DeerFlow

Open-source super agent harness orchestrating sub-agents, memory & sandboxes for deep research & extended tasks with extensible skills

DeerFlow — Super Agent Harness (OpenClaw)

DeerFlow (Deep Exploration and Efficient Research Flow) is an open-source super agent harness that orchestrates sub-agents, memory, and sandboxes to do almost anything — powered by extensible skills.

Source: C:\Users\Harry\Downloads\deer-flow\ Original: https://github.com/bytedance/deer-flow License: MIT

本技能基於 GitHub 上的 bytedance/deer-flow 修改與封裝。

Overview

DeerFlow 2.0 is a ground-up rewrite by ByteDance Volcengine for orchestrating complex AI agent workflows:

• Sub-agent orchestration — spawn and coordinate specialized sub-agents
• Persistent memory — maintain context and knowledge across sessions
• Sandboxed execution — safe code and tool execution
• Extensible skills — pluggable skill architecture
• Deep research — automated multi-step research pipelines

Structure

deer-flow/
├── backend/           # Python backend (FastAPI)
├── frontend/          # Web UI
├── skills/            # Extensible skill definitions
├── scripts/           # Utility scripts
├── contracts/         # API contracts
├── docker/            # Docker configuration
├── docs/              # Documentation
└── tests/             # Test suite

Usage in OpenClaw

When the user asks about deep research agents, multi-step agent workflows, or sub-agent orchestration:

1. Reference DeerFlow's skill architecture for extensibility patterns
2. Use the sub-agent spawning patterns documented in the project
3. Consult docs/ for architecture and deployment

Quick Install

cd C:\Users\Harry\Downloads\deer-flow
make install   # or follow Install.md