deep-research

Open Deep Research is a lightweight, open-source AI-powered research assistant designed to perform iterative, deep research on any topic. The project aims to provide the simplest possible implementation of a deep research agent—keeping the codebase under 500 lines of code—so that developers can easily understand, modify, and build upon it. The system works by taking a user query along with configurable breadth and depth parameters. It then generates targeted search queries using a large language model, executes those queries via the Firecrawl API for web search and content extraction, and processes the results to extract key learnings and new research directions. If the depth parameter is greater than zero, the agent recursively explores deeper based on its findings, building on previous learnings and maintaining context of research goals throughout the process. The final output is a comprehensive markdown report saved as report.md or answer.md, containing all findings, sources, and references organized in a clear, readable format. Key features include iterative research with recursive exploration, intelligent query generation using LLMs, configurable depth and breadth controls, smart follow-up questions to refine research needs, concurrent processing for efficiency, and support for multiple model providers including OpenAI's o3-mini, DeepSeek R1 via Fireworks, and custom OpenAI-compatible endpoints such as OpenRouter or Gemini. The project supports both Node.js and Docker deployment, with environment variables for API keys and optional concurrency tuning. The repository is licensed under MIT and includes a community Python implementation. It is created by Duet and is intended as a minimal, extensible foundation for building deep research agents.