RAG-Anything
An all-in-one multimodal retrieval-augmented generation framework, built on LightRAG, for querying documents that mix text, images, tables and equations through a single interface.
RAG-Anything is an all-in-one, multimodal document-processing RAG framework built on LightRAG. It is designed to replace using several specialized tools by handling text, images, tables, equations, charts and other multimedia within a single framework, so a document containing mixed content types can be queried through one interface.
It covers document ingestion and parsing through to multimodal query answering, with support for PDFs, Office documents, images and other file formats. A VLM-enhanced query mode integrates a vision-language model when a document contains images, combining visual and textual context for the answer, and a context configuration module lets relevant contextual information be integrated intelligently during processing.
It is aimed at use cases such as academic research, technical documentation, financial reports and enterprise knowledge management. It is written in Python and licensed under the MIT license.