Kotaemon provides an open-source, customizable RAG UI for document-based question answering, targeting both end-users seeking a chat interface for their documents and developers building RAG pipelines. It offers a clean UI, supports various LLMs (API-based and local via Ollama/llama-cpp-python), and facilitates RAG pipeline development with features like hybrid retrieval, multi-modal QA, and advanced citations.

How It Works

Kotaemon implements a RAG pipeline with a hybrid retriever combining full-text and vector search, augmented by re-ranking for improved retrieval quality. It supports complex reasoning methods like question decomposition and agent-based reasoning (ReAct, ReWOO). The architecture is built on Gradio, allowing for a customizable UI and extensible RAG pipeline strategies, including GraphRAG indexing.

Quick Start & Requirements

• Install: Via Docker (recommended) or from source (pip install -e "libs/kotaemon[all]", pip install -e "libs/ktem").
• Prerequisites: Python >= 3.10. Docker is optional but recommended. unstructured library is needed for processing file types beyond .pdf, .html, .mhtml, and .xlsx.
• Resources: Live demos available at Hugging Face Spaces. User and Developer guides are linked.
• Docker Images: ghcr.io/cinnamon/kotaemon:main-full, ghcr.io/cinnamon/kotaemon:main-ollama, ghcr.io/cinnamon/kotaemon:main-lite. Supports linux/amd64 and linux/arm64.

Highlighted Details

• Supports multi-user login, private/public collections, and shared chats.
• Offers advanced citations with in-browser PDF preview and highlights.
• Integrates with external RAG frameworks like NanoGraphRAG and LightRAG.
• Features configurable settings UI for retrieval and generation parameters.

Maintenance & Community

The project is actively developed by "The Kotaemon Team." Feedback and contributions are welcomed via GitHub issues and a contributing guide.

Licensing & Compatibility

The project is licensed under the MIT License, permitting commercial use and integration with closed-source projects.

Limitations & Caveats

Installation of optional dependencies like unstructured or specific RAG integrations (e.g., nano-graphrag, LightRAG) may introduce Python package version conflicts, requiring manual resolution. Official MS GraphRAG indexing is limited to OpenAI or Ollama APIs.