Modular multi-agent collaboration platform — Agent.md-driven agents, skill package manager, YAML workflow engine.
模块化多 Agent 协作平台 — 让每个 Agent 成为独立个体，通过 Skill 扩展能力，通过 Workflow 定义协作
- 🧩 Modular design — Agent, Skill, and Workflow are independent, composable modules
- 📝 Agent.md driven — Each agent defined by a Markdown file (persona, skills, collaboration)
- 📦 Package manager — Install skills from local, URL, Git, or npm
- 🔄 Workflow engine — YAML-defined multi-agent collaboration flows
- 🌐 Web GUI — Dark theme UI with real-time pipeline visualization
- 🛠️ CLI — Full command-line interface for agent, skill, and workflow management
- 🐳 Docker ready — One-command deployment with Docker Compose
- Python 3.10+
- An OpenAI-compatible API key (OpenAI, Xiaomi MiMo, etc.)
git clone https://github.com/JennyMa11/agentflow.git
cd agentflow
pip install -e .git clone https://github.com/JennyMa11/agentflow.git
cd agentflow
docker compose up# Start the web server
agentflow run
# Or specify port
agentflow run --port 9000Visit http://localhost:8765
# Agent management
agentflow agent list
agentflow agent info brain
agentflow agent create my_agent --role worker --icon "🔧"
# Skill management
agentflow skill list
agentflow skill create my_skill
agentflow skill install ./skills/my_skill --agent coder
# Workflow management
agentflow workflow list
agentflow workflow run default_pipeline -d "Implement a calculator"agentflow/
├── agentflow/ # Core engine / 核心引擎
│ ├── agent.py # LLM runtime + tool loop
│ ├── agent_loader.py # Agent.md parser
│ ├── cli.py # Unified CLI entry point
│ ├── config.py # Configuration system
│ ├── models.py # Data models (Message, Task, Session, ...)
│ ├── package_manager.py # Skill package manager
│ ├── pipeline.py # Multi-agent pipeline engine
│ ├── session.py # Session persistence
│ ├── skill_loader.py # Skill tool loader
│ ├── skill_manager.py # Skill CRUD operations
│ ├── workflow_engine.py # YAML workflow engine
│ └── workspace.py # File ops & shell commands
│
├── agents/ # Agent definitions / Agent 定义
│ ├── brain/Agent.md # 🧠 Orchestrator
│ ├── coder/Agent.md # 💻 Coder
│ ├── tester/Agent.md # 🧪 Tester
│ ├── reviewer/Agent.md # 🔍 Reviewer
│ ├── debater/Agent.md # ⚔️ Debater
│ └── researcher/Agent.md # 📚 Researcher
│
├── skills/ # Skill packages / 技能包
│ ├── code_review/
│ ├── security_scan/
│ └── tdd/
│
├── workflows/ # Workflow definitions / 工作流
│ └── default_pipeline.yaml
│
├── web/ # Web GUI / Web 界面
│ ├── server.py
│ └── static/index.html
│
├── tests/ # Test suite / 测试套件
├── Dockerfile
├── docker-compose.yml
├── agentflow.yaml.example # Config template
└── pyproject.toml
Each Agent is a directory with an Agent.md:
# Agent: Coder
## Identity
- id: coder
- name: 代码工匠
- icon: 💻
- role: worker
## Model
- provider: xiaomi
- model: mimo-v2.5-pro
- temperature: 0.3
## System Prompt
You are Coder, a senior software engineer.
## Skills
- tdd
- code_generation@1.2.0
## Collaboration
### Can Call
- tester: "写完代码后请Tester测试"
### Can Be Called By
- brain: "Brain分配编码任务给我"| Agent | Role | Description |
|---|---|---|
| 🧠 Brain | orchestrator | Task planning & decomposition |
| 💻 Coder | worker | Code implementation |
| 🧪 Tester | worker | Test writing |
| 🔍 Reviewer | worker | Code review |
| ⚔️ Debater | worker | Solution debate |
| 📚 Researcher | worker | Technical research |
# Install from local
agentflow skill install ./my_skill/
# Install from URL
agentflow skill install https://example.com/skill.zip
# Install from Git
agentflow skill install github:agentflow-skills/tdd
# Install to specific agent
agentflow skill install ./skills/tdd --agent coderAgentFlow loads config from multiple sources (priority: env > project > user > defaults):
# agentflow.yaml (project root)
default_provider: openai
default_model: gpt-4
server:
port: 8765Environment variables:
export AGENTFLOW_PORT=9000
export AGENTFLOW_DEFAULT_MODEL=gpt-4-turbo
export AGENTFLOW_LOG_LEVEL=DEBUGSee agentflow.yaml.example for all options.
# Setup
pip install -e ".[dev]"
# Run tests
pytest
# Lint
ruff check agentflow/ tests/
# Type check
mypy agentflow/See CONTRIBUTING.md for detailed guidelines.
MIT License — see LICENSE
- OpenAI — API format reference
- Marked — Markdown rendering
- Highlight.js — Code highlighting