Mintplex-Labs/anything-llm

AnythingLLM – an all‑in‑one self‑hosted AI chat platform

What it is – AnythingLLM is a self‑hosted web application that lets you run a private ChatGPT‑style assistant. It bundles a React front‑end, a Node/Express back‑end, document ingestion pipelines and a set of optional Docker images so you can spin up a full‑featured LLM‑powered chat service on your own hardware or cloud.

Key capabilities

• Chat with your own data – Drag‑and‑drop PDFs, Word docs, txt files, etc. The app indexes them with a vector store and retrieves relevant passages during a conversation, showing citations.
• Multi‑model support – Connect to any LLM that works with llama.cpp (local models) or to dozens of hosted providers (OpenAI, Anthropic, Azure, Gemini, Ollama, Bedrock, etc.). You can also run a built‑in local engine powered by Lemonade + llama.cpp.
• Dynamic model routing – Define rules that automatically pick the best provider/model for a given request.
• AI agents & workflows – No‑code agent builder, scheduled cron jobs, and “intelligent tool selection” that lets the LLM call external tools (web browsing, custom APIs) while keeping token usage low.
• Multi‑user & permissioning – Docker deployments support multiple users with per‑user access controls.
• Embeddable chat widget – Export a widget you can drop into any website.
• Audio & speech – Built‑in transcription (STT) and text‑to‑speech (TTS) via browser APIs or external services like OpenAI and ElevenLabs.
• Pluggable vector DBs – Default is LanceDB, but you can swap in PGVector, Pinecone, Chroma, Weaviate, Qdrant, Milvus, Astra DB, etc.
• Telemetry (opt‑out) – Sends anonymous usage events to PostHog; can be disabled via DISABLE_TELEMETRY=true.

How you run it

• Docker – One‑click deploy buttons for Docker, Railway, Render, AWS, GCP, DigitalOcean, etc.
• Bare‑metal – Instructions for manual setup on any Linux/Unix box.
• Desktop – Pre‑built binaries for macOS, Windows, Linux.
• Mobile – Open‑source Android client in a separate repo.

Typical workflow

1. Install (Docker or desktop).
2. Add your LLM credentials or enable the local engine.
3. Upload documents; the collector parses them and stores embeddings in the chosen vector DB.
4. Open the web UI, start a chat – the system retrieves relevant chunks, optionally runs an agent, and returns a response.
5. (Optional) Create scheduled tasks or custom agents for automation.

Who it’s for – Teams or individuals who need a private, customizable chat interface that can reason over proprietary documents, run local models for data‑sensitive use cases, or experiment with AI agents without writing code.

License – MIT (open source).