PixelRAG

PixelRAG is an open-source search engine that renders documents and web pages to screenshots for visual retrieval-augmented generation.

PixelRAG is a pixel-native retrieval system designed for documents, web pages, and PDFs. Instead of parsing text and losing formatting, it renders pages into screenshot tiles and indexes them for visual search. It relies on a fine-tuned Qwen3-VL-Embedding model to embed page images into a retrievable space. The system includes a Python library, a Claude Code plugin, a CLI tool called pixelshot, and a FastAPI search service. It supports custom indexes with FAISS or Qdrant backends and runs locally on Linux and macOS.

PixelRAG at a glance

Best for: Software engineer, Researcher
Runs on: macOS, Linux, API, CLI, open source
Pricing: Free
Released: May 2026
Category: AI search engine
Website: github.com/StarTrail-org/PixelRAG

Key features

Render documents and web pages to screenshot tiles
Index visual content with a fine-tuned Qwen3-VL embedding model
Search documents using text or image queries
Integrate with Claude Code via the pixelbrowse plugin
Build custom indexes with FAISS or Qdrant backends
Run search APIs with FastAPI on CPU or GPU
Process PDFs and web pages using Playwright and CDP

Use it for

Search Wikipedia pages using pre-built visual indexes
Ask Claude to read tables and charts from web screenshots
Index and search local PDF documents visually
Run visual search queries using sample images

Alternatives to PixelRAG

FlashRAG — It provides pre-processed benchmark datasets and pre-implemented algorithms.
LightRAG — It uses a dual-layer architecture to reduce LLM calls and retrieval latency.
UltraRAG — Build complex RAG workflows using modular MCP components and YAML configuration.
Hister — Protects privacy with a self-hosted search index for your pages and files.
PageIndex — It provides 98.7% accuracy on FinanceBench without vector databases.
LEANN — Efficiently index millions of documents locally with 97% storage savings.
Jina — It provides reliable reader and search APIs optimized for LLMs and RAG systems.
BGE Embeddings — Provides specialized models for high-performance information retrieval.
Orama — Runs in browsers, servers, and edge networks with zero dependencies.
Chroma — Object storage architecture with automatic query-aware data tiering and caching.
USearch — It uses SIMD instructions and optimized concurrency to outperform FAISS.
AutoRAG — Federates data in place without uploading files to a vector database.

All PixelRAG alternatives