Berkeley's PixelRAG hits 10k stars: screenshots beat text parsing for RAG retrieval
tom_doerr · x · 2026-09-19
- PixelRAG (Berkeley SkyLab & BAIR) converts web pages and PDFs into visual screenshot tiles for retrieval, preserving layout, tables, and charts that text parsing destroys.
- Backed by the paper "Web Screenshots Beat Text for Retrieval-Augmented Generation".
- Open source with 10k+ stars: full index/render/serve pipeline plus a Claude plugin and skill.
More from coding & agent
- Jev vs GLM 5.3 at Chess: Checkmated in 29 Moves but 80x Cheaper Per Move — nutlope · 2026-09-19
- 9 Use Cases for a System 1 Model: Re-ranking, Tool Pruning, Routing, Cache Decisions — udmrzn · 2026-09-19
- Devin's cloud affordances and UX details impress developer vs Codex, Cursor, Claude Code — brandon_galang · 2026-09-19
- Build self-evolving classifiers with Jev by tracking the "other" bucket — TheMoonMidas · 2026-09-19
- Jev Goes Open Source: Library Turns Any HF Model Into a Jev-Style API, Live on Featherless — Aizkmusic · 2026-09-19
- Claude Code 2.1.277 adds AGENTS.md support as fallback when CLAUDE.md is missing — pvncher · 2026-09-19