What it does Kotaemon is a self-hostable web UI for document Q&A via retrieval-augmented generation. Users upload files into private or public collections, pick from multiple LLM providers (OpenAI, Azure, Ollama, Groq, etc.), and chat with their documents. The whole thing is built on Gradio, which means the UI is customizable and the installation is a single Docker command or python app.py.

The interesting bit Most RAG tools hand you an answer and ask you to trust it. Kotaemon surfaces citations with relevance scores and renders them directly in an in-browser PDF viewer with highlights. It also warns when retrieved documents look weak. That transparency is rarer than it should be.

Key highlights

• Hybrid retrieval (full-text + vector) with re-ranking as the default pipeline
• Multi-modal parsing for figures and tables, not just text
• Support for complex reasoning via ReAct, ReWOO, and question decomposition
• GraphRAG integration via NanoGraphRAG, LightRAG, or Microsoft’s GraphRAG
• Multi-user with shared collections, configurable prompts, and local/private deployment options

Caveats

• GraphRAG setup involves version conflicts with hnswlib/chroma-hnswlib that require manual fixes
• Microsoft GraphRAG indexing is limited to OpenAI or Ollama; other providers need workarounds
• Full document type support (.doc, .docx, etc.) requires the larger full Docker image or installing Unstructured manually

Verdict Good fit for teams that need a transparent, self-hosted document chat with visible provenance. Skip it if you want a fully managed SaaS or a lightweight API-only backend.