This project provides a simple, iterative AI research assistant that deep dives into topics by combining search engines, web scraping, and LLMs. It's designed for researchers, students, or anyone needing to explore subjects in depth, offering a <500 LoC implementation for easy understanding and extension.

How It Works

The agent takes a user query, breadth, and depth parameters. It iteratively generates search queries (SERP), processes results to extract learnings and new directions, and then recursively explores deeper based on these findings if the depth parameter is not exhausted. This approach allows for refined research paths and deep dives, culminating in a comprehensive markdown report.

Quick Start & Requirements

• Install: Clone the repo and run npm install.
• Prerequisites: Node.js, Firecrawl API key, OpenAI API key (or local LLM endpoint/model).
• Setup: Rename .env.example to .env.local and set API keys. Docker setup is also available.
• Usage: Run npm start and follow prompts for query, breadth, and depth.
• Docs: README

Highlighted Details

• Iterative research with LLM-driven query generation.
• Configurable depth and breadth parameters.
• Concurrent processing for faster execution (adjustable via CONCURRENCY_LIMIT).
• Supports custom endpoints and models (e.g., DeepSeek R1 via Fireworks API).

Maintenance & Community

• Sponsored by Aomni.
• Follow on X/Twitter mentioned for project support.

Licensing & Compatibility

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

Limitations & Caveats

Free Firecrawl API usage may encounter rate limits, requiring a reduced concurrency or paid/local Firecrawl. The project is presented as a simple implementation, implying potential for further refinement and feature expansion.