LibreChat

Switch among cloud, local, and OpenAI-compatible AI providers from one conversation interface

The official README lists OpenAI, Anthropic, Google and Vertex AI, AWS Bedrock, Azure OpenAI, plus local or remote providers such as Ollama, Groq, Mistral, OpenRouter, DeepSeek, Qwen, and custom OpenAI-compatible endpoints. Models and presets can be switched during a conversation, making LibreChat a provider-neutral front end for choosing inference based on capability, cost, or deployment requirements.

Sources: [1]

Treat Agents, MCP, Skills, and code execution as first-class parts of the same platform

LibreChat Agents can combine MCP servers, tools, file search, code execution, Skills, and Subagents. The README also documents an Agent Marketplace, sharing, an Agent Management API, and experimental attached code workspaces, moving the project from simple chat toward building, governing, and operating reusable agents for multiple users.

Sources: [1]

Bundle authentication, search, RAG, administration, and observability into a Docker-based self-hosted stack

The official Local Installation guide recommends Docker Compose and bundles MongoDB, MeiliSearch, the RAG API, and a vector database. LibreChat also includes multi-user authentication, role and group management, an Admin Panel, resumable streams, and OpenTelemetry or Langfuse integrations, making it suitable for more than a single-user AI front end.

Sources: [2][1]

Best for teams that want to control the AI entry point without locking the interface to one model vendor

LibreChat fits teams that want one interface across commercial and local models, want MCP, RAG, or code execution configured centrally instead of per-user across many clients, or need a self-hosted internal AI portal with authentication and permissions. It is more operationally involved than a minimal single-user chat UI, but that tradeoff is useful when the platform itself needs to be governed.

Self-hosting the UI does not mean inference data stays local; review providers, secrets, and exposure separately

Running LibreChat on your own server does not keep prompts or attachments local when you select a cloud provider such as OpenAI or Anthropic. For a fully local path, use a local endpoint such as Ollama and review the network behavior of optional web search, RAG, Code Interpreter, and MCP integrations as well.

The official Docker path creates an .env file and the first account becomes the administrator in an unscoped single-tenant deployment. Before exposing the service, review API keys, authentication, reverse proxying, and registration policy. Apple Silicon Macs also need the documented MongoDB image override because the default image requires AVX. The repository code is MIT licensed.

Sources: [3][4][5]