LibreChat
The open-source AI platform
About LibreChat
LibreChat is a self-hosted chat application that puts every model provider you use behind a single interface. Anthropic, OpenAI, Azure, AWS Bedrock, Google, Groq, Mistral, OpenRouter and locally hosted models are all configured as endpoints and switched from one menu, and the conversation history, search, presets and file handling are the same regardless of which model answered. It grew out of a ChatGPT clone in early 2023 and has become the more enterprise-minded of the two large open-source chat front ends, the other being Open WebUI, which this directory also lists. The reason to run something like this instead of several vendor web apps is control: your data stays on your infrastructure, your users authenticate against your identity provider, and the model bill is whatever your own API keys cost.
Two things distinguish it from a bare chat UI. First, agents: LibreChat agents can carry files, run code through a built-in code interpreter, call external APIs through OpenAPI actions and connect to tools over the Model Context Protocol, so a shared internal assistant can actually do things rather than only answer. Second, the authentication story, which is the part enterprises ask about first: OAuth, SAML, LDAP and two-factor authentication are on the homepage feature list, alongside multi-user management. The licence is MIT with no branding clause, which is the caveat that has to be attached to Open WebUI. What has changed recently is the owner. ClickHouse acquired LibreChat in November 2025; the announcement says the project remains 100% open source under MIT and that community development continues, but the roadmap is now tied to ClickHouse's data and analytics ambitions, and the formerly paid Code Interpreter API is being discontinued in its current form.
You deploy LibreChat with Docker Compose, either on a laptop for a single user or on a server for a team, following the local or remote install guides in the documentation. A configuration file declares the endpoints: each provider with its API key, the models to expose, and any custom OpenAI-compatible endpoints, including local runtimes. Users sign in through whichever auth method you enabled, pick a model or an agent, and chat; conversations are stored in your own database and are searchable across messages, files and code snippets. Agents are defined in the interface with instructions, attached files, a chosen model and a set of tools, then shared with other users. MCP servers plug in through configuration, so a tool written for Claude Desktop or Cursor can be offered to every LibreChat user. Memory lets an agent keep context across conversations, and a built-in web search with reranking gives any model live internet access.
- •Every Provider, One Interface - Anthropic, OpenAI, Azure, Bedrock, Google, Groq, Mistral, OpenRouter, Vertex and local models, switchable mid-conversation
- •Agents With Tools - File handling, a code interpreter, OpenAPI actions and MCP connections, shareable across users
- •Artifacts - React, HTML and Mermaid output rendered live in the chat
- •Search and Memory - Full-text search across messages, files and code, plus persistent memory between conversations
- •Enterprise Authentication - OAuth, SAML, LDAP and two-factor, with multi-user management
- •MIT Licensed and Self-Hosted - Docker deployment on your own infrastructure; the project reports 53 million Docker pulls and over 400 contributors
Teams that want a shared, controllable AI assistant for the whole company without buying seats from one model vendor, and platform engineers asked to give hundreds of people access to several models behind corporate SSO. Developers who switch between models and want one history and one set of tools will also get a lot out of it on a single machine. It is not for anyone who wants a managed product; you run the containers, you hold the keys, and you upgrade it. If the ClickHouse acquisition matters to your risk assessment, read the announcement before adopting, and if you only need a quick local front end for Ollama, Open WebUI is the lighter choice.
