LibreChat

LibreChat is an MIT-licensed, open-source AI chat platform that you install and run on your own infrastructure, so conversations, uploaded files, and agent definitions stay in a database you control. Where most chat products lock you to a single model vendor, LibreChat puts one interface over OpenAI, Anthropic, Google, Azure, AWS Bedrock, Ollama, and any OpenAI-compatible endpoint you add yourself, including self-hosted inference behind a custom URL. Users pick a model per conversation, save presets across providers, and keep one searchable history instead of juggling four browser tabs and three subscriptions.

On top of the chat surface it ships an application platform rather than a thin wrapper. Agents are built in a no-code builder with instructions, avatars, conversation starters, a tool library, MCP servers, actions, and skills, and they can be invoked from the endpoint menu or mentioned inline with an @ sign. The same build includes a sandboxed code interpreter, generative UI artifacts for React and HTML output plus Mermaid diagrams, retrieval over uploaded files through the bundled RAG API, OCR for scanned documents, image generation and editing, web search with reranking, persistent user memory, scheduled chats, subagents, and conversation pull requests that let a team review a chat before it becomes a shared asset. Meilisearch-backed message search covers messages, files, and code snippets, while shareable links, forked chats, projects, and conversation import handle everyday collaboration.

Deployment is deliberately boring, which is the point for operations teams. Docker Compose brings up the app together with MongoDB, Meilisearch, and the RAG API in one command, an npm path covers hosts that prefer Node.js directly, and one-click Railway and remote hosting guides cover cloud installs. Configuration lives in a librechat.yaml file and environment variables, so providers, model lists, rate limits, and interface toggles are version-controlled like any other infrastructure.

Governance is where it separates from consumer chat apps. Authentication covers email and password plus Google, GitHub, Discord, and OpenID logins, and adds SSO through OAuth, SAML, and LDAP with two-factor authentication. A three-layer authorization model combines role-level feature permissions, per-resource access lists on agents, prompts, MCP servers, and shared links, and system-level admin grants, all evaluated against individual users, groups synced from Entra ID, roles, and the instance-wide public. An admin panel and usage insights sit alongside automated moderation and password reset controls.

The project publishes no pricing page and no paid plan; it is MIT licensed software you run. It lists 46.3k GitHub stars, 59.1 million Docker pulls, and 421 contributors, ships an interface translated into more than 30 languages, and displays a banner saying LibreChat is joining ClickHouse to power the open-source Agentic Data Stack. A public demo at chat.librechat.ai exists for evaluation only, so it is not a place for private data. Any hosted or vendor-supported arrangement is not publicly priced, which is why procurement here usually means hosting it yourself.