// the find
danny-avila/LibreChat
Enhanced ChatGPT Clone: Features Agents, MCP, Skills, DeepSeek, Anthropic, AWS, OpenAI, Responses API, Azure, Groq, o1, GPT-5, Mistral, OpenRouter, Vertex AI, Gemini, Artifacts, AI model switching, message search, Code Interpreter, langchain, DALL-E-3, OpenAPI Actions, Functions, Secure Multi-User Auth, Presets, open-source for self-hosting. Active
LibreChat is a self-hosted web UI that puts every major AI provider (OpenAI, Anthropic, Google, Bedrock, local Ollama, etc.) behind a single ChatGPT-style interface. It's aimed at teams or individuals who want provider flexibility and data control without paying for a SaaS wrapper. At 39k stars with daily commits, it's the most actively maintained project in this space.
MCP support is first-class and well-integrated — not bolted on — which matters now that the protocol is gaining real traction. The resumable streams feature (auto-reconnect, multi-tab/device sync, Redis-backed) is genuinely production-grade engineering, not a checkbox. The agent system supports subagents with isolated context windows, which is the right architecture for complex workflows rather than stuffing everything into one context. Test coverage is real: hundreds of spec files across controllers, middleware, and routes — not just a token test folder.
The config surface is massive — librechat.yaml alone has enough knobs to fill a small book, and getting a production deployment right requires reading through layers of docs. The RAG implementation is a separate Python service (rag_api) you have to run alongside the Node backend, which adds operational complexity most self-hosters will underestimate. MongoDB is the database — fine at this scale, but the schema is document-soup by now after years of feature additions, and there's no migration story worth speaking of. The JavaScript/Node backend is a single large Express app; at enough concurrent users the lack of any worker-process model will show.