PixelRAG Plugin Brings Visual Document Search to Claude Code via Screenshot-Based RAG

PixelRAG is an open-source Visual Retrieval-Augmented Generation tool that renders web pages, PDFs, and images as screenshots instead of extracting plain text, preserving tables, charts, and layout for accurate model querying. Unlike traditional text-based RAG systems that use embeddings and vector databases on parsed text, PixelRAG feeds image tiles directly to the model, keeping visual structure intact. The tool is built from five components — covering rendering, embedding, indexing, serving, and fine-tuning — now bundled into a single pip-installable package. Claude Code users can integrate it via a dedicated plugin, with the screenshot feature running locally through Playwright at no extra cost beyond standard token usage. GPU-dependent features like embedding and fine-tuning require Linux, though Mac and Windows users can work around this via WSL.