LibreChat is a self-hosted AI chat platform that puts many model providers behind one chat interface modeled on ChatGPT. It connects to Anthropic, OpenAI, Azure OpenAI, AWS Bedrock, Google, and Vertex AI, and to any OpenAI-compatible API through custom endpoints, which covers local and remote providers such as Ollama, Mistral AI, Deepseek, Together AI, and OpenRouter. You can switch endpoints and presets in the middle of a conversation.
Beyond chat, LibreChat runs agents. Agents are built without code, shared with users and groups or published to an agent marketplace, and use MCP servers, file search, code execution, web search, and reusable skill bundles. Subagents delegate focused work to child runs with their own context windows, and an Agent Management API (beta) manages agents, files, and skills from machine clients authenticated with OIDC. An experimental attached-workspace mode lets agents read, search, edit, and run commands in a code workspace, with Ask, Allow, or Deny approval controls.
A sandboxed code interpreter runs Python, Node.js, Go, C and C++, Java, PHP, Rust, and Fortran. Code artifacts render React, HTML, and Mermaid output in the chat, and image tools generate or edit images through OpenAI image models, Stable Diffusion, Flux, or an MCP server.
LibreChat is multi-user: it supports OAuth2, LDAP, and email login, moderation and token-spend controls, and a browser-based admin panel for users, groups, roles, and permission overrides. Streams resume after a dropped connection and sync across tabs and devices, with Redis for horizontally scaled deployments. Traces and logs export through OpenTelemetry, with a Langfuse connection for agent and model insights. It runs locally or in the cloud.
Features
- Model providers: Anthropic, OpenAI, Azure OpenAI, AWS Bedrock, Google, Vertex AI, and the OpenAI Responses API
- Custom endpoints: any OpenAI-compatible API without a proxy, including Ollama, Mistral AI, Deepseek, Qwen, Cohere, groq, Together AI, and OpenRouter
- Agents: no-code agents with MCP tools, file search, code execution, skills, subagents, and sharing with users and groups
- Agent Management API: create, update, and delete agents, files, and skills from OIDC-authenticated machine clients, with OpenAPI and Swagger UI
- Code interpreter: sandboxed execution in Python, Node.js, Go, C and C++, Java, PHP, Rust, and Fortran, with file upload and download
- Web search: combines search providers, content scrapers, and rerankers, including configurable Jina reranking
- Code artifacts: React, HTML, and Mermaid output rendered in chat, with SVG or PNG export for diagrams
- Conversation control: presets, message editing and resubmission, conversation branching and forking, and on-demand context compaction
- Multimodal input: image analysis with vision models and chat with uploaded files
- Speech: speech-to-text and text-to-speech through OpenAI, Azure OpenAI, and ElevenLabs
- Multi-user access: OAuth2, LDAP, and email login, moderation, token-spend tools, and an admin panel for roles and permissions
- Observability: OpenTelemetry export of traces and logs, a Trace Viewer for model and tool rounds, and Langfuse integration
