Berkeley Researchers Release PixelRAG, an Open-Source AI System That Searches Documents as Images to Preserve Visual Structure

Berkeley researchers unveil PixelRAG, a groundbreaking open-source AI system that searches documents as images rather than text, preserving critical visual elements like tables and charts across 8.28 million Wikipedia pages using a fine-tuned vision model and publicly available tools on Hugging Face.

• PixelRAG is an open-source retrieval-augmented generation system that renders web pages, PDFs, and documents into screenshot tiles and searches over the images directly, preserving visual structure like tables, charts, and layouts that traditional HTML parsing discards.
• The system uses a LoRA-fine-tuned Qwen3-VL-Embedding-2B model to embed page images into a searchable vector space, and ships with a live hosted API indexing 8.28M Wikipedia pages, support for FAISS and Qdrant backends, and a Claude Code plugin that gives the AI agent the ability to visually read web pages.
• Developed by researchers at Berkeley SkyLab, BAIR, and Berkeley NLP, the project is installable via pip with modular extras for rendering, embedding, indexing, and serving, and all training data, LoRA adapters, and pre-built FAISS indexes are publicly released on Hugging Face.