Skip to main content

Build AI agents that actually get things done

DeerFlow is an open-source super agent harness powered by LangGraph. Orchestrate sub-agents, manage memory, execute code in sandboxes, and extend capabilities with custom skills.

Get startedView documentation
Agent orchestration
Spawn sub-agents for complex multi-step tasks
Sandbox execution
Isolated Docker containers with full filesystem
Long-term memory
Persistent memory with fact extraction
Extensible skills
Progressive loading of custom capabilities

Quick start

Get DeerFlow running in minutes with Docker or local development

1

Clone the repository

Clone the DeerFlow repository and navigate to the directory:
git clone https://github.com/bytedance/deer-flow.git
cd deer-flow
2

Configure your model

Generate configuration files and set up your LLM model:
make config
Edit config.yaml to configure at least one model with your API key:
models:
  - name: gpt-4
    display_name: GPT-4
    use: langchain_openai:ChatOpenAI
    model: gpt-4
    api_key: $OPENAI_API_KEY
    max_tokens: 4096
3

Start DeerFlow

Launch all services with Docker (recommended):
make docker-init    # Pull sandbox image (first time only)
make docker-start   # Start all services
Or run locally:
make dev
Access the application at http://localhost:2026

Core features

Everything you need to build production-ready AI agents

Agent orchestration

LangGraph-based agent system with dynamic tool loading and sub-agent delegation

Sandbox execution

Isolated Docker containers with filesystem access and bash command execution

Skills system

Extensible capabilities loaded progressively to keep context lean

Long-term memory

Persistent memory with fact extraction and context summarization

MCP integration

Connect external tools via Model Context Protocol with OAuth support

Multi-model support

Use any OpenAI-compatible LLM with vision and reasoning capabilities

Explore documentation

Learn how to configure, extend, and deploy DeerFlow

Core concepts

Understand DeerFlow’s architecture, agent system, sandbox, and memory

Learn more

Setup guides

Docker deployment, local development, and custom skill creation

View guides

Python client

Embedded client for in-process agent access with streaming support

API reference

Advanced topics

Deep dive into middleware chain, thread state, and context engineering

Explore

Resources

Get help and contribute to the project

GitHub

View source code, report issues, and contribute

Contributing

Learn how to contribute to DeerFlow development

License

Open source under the MIT License

Ready to build powerful AI agents?

Get started with DeerFlow today and create agents that can research, code, and execute complex tasks autonomously

Get started