This project provides an automated, iterative deep research system designed for generating comprehensive reports on any topic. It targets researchers, students, and professionals needing in-depth analysis, offering a multi-agent architecture that continuously refines understanding and synthesizes findings.

How It Works

The system employs a multi-agent approach, with two primary modes: IterativeResearcher for shorter, continuous research loops, and DeepResearcher for more extensive projects. DeepResearcher first generates a report outline, then runs parallel IterativeResearcher instances for each section. Each IterativeResearcher identifies knowledge gaps, selects appropriate tools (like web search or crawlers), executes research actions, and synthesizes findings into a section, with a final agent compiling and proofreading the complete report.

Quick Start & Requirements

• Install via pip: pip install deep-researcher
• Requires Python and an .env file for API keys (OpenAI, Serper, etc.).
• Official documentation and examples are available in the repository.

Highlighted Details

• Supports a wide range of LLMs compatible with the OpenAI API spec, including AzureOpenAI, Anthropic, Gemini, and local models via Ollama.
• Offers both IterativeResearcher for simple queries and DeepResearcher for structured, multi-section reports.
• Allows custom LLM configuration at runtime for flexible model selection.
• Integrates with OpenAI's tracing feature for monitoring execution flow.

Maintenance & Community

• Developed by Jai Juneja at QX Labs.
• TODOs include adding unit tests, more search providers, caching, and specialized research tools.

Licensing & Compatibility

• Released under the MIT License.
• Compatible with commercial use and closed-source linking.

Limitations & Caveats

The system may encounter rate limits with frequent API calls, especially on lower-tier plans. LLMs can struggle with precise output length instructions, often producing 1-2,000 words per response despite requests for longer content. Gemini 2.0 Flash is noted to frequently fail tool calls.