NanoSage Project Description

What is the project about?

NanoSage is an open-source, advanced research assistant that performs deep, recursive searches and generates comprehensive reports. It operates locally on a user's machine, leveraging small, efficient language models. It's designed to mimic the human research process, exploring related concepts and building a knowledge base from both web and local data sources.

What problem does it solve?

NanoSage addresses the challenge of conducting thorough, in-depth research efficiently. It automates the process of:

1. Refining search queries: It uses a "chain-of-thought" approach to expand on the initial user query, breaking it down into more specific subqueries.
2. Gathering relevant information: It searches both the web (using DuckDuckGo) and local files (PDFs, text files, images) to build a comprehensive knowledge base.
3. Filtering out irrelevant information: It uses embedding models to calculate the relevance of each subquery and piece of information, preventing the search from going down irrelevant paths ("rabbit holes").
4. Organizing and summarizing findings: It creates a structured Table of Contents (TOC) that represents the search path and provides summaries of the information found at each step.
5. Generating a final report: It uses Retrieval-Augmented Generation (RAG) with a language model (like Gemma 2B) to synthesize the findings into a well-organized, multi-section Markdown report.

What are the features of the project?

• Recursive Search: Explores subqueries to a specified depth, mimicking the branching nature of human research.
• Relevance Filtering: Uses embedding models (ColPali, all-minilm) to ensure that only relevant subqueries and information are pursued.
• Web and Local Search: Integrates information from both web searches (via DuckDuckGo) and local files.
• Knowledge Base: Builds a structured knowledge base of relevant documents and snippets.
• Table of Contents (TOC) Tracking: Maintains a hierarchical TOC that represents the search process and organizes the findings.
• Summarization: Provides summaries of web pages, local documents, and search results.
• RAG-based Report Generation: Uses a language model (Gemma 2B, optionally) to generate a final, comprehensive report.
• Monte Carlo Exploration: Optionally samples random subqueries to discover potentially valuable but less obvious connections.
• Configurability: Allows users to customize various parameters (search depth, relevance thresholds, retrieval models, etc.) via command-line arguments and a config.yaml file.
• GPU Acceleration: Supports GPU acceleration (via CUDA) for faster processing.
• Personality: You can specify a personality for the final report generation.

What are the technologies used in the project?

• Python 3.8+
• Ollama: For running language models locally (specifically Gemma 2B).
• PyTorch: (Optional) For GPU acceleration.
• Embedding Models: ColPali (default), all-minilm (lighter alternative) for calculating relevance.
• DuckDuckGo Search API: For web searches.
• Various Python Libraries: requirements.txt lists libraries for web scraping, text processing, embedding, and more (e.g., requests, beautifulsoup4, transformers, sentence-transformers).
• YAML: For configuration files (config.yaml).
• Markdown: For generating the final report.

What are the benefits of the project?

• Efficiency: Automates the tedious aspects of research, saving time and effort.
• Thoroughness: Explores multiple avenues of inquiry, ensuring a more comprehensive understanding of the topic.
• Relevance: Focuses on relevant information, avoiding distractions and irrelevant tangents.
• Organization: Provides a structured and organized report, making it easy to understand the findings.
• Local Operation: Runs entirely on the user's machine, ensuring privacy and control over data.
• Open Source: Allows for customization, extension, and community contributions.
• Small Models: Uses relatively small language models, making it accessible to users with limited computational resources.

What are the use cases of the project?

• Academic Research: Gathering information for papers, theses, and dissertations.
• Market Research: Investigating industry trends, competitor analysis, and customer insights.
• Technical Research: Exploring new technologies, understanding complex concepts, and troubleshooting problems.
• Content Creation: Generating outlines and drafts for articles, blog posts, and reports.
• Personal Knowledge Management: Building a personal knowledge base on topics of interest.
• General Inquiry: Answering complex questions that require gathering information from multiple sources.
• Learning: Exploring a new subject in a structured and comprehensive way.