Skip to content
danilo-labsPublic

About

Ultra-lightweight local RAG. Zero Docker, zero Vector DB, zero Node.js. Pure Python + BM25 + Vanilla JS talking directly to Ollama & KoboldCpp.

Topics

Resources

Code of conduct

Contributing

Security policy

Stars

0 stars

Watchers

0 watching

Forks

Repository files navigation

LocalSift

The Ultra-lightweight, Offline RAG for Low-Spec PCs. No Docker. No Vector DB. No Node.js. Just pure Python + BM25 + Vanilla JS.

License Version Python No Docker


Ask questions about your documents. Everything runs on your machine — no cloud, no accounts, no data leaving your device.

Why choose LocalSift?

  • 🪶 Featherweight: Runs on a 4GB RAM laptop. Extraction happens in memory.
  • 🔒 Privacy-First: 100% offline. Connects only to local LLMs (Ollama/KoboldCpp) via 127.0.0.1.
  • 🧠 Smart Retrieval: Uses a custom BM25 algorithm in pure JS instead of heavy Vector Databases.
  • 🎯 Relative Filtering: Automatically discards irrelevant document chunks that fall below 50% of the best BM25 score.
  • 📄 Multi-format: Reads PDF, ODT, DOCX, TXT, MD, CSV.

🚀 Quick Start

  1. Download the latest release LocalSift-v1.1.1.zip from the Releases page and extract it.
  2. Start Ollama (or KoboldCpp) on your machine.
  3. Run LocalSift:
    • Double-click LocalSift.desktop (Linux)
    • Or run bash start_localsift.sh in your terminal.
  4. Open your browser at http://127.0.0.1:8200/index.html.

🛠 How it works

  1. Upload: The Python backend extracts text from your files using standard libraries (and pypdf).
  2. Search: The frontend uses BM25 to find the most relevant paragraphs.
  3. Filter: Only the fragments that score above the relative threshold are passed to the AI.
  4. Generate: The local LLM answers using only the provided context, with strict anti-hallucination guardrails.

📋 Requirements

  • Linux (or any OS with Python 3 and a web browser)
  • Python 3.8+
  • Ollama or KoboldCpp running locally.

🤝 Contributing

Issues and pull requests are welcome! Please check the .github folder for templates. Multilingual documentation improvements are especially appreciated.

About

Ultra-lightweight local RAG. Zero Docker, zero Vector DB, zero Node.js. Pure Python + BM25 + Vanilla JS talking directly to Ollama & KoboldCpp.

Topics

Resources

Code of conduct

Contributing

Security policy

Stars

0 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages