DeerFlow: A Deep Research Framework Powered by LLMs and Multi-Agent Systems

DeerFlow: A Deep Research Framework Powered by LLMs and Multi-Agent Systems

Summary

DeerFlow is a community-driven Deep Research framework developed by ByteDance, designed to combine language models with powerful tools for web search, crawling, and Python execution. It enables comprehensive research processes, from intelligent clarification to report generation and even podcast creation, all while giving back to the open-source community.

Repository Info

Updated on November 24, 2025
View on GitHub

Tags

agentaillmdeep-researchmulti-agentlangchainlanggraphPython

Click on any tag to explore related repositories

Introduction

DeerFlow, developed by ByteDance, is a robust, community-driven Deep Research framework that leverages the power of language models alongside specialized tools for comprehensive information gathering and analysis. It integrates capabilities such as web search, crawling, and Python code execution to facilitate in-depth research processes. The framework is built on a modular multi-agent system architecture using LangGraph, designed for automated research and code analysis, and aims to contribute significantly back to the open-source community.

Installation

Getting started with DeerFlow involves a few straightforward steps. The project is primarily developed in Python and includes a web UI built with Node.js.

Recommended Tools:

  • uv: For simplified Python environment and dependency management.
  • nvm: To manage multiple Node.js runtime versions.
  • pnpm: For installing and managing Node.js project dependencies.

Environment Requirements:

  • Python: Version 3.12+
  • Node.js: Version 22+

Quick Start Commands:

# Clone the repository
git clone https://github.com/bytedance/deer-flow.git
cd deer-flow

# Install Python dependencies (uv handles venv creation)
uv sync

# Configure .env with your API keys (e.g., Tavily, Brave Search)
cp .env.example .env

# Configure conf.yaml for your LLM model and API keys
cp conf.yaml.example conf.yaml

# Optionally, install web UI dependencies
cd deer-flow/web
pnpm install

For detailed configuration, refer to the docs/configuration_guide.md within the repository.

Examples

DeerFlow offers various examples demonstrating its capabilities in generating comprehensive research reports and interactive sessions.

Demo Replays:

Research Reports (examples from the repository):

  • OpenAI Sora Report
  • Google's Agent to Agent Protocol Report
  • What is LLM?
  • AI Adoption in Healthcare: Influencing Factors

You can run these examples or create your own research reports using the command line:

# Run with a specific query
uv run main.py "What factors are influencing AI adoption in healthcare?"

# Run in interactive mode
uv run main.py --interactive

Why Use DeerFlow?

DeerFlow stands out as a powerful framework for anyone engaged in deep research, offering a suite of features designed to enhance efficiency and accuracy:

  • Advanced LLM Integration: Supports most models via litellm and open-source models, providing a multi-tier LLM system.
  • Comprehensive Tooling: Integrates web search (Tavily, Brave Search, DuckDuckGo), crawling, and private knowledge bases (RAGFlow, Qdrant, Milvus).
  • Human-in-the-Loop Collaboration: Features intelligent clarification, interactive modification of research plans, and post-editing capabilities for reports.
  • Content Creation: Automates podcast script generation, audio synthesis, and simple PowerPoint presentations from research reports.
  • Modular Architecture: Built on LangGraph, it uses a multi-agent system (Coordinator, Planner, Research Team, Reporter) for flexible and scalable workflows.

Links