PixelRAG
The end of web parsing. The beginning of scalable pixel-native search. link: https://pixelrag.ai/
What is it?
What it is
PixelRAG is a visual retrieval-augmented generation system that renders documents (web pages, PDFs, images) as screenshots for retrieval, preserving visual structure like tables and diagrams that text-based methods lose.
Why it exists
To enable visual search and reasoning by leveraging screenshots instead of text parsing, allowing models to answer questions about visual content directly.
Who should use it
Developers building AI agents or visual search toolsResearchers in NLP/Computer VisionUsers needing document search by visual contentClaude users leveraging the pixelbrowse plugin
Who should avoid it
Users requiring only text-based searchThose without GPU access for training/embeddingProjects needing minimal setup complexity
How it works
A quick walkthrough in plain English
How PixelRAG works in 4 steps
Step 1 of 4
Something triggers the flow
A schedule, webhook, or manual click tells PixelRAG to start.
Features
Advantages
- Preserves visual structure (tables, charts, layout) that text‑only RAG loses
- Enables visual queries (image or text) for richer retrieval
- Zero‑setup hosted index available for quick experimentation
- Easy to install and use via pip or uv, with optional extras
- Custom index creation allows domain‑specific retrieval without retraining
- Fast retrieval thanks to FAISS and GPU acceleration
- Plugin integration with Claude expands use cases for agents
- Open source encourages community contributions and transparency
- Supports both CPU and GPU, making it accessible to a wide range of hardware
Disadvantages
- Large pre‑built index (~217 GB) requires significant storage
- Rendering pages can be slow and resource‑intensive, especially without GPU
- Requires Chrome/Chromium or Playwright; may fail on headless environments
- PDF rendering depends on Poppler, adding an external dependency
- Embedding step needs GPU for reasonable speed; CPU can be very slow
- Visual embeddings may miss fine‑grained textual information compared to OCR
- Training pipeline is complex and requires a separate environment
- Privacy concerns when screenshotting sensitive web content
Installation
native
pip install 'pixelrag[serve]'
# Download a pre-built index from Hugging Face. The dataset repo holds four FAISS indexes
# (base/LoRA Wikipedia pixel, Wikipedia text, news pixel); grab just the base one (~217G) here.
huggingface-cli download StarTrail-org/pixelrag-faiss-indexes \
--repo-type dataset --include "search_index_normed_v2/*" --local-dir ./index
# Serve, then query
pixelrag serve --index-dir ./index/search_index_normed_v2 --port 30001
curl -X POST http://localhost:30001/search \
-H "Content-Type: application/json" \
-d '{"queries": [{"text": "What is the capital of France?"}], "n_docs": 5}'FAQ
How do I install PixelRAG and get the pixelshot CLI on my PATH?
Install via `pip install pixelrag` (or `uv tool install pixelrag` / `pipx install pixelrag`) to make the `pixelshot` command available globally.
Can I search the pre‑built Wikipedia index without any setup?
Yes. Send a POST request to `https://api.pixelrag.ai/search` with a JSON body containing `queries` and `n_docs`; no API key or local index is required.
How do I build and serve a custom index from my own documents?
Install `pip install 'pixelrag[index]'`, create a `pixelrag.yaml` pointing to your source, run `pixelrag index build`, then `pixelrag serve --index-dir ./my_index --port 30001`.
How do I use PixelRAG as a Claude Code plugin to give Claude visual access to web pages?
Install the CLI (`uv tool install pixelrag`), add the plugin with `claude plugin marketplace add StarTrail-org/PixelRAG`, install `pixelbrowse@pixelrag-plugins`, then ask Claude to `screenshot <url>` and summarize or answer questions.
What extra dependencies are needed to render PDFs with pixelshot?
PDF rendering requires Poppler; install the extra with `pip install 'pixelrag[pdf]'` before running `pixelshot <file.pdf> -o ./tiles`.
Featured in Videos
YouTube tutorials and walkthroughs for PixelRAG
Alternatives
Similar projects ranked by category, topics, and text overlap.
