XNews

PixelRAG: Image-Based Retrieval for Web Pages Outperforms Text Parsing

PixelRAG: Image-Based Retrieval for Web Pages Outperforms Text Parsing

PixelRAG, from a Berkeley team, renders web pages and PDFs as screenshots for retrieval, preserving tables, charts, and layout that text parsing loses. It has gained over 10,000 GitHub stars and offers a pre-indexed Wikipedia dataset, plus a Claude Code plugin called pixelbrowse.

Source posts