agency-swarm-cn-git

agency-swarm

基于OpenAI Agents SDK构建的可靠多Agent编排框架。提供专业化功能用于创建、编排和管理协作式AI Agent群体，已生产级就绪。

（一）项目简介

agency-swarm填补了"单Agent demo"到"生产级多Agent系统"之间的鸿沟，让团队能够用声明式方式定义Agent角色、工具和通信流。

核心优势：

• 角色化Agent设计：CEO（协调者）、Developer（开发者）、Assistant（助手）等自定义角色，每个Agent有独立指令和行为模式
• 完全Prompt控制：开发者对每个Agent的Prompt/指令拥有完全控制权
• 类型安全工具：基于Pydantic定义工具，自动类型校验，避免运行时错误
• 编排的Agent通信：通过send_message工具实现Agent间消息传递，支持同步/异步通信
• 灵活状态持久化：支持多种状态存储后端，满足不同部署需求
• 生产级就绪：经过生产环境验证，稳定性有保障
• 多模型支持：支持OpenAI原生模型（GPT-5/4o等）和LiteLLM路由（Claude/Gemini/Grok/Azure/OpenRouter等）

（二）环境前置要求

• Python：3.12+
• 依赖：OpenAI Agents SDK（pip安装自动处理）
• API密钥：OpenAI API Key 或 LiteLLM 兼容的 API 端点

（三）快速开始/安装部署

# 安装
pip install -U agency-swarm

（四）基础使用示例

第一步：设置API密钥

import os
os.environ["OPENAI_API_KEY"] = "your-api-key"
# 或使用LiteLLM路由
os.environ["LITELLM_MODEL"] = "claude-3-5-sonnet"

第二步：定义工具

from agency_swarm import function_tool
from pydantic import Field

@function_tool
def send_email(to: str, subject: str, body: str = Field(...)) -> str:
    """发送邮件"""
    # 实现邮件发送逻辑
    return f"邮件已发送至 {to}"

第三步：定义Agent角色

from agency_swarm import Agent, Agency

ceo = Agent(
    name="CEO",
    instructions="你是公司CEO，负责协调各部门Agent完成战略任务。",
    tools=[]
)

developer = Agent(
    name="Developer",
    instructions="你是资深开发工程师，负责代码审查和技术决策。",
    tools=[send_email]
)

第四步：定义Agency并运行

agency = Agency(
   ceo=ceo,
    agents=[developer],
    shared_instructions="所有Agent共同目标是高效完成任务。"
)

# CEO向Developer委派任务
result = agency.run("帮我审查最近提交的代码")

（五）开源许可证

本项目采用 MIT 开源许可证。

🐝 Agency Swarm

Overview

The Agency Swarm is a framework for building multi-agent applications. It leverages and extends the OpenAI Agents SDK, providing specialized features for creating, orchestrating, and managing collaborative swarms of AI agents.

This framework continues the original vision of Arsenii Shatokhin (aka VRSEN) to simplify the creation of AI agencies by thinking about automation in terms of real-world organizational structures, making it intuitive for both agents and users.

Migrating from v0.x? Please see our Migration Guide for details on adapting your project to this new SDK-based version.

Key Features

• Customizable Agent Roles: Define distinct agent roles (e.g., CEO, Virtual Assistant, Developer) with tailored instructions, tools, and capabilities within the Agency Swarm framework, leveraging the underlying OpenAI Agents SDK.
• Full Control Over Prompts/Instructions: Maintain complete control over each agent’s guiding prompts (instructions) for precise behavior customization.
• Type-Safe Tools: Develop tools using Pydantic models for automatic argument validation, compatible with the OpenAI Agents SDK’s FunctionTool format.
• Orchestrated Agent Communication: Agents communicate via a dedicated send_message tool, with interactions governed by explicit, directional communication_flows defined on the Agency.
• Flexible State Persistence: Manage conversation history by providing load_threads_callback and save_threads_callback to the Agency, enabling persistence across sessions (e.g., DB/file storage).
• Multi-Agent Orchestration: Build agent workflows on the OpenAI Agents SDK foundation, enhanced by Agency Swarm’s structured orchestration layer.
• Production-Ready Focus: Built for reliability and designed for easy deployment in real-world environments.

Installation

pip install -U agency-swarm

v1.x note: The framework targets the OpenAI Agents SDK + Responses API. Migrating from v0.x? See the Migration Guide.

Compatibility

• Python: 3.12+
• Model backends:
  • OpenAI (native): GPT-5 family, GPT-4o, etc.
  • Via LiteLLM (router): Anthropic (Claude), Google (Gemini), Grok (xAI), Azure OpenAI, OpenRouter (gateway), etc.
• OS: macOS, Linux, Windows

If you hit environment issues, see the Installation guide.

Getting Started

Recommended: Start with the Agency Starter Template before you customize anything.

1. Set Your OpenAI Key:

   • Create a .env file with OPENAI_API_KEY=your_key (auto-loaded), or export it in your shell:

     export OPENAI_API_KEY="YOUR_API_KEY"
     

2. Create Tools: Define tools using the modern @function_tool decorator (recommended), or extend BaseTool (compatible):

   from agency_swarm import function_tool
   
   @function_tool
   def my_custom_tool(example_field: str) -> str:
       """A brief description of what the custom tool does."""
       return f"Result: {example_field}"
   

   or with BaseTool:

   from agency_swarm.tools import BaseTool
   from pydantic import Field
   
   class MyCustomTool(BaseTool):
       """
       A brief description of what the custom tool does.
       The docstring should clearly explain the tool's purpose and functionality.
       It will be used by the agent to determine when to use this tool.
       """
   
       # Define the fields with descriptions using Pydantic Field
       example_field: str = Field(
           ..., description="Description of the example field, explaining its purpose and usage for the Agent."
       )
   
       def run(self):
           """
           The implementation of the run method, where the tool's main functionality is executed.
           """
           # Your custom tool logic goes here
           # do_something(self.example_field)
   
           # Return the result of the tool's operation
           return "Result of MyCustomTool operation"
   

   or convert from OpenAPI schemas:

   from agency_swarm.tools import ToolFactory
   # using local file
   with open("schemas/your_schema.json") as f:
       tools = ToolFactory.from_openapi_schema(
           f.read(),
       )
   
   # using requests
   import requests
   tools = ToolFactory.from_openapi_schema(
       requests.get("https://api.example.com/openapi.json").json(),
   )
   

3. Define Agent Roles: Start by defining the roles of your agents. For example, a CEO agent for managing tasks and a developer agent for executing tasks.

   from agency_swarm import Agent, ModelSettings
   
   ceo = Agent(
       name="CEO",
       description="Responsible for client communication, task planning and management.",
       instructions="You must converse with other agents to ensure complete task execution.", # can be a file like ./instructions.md
       files_folder="./files", # files to be uploaded to OpenAI
       schemas_folder="./schemas", # OpenAPI schemas to be converted into tools
       tools=[my_custom_tool],  # FunctionTool returned by @function_tool (or adapt BaseTool via ToolFactory)
       model="gpt-5.4-mini",
       model_settings=ModelSettings(
           max_tokens=25000,
       ),
   )
   

   Working from examples:

   • Browse ./examples for runnable demos and patterns you can adapt.
   • Use the .cursorrules file at the repo root with your AI coding agent (Cursor, Claude Code, etc.).
   • Follow the Cursor IDE guide: https://agency-swarm.ai/welcome/getting-started/cursor-ide

4. Define Agency Communication Flows: Establish how your agents will communicate with each other.

   from agency_swarm import Agency
   # if importing from local files
   from Developer import Developer
   from VirtualAssistant import VirtualAssistant
   
   dev = Developer()
   va = VirtualAssistant()
   
   agency = Agency(
       ceo,  # CEO will be the entry point for communication with the user
       communication_flows=[
           ceo > dev,  # CEO can initiate communication with Developer
           ceo > va,   # CEO can initiate communication with Virtual Assistant
           dev > va    # Developer can initiate communication with Virtual Assistant
       ],
       shared_instructions='agency_manifesto.md', # shared instructions for all agents
   )
   

   In Agency Swarm, communication flows are directional. The > operator defines allowed initiations (left can initiate a chat with right).

5. Run a Demo

Web UI:

agency.copilot_demo()

Terminal:

agency.tui()

On first run, Agency Swarm sets up the terminal app automatically, shows a short setup message, and reuses it on later runs.

See the terminal workflow guide: https://agency-swarm.ai/core-framework/agencies/agent-swarm-cli

Programmatic (async):

import asyncio

async def main():
    resp = await agency.get_response("Create a project skeleton.")
    print(resp.final_output)

asyncio.run(main())

Need sync? agency.get_response_sync(...) exists, but async is recommended.

Folder Structure

Recommended agent folder structure:

/your-specified-path/
│
├── agency_manifesto.md or .txt # Agency's guiding principles (created if not present)
└── AgentName/                  # Directory for the specific agent
    ├── files/                  # Directory for files that will be uploaded to OpenAI
    ├── schemas/                # Directory for OpenAPI schemas to be converted into tools
    ├── tools/                  # Directory for tools to be imported by default.
    ├── AgentName.py            # The main agent class file
    ├── __init__.py             # Initializes the agent folder as a Python package
    ├── instructions.md or .txt # Instruction document for the agent
    └── tools.py                # Custom tools specific to the agent's role.

This structure ensures that each agent has its dedicated space with all necessary files to start working on its specific tasks. The tools.py can be customized to include tools and functionalities specific to the agent's role.

Learn More

• Installation: https://agency-swarm.ai/welcome/installation
• From Scratch guide: https://agency-swarm.ai/welcome/getting-started/from-scratch
• Cursor IDE workflow: https://agency-swarm.ai/welcome/getting-started/cursor-ide
• Tools overview: https://agency-swarm.ai/core-framework/tools/overview
• Agents overview: https://agency-swarm.ai/core-framework/agents/overview
• Agencies overview: https://agency-swarm.ai/core-framework/agencies/overview
• Communication flows: https://agency-swarm.ai/core-framework/agencies/communication-flows
• Running an agency: https://agency-swarm.ai/core-framework/agencies/running-agency
• Agent Swarm CLI: https://agency-swarm.ai/core-framework/agencies/agent-swarm-cli
• Observability: https://agency-swarm.ai/additional-features/observability

Contributing

For details on how to contribute to Agency Swarm, please refer to the Contributing Guide.

License

Agency Swarm is open-source and licensed under MIT.

Need Help?

If you need help creating custom agent swarms for your business, check out our Agents-as-a-Service subscription, or schedule a consultation with me at https://calendly.com/vrsen/ai-readiness-call