An open-source super agent harness that orchestrates sub-agents, persistent memory, and secure sandboxes for deep research and complex, multi-step tasks.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install deer-flow
2. Manual Installation
Copy the skill folder to one of these locations
~/.openclaw/skills/ <project>/skills/ Priority: Workspace > Local > Bundled
3. Prompt Installation
Copy this prompt to OpenClaw to install it automatically.
Help me install deer-flow using Clawhub. If Clawhub is not installed, install it first (npm i -g clawhub).
Prefer to download?
Get the raw skill files in a ZIP archive.
What is DeerFlow?
DeerFlow is an advanced, open-source super agent harness originally developed by ByteDance Volcengine. It is designed to coordinate complex AI agent workflows by seamlessly managing sub-agents, persistent memory systems, and sandboxed execution environments. By acting as an orchestration layer, DeerFlow empowers developers to build and deploy agents capable of handling deep research and extended, multi-step tasks.
As part of the Openclaw Skills ecosystem, this orchestration framework allows developers to extend AI agent capabilities through a highly modular and pluggable skill architecture. Whether you are building complex research pipelines or deploying autonomous agents in secure environments, DeerFlow provides the structured foundation needed to scale agent-based automation.
DeerFlow Use Cases
- Orchestrating specialized sub-agents to tackle multi-threaded or modular tasks simultaneously.
- Building deep research pipelines that require multi-step information gathering, analysis, and synthesis.
- Executing untrusted code or automated tools safely using sandboxed environments.
- Managing persistent context and memory across multiple sessions to maintain coherence in long-running agent workflows.
How DeerFlow Works
- Triggering the Workflow: The user initiates a complex task, which is received by the FastAPI backend.
- Deconstruction & Planning: DeerFlow analyzes the objective and breaks it down into structured sub-tasks.
- Sub-Agent Orchestration: Specialized sub-agents are spawned to execute specific parts of the plan.
- Execution in Sandboxes: Code execution and tool usage occur within safe, sandboxed environments.
- Context Maintenance: A persistent memory layer passes data and state between sub-agents and across execution loops.
- Aggregation: The parent agent synthesizes the results from all sub-agents and delivers the final output.
DeerFlow Setup
To get started with the DeerFlow orchestration harness, navigate to your directory and run the install script:
# Navigate to the project directory
cd deer-flow
# Run the installation command
make install
Refer to the Install.md file in the repository for detailed configuration guidelines and environment setup instructions.
DeerFlow Data Schema & Taxonomy
The DeerFlow directory structure is organized logically to separate the backend, UI, and custom capabilities:
| Directory | Description |
|---|---|
backend/ |
Python FastAPI backend handling API endpoints and orchestrations |
frontend/ |
Web User Interface for visual interaction and monitoring |
skills/ |
Directory for writing and extending custom agent skill definitions |
contracts/ |
API contracts ensuring strict communication interfaces |
docker/ |
Docker configuration files for sandboxed execution |
docs/ |
Comprehensive architectural and deployment documentation |
DeerFlow Advanced Features
- Pluggable Skill Architecture: Dynamically extend the capabilities of your agents using the highly modular layout designed for Openclaw Skills.
- Sub-Agent Spawning Patterns: Built-in templates to spin up, monitor, and tear down task-specific sub-agents.
- Robust Memory Management: Persistent storage schemas designed to prevent context loss in deep research operations.
- Sandboxed Tool Execution: Secure environment controls to execute generated code without compromising the host machine.
SKILL.md
Loading
Related Openclaw Skills
Letta (formerly MemGPT)
Letta enables developers to build stateful AI agents featuring advanced persistent memory, pre-built skills, and multi-agent systems.
The Women - AI Literature & Recovery Assistant
An interactive AI agent skill based on Kristin Hannah's novel to help users process trauma, understand veterans' experiences, and navigate complex family dynamics.
Prompt Optimizer
A versatile AI prompt engineering and optimization tool available as a web app, desktop app, Chrome extension, and self-hosted Docker container.
Grain Brain
An AI-powered nutritional and lifestyle guidance skill based on Dr. David Perlmutter's 'Grain Brain' to protect brain health, reduce inflammation, and prevent cognitive decline.
Agentic Plugin Marketplace
A comprehensive marketplace of 84 production-ready agentic plugins, 192 agents, and 156 skills designed to supercharge Claude Code, Cursor, and Openclaw Skills.
Ruflo Multi-Agent AI Harness
Ruflo is an enterprise-grade multi-agent AI harness designed to orchestrate over 100 specialized agents across distributed environments with federated communication and self-learning memory.
Featured*
