XNews

PixelRAG: Webpage Screenshot-Based Retrieval Outperforms Text Chunking

PixelRAG: Webpage Screenshot-Based Retrieval Outperforms Text Chunking

PixelRAG from UC Berkeley renders web pages and PDFs as screenshots for retrieval, preserving tables, charts, and layout. It indexed all 8.28 million Wikipedia pages and offers a plugin for Claude Code. Mac M-series indexes a PDF in ~3 minutes without GPU.

Source posts