling: Your API key is stored in your browser's local storage. Requests are sent directly to the AI provider. Scores: key handling 62/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 13/25: The key is persisted in browser local storage, which any script on the page can read. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under MIT, so anyone can check how the key is handled. - Privacy 7/15: Nothing specific is known about its data handling beyond the key itself. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-07-16, about 2 month(s) ago. (https://github.com/algopian/chromeclaw/commits) ### Read Frog - Page: https://www.byoaik.com/tools/read-frog/ - Website: https://readfrog.app/ - Source: https://github.com/mengxi-ream/read-frog - Category: Browser Extensions | Pricing: OPEN_SOURCE | Open source: yes (GPL-3.0 (dual-licensed; commercial license available)) | Trust score: 65/100 - Platforms: Browser Extension - Bring-your-own-key providers: OpenAI, Anthropic, Google Gemini, DeepSeek, xAI Grok, Groq, Mistral, Ollama, OpenRouter, Custom / OpenAI-compatible - Key storage: BROWSER_LOCAL_STORAGE | Key transmission: DIRECT_TO_PROVIDER | Risk: MEDIUM - Maintenance: Actively developed (last commit today) Read Frog is an open-source (GPL-3.0) browser extension for Chrome, Edge, and Firefox that delivers immersive bilingual translation, article/context analysis, and language-learning explanations directly in the page. It connects to 20+ AI providers through the Vercel AI SDK and lets the end user bring their own API key: an "Add Provider" UI exposes per-provider API key fields for OpenAI, Anthropic Claude, Google Gemini, DeepSeek, xAI Grok, Groq, Mistral, Ollama, OpenRouter and more, plus a fully customizable OpenAI-compatible provider option with configurable base URL for self-hosted or third-party endpoints. Free translation engines (Google Translate, Microsoft Translate, DeepLX) are also offered for no-cost basic translation. The extension is actively maintained (latest release v1.35.1, June 2026; ~8k GitHub stars) and dual-licensed with an optional commercial license; a GitHub Sponsors premium tier adds batch requests to cut API costs. Key handling: Your API key is stored in your browser's local storage. Requests are sent directly to the AI provider. Scores: key handling 57/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 13/25: The key is persisted in browser local storage, which any script on the page can read. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under GPL-3.0 (dual-licensed; commercial license available), so anyone can check how the key is handled. - Privacy 4/15: Analytics or error-reporting libraries are present; what they send was not established. - Maintenance 10/10: Actively developed, with commits within the last three months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [STRONG via SOURCE_SCAN] Supports a local model backend, so it can run with no cloud provider key at all. (https://github.com/mengxi-ream/read-frog/blob/HEAD/README.md) - [REPORTED via SOURCE_SCAN] References PostHog, so some analytics or error reporting is present. This scan cannot tell whether it is opt-in or what it sends. (https://github.com/mengxi-ream/read-frog/blob/HEAD/.env.example) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-09-29, about 0 month(s) ago. (https://github.com/mengxi-ream/read-frog/commits) ### GPT Pilot - Page: https://www.byoaik.com/tools/gpt-pilot/ - Website: https://github.com/Pythagora-io/gpt-pilot - Source: https://github.com/Pythagora-io/gpt-pilot - Category: App Builders | Pricing: FREE | Open source: no | Trust score: 64/100 - Platforms: CLI, Local-only, Self-hosted, Docker - Bring-your-own-key providers: OpenAI, Anthropic, Custom / OpenAI-compatible - Key storage: ENVIRONMENT_VARIABLE | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Maintenance: Slowing down (last commit about 5 months ago) GPT Pilot writes applications incrementally, pausing for review rather than emitting a finished codebase, and runs locally against a model key you supply. Read the licence before committing to it: the file is the Functional Source License 1.1 with an MIT Future License, meaning it is source-available now, restricted against competing uses, and converts to MIT two years after each release. That is a real grant on a delay, not open source today, and the distinction matters if you are deciding what to build a business on. Key handling: Your API key is supplied via an environment variable or local config file. Requests are sent directly to the AI provider. Because it can be self-hosted, your key never has to touch a third-party backend. Scores: key handling 77/100 (assessable for any licence), verifiability 33/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 22/25: The key is supplied by an environment variable or local config file you control. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 9/20: A public repository exists but the project is not open source. - Privacy 7/15: Can be self-hosted, so the data path stays inside infrastructure you control. Analytics or error-reporting libraries are present; what they send was not established. - Maintenance 8/10: Commits within the last six months. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [REPORTED via SOURCE_SCAN] References PostHog, so some analytics or error reporting is present. This scan cannot tell whether it is opt-in or what it sends. (https://github.com/Pythagora-io/gpt-pilot/blob/HEAD/Dockerfile) - [CONFIRMED via SOURCE_SCAN] Ships a container definition, so it can be self-hosted on your own infrastructure. (https://github.com/Pythagora-io/gpt-pilot/blob/HEAD/Dockerfile) - [REPORTED via SOURCE_SCAN] References Sentry, so some analytics or error reporting is present. This scan cannot tell whether it is opt-in or what it sends. (https://github.com/Pythagora-io/gpt-pilot/blob/HEAD/requirements.txt) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-06-12, about 4 month(s) ago. (https://github.com/Pythagora-io/gpt-pilot/commits) ### ChatHub - Page: https://www.byoaik.com/tools/chathub/ - Website: https://github.com/chathub-dev/chathub - Source: https://github.com/chathub-dev/chathub - Category: Browser Extensions | Pricing: OPEN_SOURCE | Open source: yes (GPL-3.0) | Trust score: 60/100 - Platforms: Browser Extension - Bring-your-own-key providers: OpenAI, Anthropic, Custom / OpenAI-compatible - Key storage: BROWSER_LOCAL_STORAGE | Key transmission: DIRECT_TO_PROVIDER | Risk: MEDIUM - Maintenance: Actively developed (last commit 19 days ago) ChatHub puts multiple assistants in one browser panel so answers can be compared side by side, and supports each through either a web session or an API key you supply. The API path is the one that concerns this directory, and it is genuinely bring-your-own-key. Read the maintenance badge before installing: the default branch has not been committed to for several months and the README still lists services that have since been discontinued, so treat the provider list as historical rather than current. GPL-3.0. Key handling: Your API key is stored in your browser's local storage. Requests are sent directly to the AI provider. Scores: key handling 57/100 (assessable for any licence), verifiability 70/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 13/25: The key is persisted in browser local storage, which any script on the page can read. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 20/20: Source is public under GPL-3.0, so anyone can check how the key is handled. - Privacy 4/15: Analytics or error-reporting libraries are present; what they send was not established. - Maintenance 5/10: Last commit was within the last year. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. - [REPORTED via SOURCE_SCAN] References Sentry, so some analytics or error reporting is present. This scan cannot tell whether it is opt-in or what it sends. (https://github.com/chathub-dev/chathub/blob/HEAD/package.json) - [CONFIRMED via SOURCE_SCAN] Most recent commit 2026-02-27, about 7 month(s) ago. (https://github.com/chathub-dev/chathub/commits) ### Pal Chat - Page: https://www.byoaik.com/tools/pal-chat/ - Website: https://apps.apple.com/app/id6447545085 - Category: AI Chat Interfaces | Pricing: FREE | Open source: no | Trust score: 59/100 - Platforms: Mobile - Bring-your-own-key providers: OpenAI, Anthropic, Google Gemini, OpenRouter, Groq, Mistral, DeepSeek - Key storage: LOCAL_DEVICE | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW Pal Chat is a free-to-download, closed-source native app for iOS, iPadOS and Apple Silicon Macs for chatting with models from OpenAI, Anthropic, Google, Groq, OpenRouter, DeepSeek and Mistral, plus image generation and local models. The App Store listing instructs users to obtain their own API key and enter it in the app; no Pal-hosted proxy or account is required for core chat, and usage is billed by the provider. An optional Pro purchase unlocks extra features but not the bring-your-own-key capability itself. No public repository exists, so key handling cannot be inspected. Key handling: Your API key is stored locally on your device. Requests are sent directly to the AI provider. Scores: key handling 78/100 (assessable for any licence), verifiability 27/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 23/25: The key is held on your own device. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 3/20: Closed source with no public repository, so the key handling cannot be independently inspected. - Privacy 7/15: Nothing specific is known about its data handling beyond the key itself. - Maintenance 4/10: Maintenance activity has not been established for this listing. - Verification Confidence 5/10: Key handling is quoted verbatim from official documentation. ### Msty - Page: https://www.byoaik.com/tools/msty/ - Website: https://msty.ai - Category: Desktop AI Assistants | Pricing: FREE | Open source: no | Trust score: 59/100 - Platforms: Desktop - Bring-your-own-key providers: OpenAI, Anthropic, Google Gemini, OpenRouter, Ollama, xAI Grok - Key storage: LOCAL_DEVICE | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW Msty is a closed-source but free desktop app for chatting with both local models (via built-in Ollama integration) and cloud models using your own API keys. Its documentation includes a dedicated 'Find API Keys' guide directing users to obtain and paste keys for OpenAI, Anthropic, Gemini, xAI Grok and others, and users can route through OpenRouter with their own key. The free tier covers chat, personas, agent mode and knowledge stacks with no subscription required for bring-your-own-key usage; a paid tier unlocks extra features. Available for Windows, macOS and Linux. Key handling: Your API key is stored locally on your device. Requests are sent directly to the AI provider. Scores: key handling 85/100 (assessable for any licence), verifiability 13/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 23/25: The key is held on your own device. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 3/20: Closed source with no public repository, so the key handling cannot be independently inspected. - Privacy 11/15: Local-first: it works without sending your data anywhere. - Maintenance 4/10: Maintenance activity has not been established for this listing. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. ### AI Actions for Shortcuts - Page: https://www.byoaik.com/tools/ai-actions-for-shortcuts/ - Website: https://sindresorhus.com/ai-actions - Category: Productivity Tools | Pricing: FREE | Open source: no | Trust score: 59/100 - Platforms: Mobile, Desktop - Bring-your-own-key providers: OpenAI, Anthropic, Google Gemini, Groq, OpenRouter, xAI Grok, Ollama, Custom / OpenAI-compatible - Key storage: LOCAL_DEVICE | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW A free app by Sindre Sorhus that adds AI-powered actions to Apple's Shortcuts app on iOS/iPadOS, macOS (Apple silicon), and visionOS. It lets you build shortcuts that call ChatGPT/OpenAI, Anthropic (Claude), and OpenAI-compatible endpoints (Ollama for local models, Groq, Google Gemini, OpenRouter, xAI Grok) directly from your own provider account. The app requires you to supply your own API key, which is stored securely in the system keychain rather than embedded in the shortcut, so shortcuts can be shared without leaking credentials. Supports image inputs and DALL-E image generation; App Store description and reviews also cite image and PDF handling. No subscription, no ads, no in-app purchases. Note: the GitHub repo github.com/sindresorhus/Actions belongs to a different (and now closed-source) app, the general-purpose 'Actions' shortcuts app; AI Actions itself has no public open-source repository. Key handling: Your API key is stored locally on your device. Requests are sent directly to the AI provider. Scores: key handling 85/100 (assessable for any licence), verifiability 13/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 23/25: The key is held on your own device. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 3/20: Closed source with no public repository, so the key handling cannot be independently inspected. - Privacy 11/15: Local-first: it works without sending your data anywhere. - Maintenance 4/10: Maintenance activity has not been established for this listing. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. ### Granth - Page: https://www.byoaik.com/tools/granth/ - Website: https://granthwriter.com - Category: Writing Tools | Pricing: FREE | Open source: no | Trust score: 59/100 - Platforms: Desktop - Bring-your-own-key providers: OpenAI, Anthropic, Google Gemini, OpenRouter, Ollama - Key storage: LOCAL_DEVICE | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW - Disclosure: Granth is built by the maintainer of BYOAIK. It is listed on the same published criteria as every other tool, and its trust score is unadjusted. - Developer claimed: ownership proven 2026-08-10; key handling below is the developer's stated architecture Granth is a free, local-first 'Writer Operating System' for long-form fiction and nonfiction. Projects live as human-readable .granth folders of plain Markdown and YAML on your own disk, Git-friendly and usable without the app, and writing, organizing, search, and export all work fully offline with no account. AI is an optional bring-your-own-key layer: connect OpenAI, Anthropic, Google Gemini, OpenRouter, or a local model via Ollama. The API key is stored in the operating system's secure keychain: never in a file, database, or the app's config. Before any text leaves the machine, a consent dialog shows the exact payload and destination provider; every AI interaction is logged to an audit file (ai/usage/log.jsonl) inside the project folder; and AI output is suggestion-only until explicitly accepted. Features include a distraction-free editor with focus mode, typewriter scrolling, and snapshots; a 'Project Brain' that answers questions about the whole manuscript with citations into your own material; story bible, relationship map, threads, and timeline for fiction; a research vault, claim tracker, and citation engine (APA/MLA/Chicago) for nonfiction; and publication-grade export to print-ready PDF, validated EPUB, and DOCX with KDP and IngramSpark profiles. Desktop app for macOS, Windows, and Linux (v0.1.0) with no telemetry, no analytics, and no training on user content. Closed source with no public repository. Key handling: Your API key is stored locally on your device. Requests are sent directly to the AI provider. Scores: key handling 85/100 (assessable for any licence), verifiability 13/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 23/25: The key is held on your own device. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 3/20: Closed source with no public repository, so the key handling cannot be independently inspected. - Privacy 11/15: Local-first: it works without sending your data anywhere. - Maintenance 4/10: Maintenance activity has not been established for this listing. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. ### Raycast - Page: https://www.byoaik.com/tools/raycast/ - Website: https://manual.raycast.com/ai/bring-your-own-keys - Category: Productivity Tools | Pricing: FREE | Open source: no | Trust score: 59/100 - Platforms: Desktop, Mobile - Bring-your-own-key providers: OpenAI, Anthropic, Google Gemini, OpenRouter - Key storage: LOCAL_DEVICE | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW Raycast is a free, closed-source launcher and productivity app for macOS with an iOS companion. Its Bring Your Own Key feature lets users add a personal key per provider under Settings, AI, Custom API Keys and use Raycast's AI chat, commands and extensions without a Pro subscription, with iOS additionally supporting OpenRouter. Users are billed directly by their chosen provider at standard rates. The base app and BYOK usage are free; Pro remains optional for cloud sync and other non-AI features. No public repository, so key handling cannot be inspected in source. Key handling: Your API key is stored locally on your device. Requests are sent directly to the AI provider. Scores: key handling 78/100 (assessable for any licence), verifiability 27/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 23/25: The key is held on your own device. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 3/20: Closed source with no public repository, so the key handling cannot be independently inspected. - Privacy 7/15: Nothing specific is known about its data handling beyond the key itself. - Maintenance 4/10: Maintenance activity has not been established for this listing. - Verification Confidence 5/10: Key handling is quoted verbatim from official documentation. ### GitHub Copilot (BYOK) - Page: https://www.byoaik.com/tools/github-copilot-byok/ - Website: https://docs.github.com/en/copilot/how-tos/administer-copilot/manage-for-enterprise/use-your-own-api-keys - Category: Coding Assistants | Pricing: SUBSCRIPTION | Open source: no | Trust score: 59/100 - Platforms: Desktop, Self-hosted - Bring-your-own-key providers: Anthropic, OpenAI, Google Gemini, Azure OpenAI, OpenRouter, xAI Grok, Ollama, Custom / OpenAI-compatible - Key storage: LOCAL_DEVICE | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW GitHub Copilot's Bring Your Own Key (BYOK) lets developers use Copilot Chat and agents in VS Code (and supported IDEs) with their own LLM provider API keys instead of GitHub-hosted models. Configured via the "Chat: Manage Language Models" command in VS Code, BYOK supports models from Anthropic, OpenAI, Google Gemini, Azure OpenAI, OpenRouter, xAI Grok, AWS Bedrock, and any OpenAI-compatible endpoint, plus local models via Ollama and Foundry Local. BYOK usage is billed directly by the chosen provider and does not count against Copilot request quotas. Generally available as of April 2026 for Copilot Business and Enterprise customers (admins can disable the policy). BYOK applies to chat/agent models, not code completions. Copilot itself is a closed-source, paid product. Key handling: Your API key is stored locally on your device. Requests are sent directly to the AI provider. Scores: key handling 85/100 (assessable for any licence), verifiability 13/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 23/25: The key is held on your own device. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 3/20: Closed source with no public repository, so the key handling cannot be independently inspected. - Privacy 11/15: Local-first: it works without sending your data anywhere. - Maintenance 4/10: Maintenance activity has not been established for this listing. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. ### MacWhisper - Page: https://www.byoaik.com/tools/macwhisper/ - Website: https://goodsnooze.gumroad.com/l/macwhisper - Category: Voice and Transcription Tools | Pricing: FREE | Open source: no | Trust score: 59/100 - Platforms: Desktop, Mobile, Web - Bring-your-own-key providers: OpenAI, Anthropic, Groq, Ollama, xAI Grok, DeepSeek, Azure OpenAI, Custom / OpenAI-compatible, Google Gemini, OpenRouter, Mistral, Fireworks - Key storage: LOCAL_DEVICE | Key transmission: DIRECT_TO_PROVIDER | Risk: LOW MacWhisper (by Jordi Bruin / Good Snooze) is a popular paid macOS transcription app that runs OpenAI's Whisper models locally for private, offline transcription of audio, meetings, and YouTube videos. Beyond local transcription, it offers optional cloud speech-to-text and LLM post-processing using the end user's own API keys. For cloud transcription, users can plug in keys for OpenAI, Groq, ElevenLabs, Deepgram, or any custom OpenAI-compatible endpoint. For AI cleanup, summarization, and translation of transcripts and dictation, users supply their own keys for OpenAI/ChatGPT, Anthropic Claude, Groq, Ollama (local), xAI, DeepSeek, Azure AI, and custom OpenAI-compatible endpoints. Official docs explicitly state users must "fill in your OpenAI API key." It is closed-source and sold as a one-time purchase (with an optional hosted Assistant subscription that does not require keys). The app is widely used (220k+ sales, 4.8/5 on Product Hunt across ~1,900 ratings) and actively maintained. Key handling: Your API key is stored locally on your device. Requests are sent directly to the AI provider. Scores: key handling 85/100 (assessable for any licence), verifiability 13/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 23/25: The key is held on your own device. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 3/20: Closed source with no public repository, so the key handling cannot be independently inspected. - Privacy 11/15: Local-first: it works without sending your data anywhere. - Maintenance 4/10: Maintenance activity has not been established for this listing. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it. ### Monet - Page: https://www.byoaik.com/tools/monet/ - Website: https://www.monet.gg - Category: Developer Infrastructure | Pricing: FREE | Open source: no | Trust score: 43/100 - Platforms: Web - Bring-your-own-key providers: OpenAI, Anthropic, Google Gemini, Custom / OpenAI-compatible - Key storage: SERVER_SIDE_ENCRYPTED | Key transmission: DIRECT_TO_PROVIDER | Risk: MEDIUM Monet is an OAuth-style consent flow for AI provider credentials. You connect your own OpenAI, Anthropic or Google key once, and an integrating app receives a scoped, revocable bearer token rather than your raw secret, calling one OpenAI-compatible endpoint through Monet. The vendor states credentials are held AES-256-GCM encrypted and decrypted in-process only to proxy upstream calls, that the integrating app never touches the raw credential, and that you can revoke access yourself. Note what that means for the key journey: your key rests with Monet and every request is proxied through it, so this trades direct provider contact for revocability and per-app scoping. Closed source with no public repository, so none of it can be independently inspected. Free through beta. Key handling: Your API key is stored encrypted on the server instance you control. Requests are sent directly to the AI provider. Scores: key handling 58/100 (assessable for any licence), verifiability 13/100 (how much BYOAIK could confirm; lower for closed source). No popularity signal. Breakdown: - Key Safety 11/25: The key is stored server-side, encrypted, on an instance you control. - Request Routing 17/20: Requests go straight from you to the AI provider. - Transparency 3/20: Closed source with no public repository, so the key handling cannot be independently inspected. - Privacy 7/15: Nothing specific is known about its data handling beyond the key itself. - Maintenance 4/10: Maintenance activity has not been established for this listing. - Verification Confidence 1/10: Compiled from public documentation by an AI-assisted pass, not independently confirmed. What was checked (verification tier RESEARCH_ASSISTED, derived from this evidence): - [REPORTED via RESEARCH] This listing was compiled by an AI-assisted research pass over the tool's public website, README and documentation. No person independently confirmed it.