Kotaemon
lastmod 2025-01-15created 2025-01-151 edits
An open-source RAG-based tool that lets you chat with your documents
About
- A user-friendly UI for RAG (retrieval-augmented generation) based QA
- Designed for end users who want to do QA on their documents and for developers who want to build their own RAG pipelines
- Supports various LLMs: LLM API providers such as OpenAI, AzureOpenAI, and Cohere, as well as local LLMs (ollama, llama-cpp-python)
- Easy installation: provides scripts for a quick start
- Test your own RAG pipeline with a UI built with Gradio (themes provided)
Key features
- Hosted document QA web UI:
- Supports multi-user login
- Organize documents into private/public collections
- Share and collaborate on chat history
- Supports local LLMs and popular API providers such as OpenAI, Azure, Ollama, and Groq
- By default, combines a hybrid (full-text and vector) retriever with re-ranking to ensure the best retrieval quality
- Can answer questions over multiple documents containing figures and tables
- Multimodal document parsing options can be selected in the UI
- Provides detailed citation information to ensure accuracy
- View citations and relevance scores with highlights in the in-browser PDF viewer
- Shows a warning when returning low-relevance documents
- Can answer complex/multi-hop questions through question decomposition
- Supports agent-based reasoning using ReAct, ReWOO, and other agents
- Key parts of the retrieval and generation process can be adjusted directly in the UI (e.g., prompt settings)
- Built on Gradio, so UI elements can be freely customized
- Support for various document indexing and retrieval strategies is planned. A GraphRAG indexing pipeline is provided as an example