StarTrail-org/PixelRAG ? reverse-engineered prompt

Build me a Python tool that lets me search documents by how they look, not just by extracted text. I want to be able to point it at a web page, PDF, image, or a local folder of documents, have it render them into screenshot tiles, and then search those images with either a text question or an image query. The results should show the matching screenshots so a vision model can answer questions about tables, charts, layouts, and infographics that normal text parsing would miss.

Please include a simple command line flow for rendering pages, building an index from my own files, and serving a local search API. If there is already a hosted Wikipedia index and public API in the docs, support that too so I can try it with no setup. A small browser demo or minimal web UI would be great for testing.

Also, if it fits the repo, add the Claude plugin style workflow where Claude can screenshot a page and read the image. Look up current docs online if you need to.