deer-flow

An open-source long-horizon SuperAgent harness that researches, codes, and creates. With the help of sandboxes, memories, tools, skill, subagents and message gateway, it handles different levels of ta

open-sourcememory-knowledge
54.8k
Stars
+35873
Stars/month
0
Releases (6m)

Star Growth

+5.8k (11.5%)
49.0k52.9k56.8kMar 27Apr 1

Overview

DeerFlow (Deep Exploration and Efficient Research Flow) is an open-source super agent harness designed to orchestrate complex AI workflows through sub-agents, memory systems, and sandboxes. The platform enables AI agents to handle long-horizon tasks including research, coding, and content creation through an extensible skills framework. Version 2.0 represents a complete ground-up rewrite with no shared code from the original version, indicating a mature evolution of the architecture. The system integrates BytePlus's InfoQuest for intelligent search and crawling capabilities, and supports multiple AI models including Doubao-Seed-2.0-Code, DeepSeek v3.2, and Kimi 2.5. Built with Python 3.12+ and Node.js 22+, DeerFlow aims to provide a comprehensive framework for building sophisticated AI agent systems that can coordinate multiple components to accomplish complex, multi-step tasks. With nearly 50,000 GitHub stars, it has gained significant community adoption and represents a notable advancement in agent orchestration technology.

Deep Analysis

Key Differentiator

vs AutoGPT: purpose-built for deep research with sub-agent orchestration and sandbox; vs LangGraph: higher-level harness with built-in memory, sandbox, and skill system rather than bare graph framework

Capabilities

  • Super agent harness with sub-agent orchestration
  • Extensible skill system
  • Sandboxed code execution
  • Long-term memory management
  • MCP server integration
  • Claude Code and Codex CLI integration
  • IM channel notifications
  • Context engineering for complex tasks

🔗 Integrations

OpenAI GPT-4/5DeepSeekKimiDoubao-SeedOpenRouterLangSmithClaude CodeCodex CLITavilyBytePlus InfoQuest

Best For

  • Deep research and exploration tasks
  • Building multi-agent systems with sub-agent orchestration
  • Teams wanting coding agent integration (Claude Code/Codex)

Not Ideal For

  • Simple single-task automation
  • Teams needing stable API with long-term backward compatibility

Languages

PythonTypeScript

Deployment

Docker (recommended)Local developmentCloud deployment

Pricing Detail

Free: Fully open-source MIT license
Paid: N/A - free; pay for LLM APIs

Known Limitations

  • v2.0 is complete rewrite - no migration from v1
  • Requires multiple API keys for full functionality
  • Complex configuration for model providers
  • Early-stage project with rapid breaking changes

Pros

  • + Comprehensive agent orchestration system that coordinates sub-agents, memory, and sandboxes for complex multi-step tasks
  • + Extensible skills framework allows customization and expansion of agent capabilities beyond basic functionality
  • + Active development with a complete 2.0 rewrite showing commitment to architectural improvements and long-term maintenance

Cons

  • - Version 2.0 is a complete rewrite with no backward compatibility, requiring migration effort for existing users
  • - Complex architecture with multiple components may require significant setup and configuration effort
  • - Limited documentation visible in the provided materials, potentially creating a steep learning curve

Use Cases

  • Automated research workflows that require gathering information from multiple sources and synthesizing findings
  • Software development projects requiring coordination between planning, coding, testing, and deployment phases
  • Content creation tasks that involve research, writing, editing, and publication across multiple platforms

Getting Started

1. Install dependencies for Python 3.12+ and Node.js 22+ environments. 2. Clone the repository and configure your preferred AI model (Doubao-Seed-2.0-Code, DeepSeek v3.2, or Kimi 2.5). 3. Visit the official website at deerflow.tech for setup documentation and run your first agent workflow.

Compare deer-flow