RAG-Anything
Multimodal RAG framework for documents with images, tables, and equations
What it is
A Python framework built on LightRAG that parses PDFs, Office files, and images, then indexes their text, figures, tables, and math into a multimodal knowledge graph for retrieval. It uses MinerU-based parsing or accepts pre-parsed content lists, and can pass retrieved images to a vision-language model at query time. Developers building question answering over mixed-content material such as research papers, technical manuals, or financial reports can use it as their retrieval layer.