LibreChat

What LibreChat does

LibreChat is a web application that gives a team or a whole organization one ChatGPT-like interface in front of many AI models. Users get conversation history, file uploads, editing and branching of messages, saved presets and the option to change model in the middle of a chat. Administrators decide which providers and models are available, who may use agents or code execution, and how much each user can spend in tokens.

The feature list goes well beyond chat. Agents can combine instructions, file search, code execution, MCP tools and subagents, and can be shared with specific users or groups. Web search pulls live results into the context, artifacts render React, HTML and Mermaid output beside the conversation, and a memory feature carries context from one conversation to the next. Image generation and voice input and output are available when a suitable provider is configured.

Models and providers

LibreChat does not run models itself. It connects to OpenAI, Azure OpenAI, Anthropic, Google (Gemini and Vertex AI), AWS Bedrock and a long list of other APIs, and any OpenAI-compatible server can be added as a custom endpoint. The documentation has setup pages for local servers such as Ollama, vLLM, LiteLLM and Apple MLX.

The open source license covers the chat application, not the models. If an endpoint points at OpenAI, Anthropic or Google, prompts and uploaded files are sent to that company under its terms. Keeping conversations entirely on your own hardware means using local servers only.

Privacy and data

Conversations, accounts and settings are stored in your own MongoDB database. Meilisearch indexes messages for search, and files used for document chat are embedded by the RAG API into PostgreSQL with pgvector. The example configuration includes OpenTelemetry tracing, Langfuse and browser monitoring with optional session replay, all switched off by default and sending data only to collectors the operator sets up. We found no documented telemetry that reports to the LibreChat developers.

Web search, reranking and image tools may call outside services such as Serper, Firecrawl, Jina or Cohere. SearXNG can be used as the search provider instead.

Self-hosting LibreChat

The recommended install is Docker Compose, which starts LibreChat with MongoDB, Meilisearch, the RAG API, a pgvector database and the admin panel. It can also be installed with npm (the docs list Node.js v24 and an external MongoDB) or deployed to Kubernetes with a Helm chart; Redis supports horizontally scaled setups. There is no desktop app: the docs say plainly that LibreChat is a self-hosted web application used in a browser.

Sign-in options include email and password, OAuth2 providers (GitHub, Google, Discord, Facebook, Apple), OpenID Connect, SAML and LDAP, with two-factor authentication. The browser-based admin panel manages users, groups, roles and permissions without redeploying. Code execution needs a separate sandbox service; the docs point to ClickHouse’s open source code-interpreter (Apache-2.0), which you host yourself.

Hosted version

There is no hosted subscription. The project runs a public demo at chat.librechat.ai. ClickHouse acquired LibreChat in November 2025 and says existing deployments are unaffected.

Limitations

LibreChat is the heaviest of the common chat UIs to run: four supporting services in the default stack, and configuration split between a .env file and librechat.yaml with many options. Web search and code execution only work after extra services or API keys are configured. Local models need a separately installed server. The project warns that some upgrades contain breaking changes, so read the changelog before updating.

Who it suits

LibreChat fits companies, universities and teams that want an internal ChatGPT replacement for many users, with single sign-on, roles, spending limits and a choice of providers. For a single person who mainly wants to run local models, Jan or AnythingLLM need far less setup. See the AI chat guide for how the options compare.