Skip to main content

Documentation Index

Fetch the complete documentation index at: https://mintlify.com/StarTrail-org/PixelRAG/llms.txt

Use this file to discover all available pages before exploring further.

PixelRAG is an open-source visual RAG system from Berkeley SkyLab that replaces text parsing with screenshot rendering. Instead of extracting plain text from web pages and PDFs — and losing every table, chart, and diagram in the process — PixelRAG renders each page as tiled JPEG images, embeds those images with a fine-tuned Qwen3-VL-Embedding-2B model, and retrieves them by visual similarity. A hosted Wikipedia index of 8.28 million pages is available at api.pixelrag.ai with no setup required.

Quickstart

Search the hosted Wikipedia index or build your own in minutes

How It Works

Understand the render → embed → index → search pipeline

Build an Index

Index your own documents from local files, PDFs, or URLs

REST API

Full reference for the search, status, and tile endpoints

​
What makes PixelRAG different

Text-based RAG parses HTML or PDF to text chunks and discards visual structure. A table with benchmark numbers, a chart showing a trend, an infographic — all become ambiguous prose. PixelRAG keeps those structures intact by treating pages as images from the start.

Visual Structure

Tables, charts, and layouts are preserved as rendered pixels, not mangled text

Wikipedia Index

8.28M pre-rendered Wikipedia pages ready to query with no download

Claude Plugin

Give Claude eyes: screenshot any page and read it visually

​
Get up and running

1

Install PixelRAG

pip install pixelrag
The core package includes the pixelshot renderer and the pixelrag CLI. Add [embed], [serve], or [index] extras for the ML pipeline stages.
2

Render a page to screenshot tiles

pixelshot https://en.wikipedia.org/wiki/Python --output ./tiles
Headless Chromium captures the full page as tiled JPEG images, one directory per URL.
3

Search the hosted Wikipedia index

curl -X POST https://api.pixelrag.ai/search \
  -H "Content-Type: application/json" \
  -d '{"queries": [{"text": "What is the capital of France?"}], "n_docs": 5}'
The live API serves 8.28M Wikipedia pages. No API key needed.
4

Or build your own index

pip install 'pixelrag[index]'
pixelrag index build   # uses pixelrag.yaml in your working directory
pixelrag serve --index-dir ./my_index --port 30001
Point pixelrag.yaml at local files, PDFs, or a list of URLs — PixelRAG handles the rest.

​
Explore the documentation

Installation

All install options — core, extras, and platform notes for Linux and macOS

Pipeline Overview

Render → chunk → embed → index: what each stage does and when to run it

Render Documents

Capture web pages, PDFs, and local files with pixelshot

Qdrant Backend

Use Qdrant for quantization, disk-backed vectors, and multi-server sharing

CLI Reference

Full flag reference for pixelshot and every pixelrag subcommand

Python SDK

render_url, render_urls, render_pdf — programmatic rendering API

Build docs developers (and LLMs) love

Get started for freeTalk to us