softrag

Minimal local-first multimodal Retrieval-Augmented Generation (RAG) library powered by SQLite + sqlite-vec.
Everything—documents, embeddings, cache—lives in a single .db file.

created by Julio Peixoto.

🌟 Features

• Local-first – All processing happens locally, no external services required for storage
• SQLite + sqlite-vec – Documents, embeddings, and cache in a single .db file
• Model-agnostic – Works with OpenAI, Hugging Face, Ollama, or any compatible models
• Blazing-fast – Optimized for minimal overhead and maximum throughput
• Multi-format support – PDF, DOCX, Markdown, text files, web pages, and images
• Image understanding – Uses GPT-4 Vision to analyze and describe images for semantic search
• Hybrid retrieval – Combines keyword search (FTS5) and semantic similarity
• Unified search – Query across text documents and image descriptions seamlessly

🚀 Quick Start

pip install softrag

from softrag import Rag
from langchain_openai import ChatOpenAI, OpenAIEmbeddings

# Initialize
rag = Rag(
    embed_model=OpenAIEmbeddings(model="text-embedding-3-small"),
    chat_model=ChatOpenAI(model="gpt-4o")
)

# Add different types of content
rag.add_file("document.pdf")
rag.add_web("https://example.com/article")
rag.add_image("photo.jpg")  # 🆕 Image support!

# Query across all content types
answer = rag.query("What is shown in the image and how does it relate to the document?")
print(answer)

📚 Documentation

For complete documentation, examples, and advanced usage, see: docs/softrag.md

🛠️ Next Steps

• Documentation Creation: Develop comprehensive documentation using tools like Sphinx or MkDocs to provide clear guidance on installation, usage, and contribution.
• Image Support in RAG: Integrate capabilities to handle image data, enabling the retrieval and generation of content based on visual inputs. This could involve incorporating models like CLIP for image embeddings.
• Automated Testing: Implement unit and integration tests using frameworks such as pytest to ensure code reliability and facilitate maintenance.
• Support for Multiple LLM Backends: Extend compatibility to include various language model providers, such as OpenAI, Hugging Face Transformers, and local models, offering users flexibility in choosing their preferred backend.
• Enhanced Context Retrieval: Improve the relevance of retrieved documents by integrating reranking techniques or advanced retrieval models, ensuring more accurate and contextually appropriate responses.
• Performance Benchmarking: Conduct performance evaluations to assess Softrag's efficiency and scalability, comparing it with other RAG solutions to identify areas for optimization.
• Monitoring and Logging: Implement logging mechanisms to track system operations and facilitate debugging, as well as monitoring tools to observe performance metrics and system health.

🤝 Contributing

We welcome contributions! Here's how to get started:

Development Setup

This project uses uv for dependency management. Make sure you have it installed:

# Install uv if you haven't already
curl -LsSf https://astral.sh/uv/install.sh | sh

Getting Started

1. Fork and clone the repository:

   git clone https://github.com/yourusername/softrag.git
   cd softrag
   

2. Install dependencies with uv:

   uv sync --dev
   

3. Activate the virtual environment:

   source .venv/bin/activate  # On Windows: .venv\Scripts\activate
   

Making Changes

1. Create a new branch for your feature/fix
2. Make your changes
3. Add tests if applicable
4. Ensure all tests pass
5. Submit a pull request

Project Structure

• src/softrag/ - Main library code
• docs/ - Documentation
• examples/ - Usage examples
• tests/ - Test suite

📜 License

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

Give to us your star ⭐

Developed with ❤️ for community