RAG-Anything
A Python RAG framework builds searchable document collections from text, images, tables and equations, using LightRAG and vision models.
RAG-Anything is a Python framework for developers building question-answering systems over documents that mix prose with visual or structured content. Built on LightRAG, it suits research papers, technical manuals and financial reports where answers depend on figures, tables or formulas.
- Mixed document formats: Process PDFs, Office documents and images while retaining document structure and the relationships between their contents.
- Images, tables and equations: Use dedicated processors to interpret visual content, identify patterns in tables and handle mathematical expressions with LaTeX support.
- Connected document search: Combine similarity search with a knowledge graph that links text to related visual and structured elements.
- Visual question answering: Include document images alongside text in queries sent to a vision-language model.
- Existing content pipelines: Accept content you've already parsed elsewhere, or use document parsers such as MinerU, Docling and PaddleOCR.
The framework is open source under the MIT license and runs as an installable Python package, with local storage for its document index. Model connections determine where analysis happens: using cloud APIs sends text or images to those providers. It supports OpenAI-compatible connections, with examples using GPT-4o, GPT-4o-mini and OpenAI embeddings. Office document processing requires LibreOffice.
Logotyper
