AnythingLLM

AnythingLLM is an open source, self-hosted alternative to ChatGPT that lets you chat with your own documents, run AI agents, and connect to virtually any LLM provider from one interface. It targets teams and individuals who want a private, multi-user chat application without giving up document context, agent tooling, or control over which model handles their data. The project ships as a Docker deployment for multi-user setups or a desktop app for Mac, Windows, and Linux, and it aims to run with zero setup friction locally by default.

Key features

• Multi-provider LLM support: connect OpenAI, Anthropic, Azure OpenAI, AWS Bedrock, Ollama, LM Studio, Groq, Mistral, DeepSeek, and dozens of other closed and open source providers from one workspace.
• Document chat with citations: ingest PDFs, TXT, DOCX, and other formats, then chat with source citations attached to responses.
• Built-in AI agents: agents can browse the web and perform multi-step tasks inside a workspace, with a no-code agent flow builder for custom automations and intelligent tool selection to cut token usage.
• MCP compatibility: connect Model Context Protocol servers to extend what agents can do beyond the built-in tool set.
• Multi-user support with permissioning: the Docker version supports multiple users on a single instance with per-user access control, without compromising the security of the shared instance.
• Scheduled tasks: run recurring prompts or agent tasks on a cron schedule with full agent capabilities attached.
• Embeddable chat widget: drop a customizable chat widget into any website, backed by the same workspace and documents, in the Docker version.

Self-hosting and deployment

AnythingLLM runs via Docker for production and multi-user deployments, or as a downloadable desktop app for local, single-user use with no server to manage. It ships with a native embedder and vector database by default, so a working setup needs no external services, though it supports Pinecone, Chroma, Weaviate, and other vector stores if you want to swap them in. Documents are processed through built-in ingestion pipelines that the project describes as optimized for large document sets, aiming for lower costs and faster responses than comparable chat UIs.

Ideal use cases

Teams building an internal knowledge assistant use AnythingLLM to point it at product docs, wikis, and PDFs, then chat against that context without exposing the documents to a public API. Because it works with any LLM provider, including fully local options through Ollama or LM Studio, it also fits regulated environments that need to keep prompts and documents off external servers entirely.

Tech stack

The application is written in JavaScript, with a Node.js server and a React frontend, and it exposes a full developer API for building custom integrations on top of a workspace. It's distributed under the MIT license and is also available as an embeddable widget package for adding the same chat experience to an existing website.