NanoSage is a local, recursive search and knowledge exploration tool designed for deep research. It assists users by systematically breaking down queries, building a knowledge base from local and web data, and dynamically exploring subqueries with a focus on relevance and depth. The tool is ideal for researchers, students, and power users seeking structured, in-depth reports generated via retrieval-augmented generation (RAG) on their own hardware.

How It Works

NanoSage employs a structured, relevance-driven recursive search pipeline. It refines user queries, builds a knowledge base from local files and web searches, and tracks its exploration progress via a Table of Contents (TOC). Monte Carlo-based exploration balances search breadth and depth, ranking subqueries by relevance to maintain precision. The system then generates a detailed Markdown report using RAG, integrating insights from the most valuable findings.

Quick Start & Requirements

• Install: pip install -r requirements.txt
• Prerequisites: Python 3.8+, Ollama with Gemma 2B model (ollama pull gemma2:2b). Optional GPU acceleration requires PyTorch with CUDA.
• Run: python main.py --query "Your query here" --web_search --max_depth 2 --device cpu
• Docs: Example Report

Highlighted Details

• Recursive search with relevance-based subquery pruning.
• Integrates local files (PDFs, text) and web search results.
• Supports RAG with models like Gemma 2B for report generation.
• Generates structured Markdown reports detailing the research journey.

Maintenance & Community

• Project maintained by Foad Abo Dahood.
• BibTeX and APA citation provided for academic use.

Licensing & Compatibility

• MIT License. Permissive for commercial use and closed-source linking.

Limitations & Caveats

The system's effectiveness is dependent on the quality of the chosen retrieval model and the LLM used for summarization. Performance may vary based on hardware, especially when using CPU-only mode. Recursion depth and relevance thresholds require tuning for optimal results.