LibreChat

LibreChat is a free, MIT-licensed, self-hosted AI chat platform from Danny Avila and the LibreChat contributors. With 36,000+ GitHub stars, MIT licensing, and a 2025 acquisition by ClickHouse, it has grown from a weekend ChatGPT clone into a serious open-source platform that small teams use as a private, multi-provider replacement for ChatGPT Team and Claude Teams.

What LibreChat does:

• •Switch any model mid-conversation chat with GPT-5, Claude Opus 4.6, Gemini 3.1 Pro, DeepSeek, Mistral, Groq, or a local Llama on Ollama in the same thread, with context preserved across providers.
• •Agents with tool use build custom AI agents that handle file operations, run code, hit your APIs through OpenAPI Actions, and call MCP servers. Each agent has its own instructions, tools, and model.
• •Built-in code interpreter run Python, JavaScript, TypeScript, Go, and other languages in a sandboxed environment with zero setup. The agent writes and executes code without leaving the chat.
• •MCP support with dynamic servers connect any Model Context Protocol server (file system, databases, web search, custom tools) and expose it to your agents with per-user access control.
• •Document RAG and file uploads upload PDFs, code, spreadsheets, or images. LibreChat ingests them, builds embeddings, and answers questions grounded in your documents without sending them to a third-party RAG service.
• •Web search with reranking live internet access through built-in search and reranking, so agents pull fresh sources and cite them in answers.
• •Artifacts panel generate React components, HTML, and Mermaid diagrams that render side-by-side with the chat, like Claude artifacts but in your own deployment.
• •Enterprise SSO OAuth (Google, GitHub, Discord, Apple, Facebook), Azure AD, AWS Cognito, OpenID Connect, SAML, LDAP, and 2FA, all in the free self-hosted version.
• •Persistent memory the platform remembers context across conversations per user, so your agents do not start from zero every time.
• •Customer base proves it works Shopify, Daimler Truck, Boston University, ClickHouse, and Stripe self-host LibreChat for internal AI access.

Pricing:

• •Self-hosted $0: free forever under the MIT licence. You pay only for the server you run it on (a $5/month VPS handles small teams) plus any LLM API costs if you use cloud models.
• •Cloud / managed hosting not officially offered by the project: LibreChat is self-host-first. Third-party providers (Pinggy, Coolify, Railway templates) host it for you starting around $5-10/month.
• •LLM API costs pass-through: bring your own keys for OpenAI, Anthropic, Google, etc. Pair with Ollama for fully free local inference if you have the hardware.

Limitations:

• •Docker familiarity required the recommended install path is Docker Compose, so non-technical operators will need help setting it up and applying updates.
• •Smaller community than Open WebUI 36k stars vs Open WebUI's 136k, which means fewer community plugins, integrations, and Stack Overflow answers.
• •No first-party hosted version the LibreChat team does not run a managed service, so if you want hosted-and-supported you go through a third party or run it yourself.
• •Local-model performance is hardware-bound a small VPS will not run 70B-parameter models comfortably without a GPU, so heavy local workloads need a real GPU box or paired cloud-API access.