What is Deep Researcher?
Deep Researcher is an AI-powered research assistant that performs iterative, deep research on any topic by combining search engines, web scraping, and large language models. The tool is designed to refine its research direction over time, diving deeper into a subject to produce comprehensive reports. It is built with simplicity in mind, keeping the codebase under 500 lines of code (LoC) for easy understanding and customization.
Key features include recursive exploration, where the agent starts with an initial query and iteratively searches for more information, following leads and narrowing down the focus. It supports multiple LLM backends, including OpenAI's o3-mini and DeepSeek's R1 via Fireworks, with automatic switching when the appropriate API key is detected. Users can also configure local LLMs by setting custom endpoints and models. The tool integrates with search engines and web scraping services like Firecrawl, allowing it to gather data from various online sources.
Benefits include time savings for researchers, students, and professionals who need to quickly gather in-depth information on complex topics. The iterative approach ensures that the final report is thorough and well-structured, covering multiple angles. Use cases range from academic research and market analysis to competitive intelligence and personal learning.
Technical details: The tool is available as a Python package (via GitHub) and can be run locally using Docker. It requires API keys for OpenAI or Fireworks (for R1), and optionally for Firecrawl (with configurable concurrency limits). The output is saved as a Markdown file (report.md or answer.md). The project is open-source under the MIT License, encouraging community contributions and modifications.
Use Cases
Key Features
iterative deep research recursive exploration multiple LLM backends automatic API switching local LLM configuration search engine integration web scraping comprehensive report generation under 500 LoCWho is it for?
researchers, students, professionals, content creators, analysts, journalists
