danny-avila/LibreChat

A self-hosted, open-source AI chat platform that unifies many model providers behind one ChatGPT-style interface with agents, tools, and multi-user auth.

What it is

LibreChat is a self-hosted web application for chatting with AI models from a single, privacy-focused interface. It connects to major providers and any OpenAI-compatible endpoint, and adds agents, Model Context Protocol (MCP) support, artifacts, a code interpreter, web search, and enterprise-style multi-user authentication on top of plain chat. It is for teams and individuals who want to run their own AI chat infrastructure rather than depend on a single hosted product.

Key features

Model selection across Anthropic, AWS Bedrock, OpenAI, Azure OpenAI, Google, Vertex AI, and the OpenAI Responses API, plus custom endpoints for any OpenAI-compatible API and local/remote providers such as Ollama, Mistral, Groq, and OpenRouter.
A sandboxed Code Interpreter API that runs Python, Node.js, Go, C/C++, Java, PHP, Rust, and Fortran with file upload and download.
Agents and tools: no-code custom assistants, an agent marketplace, reusable SKILL.md bundles, subagents with isolated context windows, and MCP server support.
Generative UI with code artifacts (React, HTML, Mermaid), plus image generation and editing via GPT-Image-1, DALL·E, Stable Diffusion, Flux, or MCP servers.
Resumable streams that reconnect dropped responses and sync across tabs and devices, production-ready with Redis for horizontally scaled deployments.
Multi-user secure authentication (OAuth2, LDAP, email), a browser-based admin panel for users/roles/config, message search, import/export, speech-to-text and text-to-speech, and a multilingual UI.

Tech stack

Primary language: TypeScript (with JavaScript across the API), organized as an npm workspaces monorepo covering api, client, and packages/*.
Package manager pinned to npm@11.13.0, builds orchestrated with Turbo, end-to-end tests with Playwright, and linting/formatting via ESLint and Prettier.
Search indexing via Meilisearch (present in the API config), and Redis for resumable streams and scaled deployments.
Ships with Dockerfiles and Docker Compose stacks, with one-click deploy templates for Railway, Zeabur, and Sealos.
License: MIT. Repository version reported as v0.8.7.

When to reach for it

Running a shared, self-hosted chat platform for a team that needs multi-user auth, roles, and an admin panel.
Consolidating many model providers, including local and OpenAI-compatible endpoints, behind one UI.
Building and sharing agents that use tools, MCP servers, file search, and code execution.
Keeping conversation data and AI infrastructure under your own control for privacy or compliance reasons.

When not to reach for it

You just want a personal desktop chat app with minimal setup; a full self-hosted stack with database, search, and optional Redis is more than that requires.
You are content with a single hosted assistant and do not need provider switching, self-hosting, or multi-user features.
You lack the operational capacity to deploy and maintain a Dockerized, multi-service application.

Maturity signal

LibreChat has been developed since early 2023, carries a very large star count, and shows a push date within the last day, all of which point to an actively developed, widely adopted project. A sizable open-issue backlog is expected for software of this breadth and user base rather than a sign of neglect. The MIT license is permissive, the repository is not archived, and the README stresses consulting the changelog for breaking changes before updating — consistent with a fast-moving, feature-rich codebase.

Alternatives

Open WebUI — reach for it when your priority is a local, Ollama-centric chat UI with a lighter setup.
AnythingLLM — reach for it when the core need is document-centric RAG workspaces rather than broad multi-provider chat.
Jan or LM Studio — reach for these when you want a desktop app for local models instead of a self-hosted multi-user server.

Notes

The Code Interpreter is open-source and self-hostable, powered by ClickHouse/code-interpreter, so sandboxed execution can run without sending code to a third party.
The UI is translated into a large number of languages, and the project maintains a separate RAG API repository and website repository outside the main codebase.

Wiki: danny-avila/LibreChat