Kotaemon is an open-source RAG (Retrieval-Augmented Generation) framework designed for document-based question answering. It provides both a polished chat interface for end users and a flexible pipeline architecture for developers building custom RAG solutions. With over 25,000 GitHub stars, Kotaemon has become one of the most popular open-source document QA tools available. ## Why Kotaemon Matters Document QA is one of the most practical applications of large language models, yet building a production-ready system requires integrating retrieval, ranking, parsing, and generation into a coherent pipeline. Kotaemon packages all of these components into a single deployable application with sensible defaults while maintaining full extensibility for advanced use cases. The framework takes a hybrid retrieval approach, combining full-text search with vector retrieval and re-ranking to maximize answer quality. This layered strategy outperforms either method alone, particularly on documents with mixed content types like tables, figures, and structured data. ## Multi-User Document Management Kotaemon supports multi-user authentication with SSO integration, allowing teams to organize documents into private and public collections. Users can share specific collections with colleagues, enabling collaborative knowledge bases without exposing sensitive documents. The permission system is granular enough for enterprise deployments where data access controls are critical. ## Hybrid Retrieval Pipeline The retrieval engine is Kotaemon's strongest technical component. It combines Elasticsearch or LanceDB for full-text search with ChromaDB, Milvus, or Qdrant for vector similarity, then applies a re-ranking stage to surface the most relevant passages. Developers can swap any component in the pipeline without modifying the rest of the system. Multiple indexing strategies are supported, including traditional chunking approaches and more advanced methods like GraphRAG through NanoGraphRAG and LightRAG integrations. This flexibility allows teams to optimize retrieval quality for their specific document types. ## Multi-Modal Document Processing Kotaemon handles documents containing figures, tables, and mixed content through integration with Azure Document Intelligence, Adobe PDF Extract, and Docling. The multi-modal parsing pipeline extracts structured information from complex documents, enabling the QA system to answer questions that reference charts, tables, or diagrams embedded in PDFs. ## Advanced Reasoning Beyond simple retrieval-and-answer patterns, Kotaemon supports complex reasoning methods including question decomposition and agent-based approaches using ReAct and ReWOO frameworks. These capabilities allow the system to handle multi-hop questions that require synthesizing information from multiple document sections or even multiple documents. ## Broad LLM Provider Support Kotaemon integrates with a wide range of LLM providers including OpenAI, Azure OpenAI, Ollama for local models, Cohere, Groq, and llama-cpp-python for GGUF format models. This provider flexibility means teams can choose between cloud APIs for convenience or local inference for data privacy, switching between providers through configuration changes alone. ## Deployment Flexibility The project offers multiple deployment paths: Docker images in lite, full, and Ollama-bundled variants for quick setup; direct pip installation for development environments; and cloud options through HuggingFace Spaces and Google Colab notebooks for evaluation. The Gradio-based UI is fully configurable, allowing teams to adjust retrieval parameters, generation settings, and interface layout. ## Citation and Transparency Every answer includes advanced citations with in-browser PDF viewer support and relevance scoring. Users can click through to the exact source passages that informed each response, providing the transparency needed for professional and research applications where answer provenance matters.