HKUDS/RAG-Anything ? reverse-engineered prompt

I want to build an advanced AI tool for "chatting with my documents," but with a major upgrade. The documents I work with—like PDFs, research papers, and reports—are full of more than just text. They have important images, charts, data tables, and even mathematical formulas mixed in.

I need a system that acts as an all-in-one framework to handle this. It should be able to ingest and process these entire complex documents, understanding all the different types of content. When I ask a question, the AI should be able to pull context from the text, a related image, and a data table to give a truly comprehensive answer. The goal is a single, unified system that can query this mixed-media knowledge base without needing specialized tools for each content type. Use Python for this.