README.md

---
title: Visual Rag Toolkit
emoji: 👀
colorFrom: green
colorTo: red
sdk: docker
pinned: false
license: mit
short_description: Efficient Multi Vector Visual RAG Toolkit
---

# Visual RAG Toolkit Demo

Interactive demo for the Visual RAG Toolkit - end-to-end visual document retrieval with ColPali models.

## Features

- **📤 Upload Tab**: Index your own PDF documents with visual embeddings
- **📊 Benchmarking Tab**: Run evaluations on ViDoRe datasets (requires Qdrant)
- **🎮 Playground Tab**: Interactive search and visualization

## Environment Variables

Set these in your Space's settings:

| Variable | Description |
|----------|-------------|
| `QDRANT_URL` | Qdrant Cloud URL (e.g., `https://xxx.qdrant.io`) |
| `QDRANT_API_KEY` | Qdrant API key |

## Models

This demo supports:
- `vidore/colpali-v1.3` - Full ColPali model
- `vidore/colSmol-500M` - Smaller, faster ColSmol model

## Links

- [GitHub Repository](https://github.com/Ara-Yeroyan/visual-rag-toolkit)
- [Documentation](https://github.com/Ara-Yeroyan/visual-rag-toolkit#readme)