DeerFlow is a community-driven deep research framework that integrates large language models with specialized tools like web search, crawling, and Python execution. It aims to automate complex research tasks, generate comprehensive reports, and create multimedia content like podcasts, targeting researchers, developers, and power users seeking efficient knowledge discovery.

How It Works

DeerFlow employs a modular multi-agent system architecture built on LangGraph. A Coordinator manages the workflow, delegating tasks to a Planner that decomposes research objectives into structured execution plans. Specialized agents within a "Research Team" (Researcher for information gathering, Coder for code execution) execute these plans using tools like web search engines and Python REPLs. A Reporter agent aggregates findings and generates reports. This state-based workflow allows for flexible agent communication and task management.

Quick Start & Requirements

• Install: Clone the repository, then use uv sync to install Python dependencies. Configure .env with API keys (Tavily, Brave Search, Volcengine TTS) and conf.yaml for LLM settings. Install marp-cli for PPT generation.
• Prerequisites: Python 3.12+, Node.js 22+. Recommended tools: uv, nvm, pnpm.
• Running:
  • Console UI: uv run main.py
  • Web UI: ./bootstrap.sh -d (macOS/Linux) or bootstrap.bat -d (Windows)
• Docs: Configuration Guide

Highlighted Details

• Integrates with most LLMs via litellm, supporting open-source models and OpenAI-compatible APIs.
• Features a multi-agent architecture orchestrated by LangGraph for complex research workflows.
• Supports human-in-the-loop for plan review and modification, with auto-acceptance options.
• Enables content creation, including AI-powered podcast script generation and audio synthesis via Volcengine TTS.
• Offers a Web UI for a more interactive experience and integrates with LangSmith for tracing.

Maintenance & Community

• Core contributors include Daniel Walnut and Henry Li.
• Built upon LangChain and LangGraph frameworks.
• Community-driven with a focus on giving back to open source.

Licensing & Compatibility

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

Limitations & Caveats

The project requires API keys for several supported search engines and TTS services, which may incur costs. While it supports various LLMs via litellm, optimal performance may depend on specific model configurations.