A lightweight multimodal RAG (Retrieval-Augmented Generation) library that uses vision AI instead of traditional embeddings or vector databases. DocPixie processes documents as images and uses vision language models for both document understanding and intelligent page selection.

• Vision-First Approach: Documents processed as images using PyMuPDF, preserving visual information and formatting
• No Vector Database Required: Eliminates the complexity of embeddings and vector storage
• Adaptive RAG Agent: Single intelligent agent that dynamically plans tasks and selects relevant pages
• Multi-Provider Support: Works with OpenAI GPT-4V, Anthropic Claude, and OpenRouter
• Modern CLI Interface: Beautiful terminal UI built with Textual
• Conversation Aware: Maintains context across multiple queries
• Pluggable Storage: Local filesystem or in-memory storage backends

# use uv (recommended)
uv pip install docpixie

# or pip
pip install docpixie

Try the CLI:

docpixie

import asyncio
from docpixie import DocPixie

async def main():
    # Initialize with your API key
    docpixie = DocPixie()

    # Add a document
    document = await docpixie.add_document("path/to/your/document.pdf")
    print(f"Added document: {document.name}")

    # Query the document
    result = await docpixie.query("What are the key findings?")
    print(f"Answer: {result.answer}")
    print(f"Pages used: {result.page_numbers}")

# Run the example
asyncio.run(main())

Start the interactive terminal interface:

docpixie

The CLI provides:

• Interactive document chat
• Document management
• Conversation history
• Model configuration
• Command palette with shortcuts

DocPixie uses environment variables for API key configuration:

# For OpenAI (default)
export OPENAI_API_KEY="your-openai-key"

# For Anthropic Claude
export ANTHROPIC_API_KEY="your-anthropic-key"

# For OpenRouter (supports many models)
export OPENROUTER_API_KEY="your-openrouter-key"

You can also specify the provider:

from docpixie import DocPixie, DocPixieConfig

config = DocPixieConfig(
    provider="anthropic",  # or "openai", "openrouter"
    model="claude-3-opus-20240229",
    vision_model="claude-3-opus-20240229"
)

docpixie = DocPixie(config=config)

• PDF files (.pdf) - Full multipage support
• More file types coming soon

DocPixie uses a clean, modular architecture:

📁 Core Components
├── 🧠 Adaptive RAG Agent - Dynamic task planning and execution
├── 👁️  Vision Processing - Document-to-image conversion via PyMuPDF
├── 🔌 Provider System - Unified interface for AI providers
├── 💾 Storage Backends - Local filesystem or in-memory storage
└── 🖥️  CLI Interface - Modern terminal UI with Textual

📁 Processing Flow
1. Document → Images (PyMuPDF)
2. Vision-based summarization
3. Adaptive query processing
4. Intelligent page selection
5. Response synthesis

• Provider-Agnostic: Generic model configuration works across all providers
• Image-Based Processing: All documents converted to images, preserving visual context
• Business Logic Separation: Raw API operations separate from workflow logic
• Adaptive Intelligence: Single agent mode that dynamically adjusts based on findings

• Research & Analysis: Query academic papers, reports, and research documents
• Document Q&A: Interactive questioning of PDFs, contracts, and manuals
• Content Discovery: Find specific information across large document collections
• Visual Document Processing: Handle documents with charts, diagrams, and complex layouts

• Getting Started Guide - Detailed examples and tutorials
• CLI Tool Guide - Complete CLI documentation

1. Fork the repository
2. Create a feature branch (git checkout -b feature/amazing-feature)
3. Commit your changes (git commit -m 'Add amazing feature')
4. Push to the branch (git push origin feature/amazing-feature)
5. Open a Pull Request

This project is licensed under the MIT License - see the LICENSE file for details.

• Built with PyMuPDF for PDF processing
• CLI powered by Textual
• Supports OpenAI, Anthropic, and OpenRouter APIs