主流 Agent 开发框架集成指南

本文档向您展示如何将不同 Agent 开发框架开发的 Agent 程序接入到 PPIO Agent Runtime 中。提供了多个主流 AI Agent 开发框架的集成示例,包括:

LangGraph - 构建复杂的多步骤 Agent 工作流
OpenAI Agents SDK - OpenAI 官方 Agent 框架
Microsoft AutoGen - 微软的多 Agent 对话框架
Google ADK - Google Agent Development Kit

核心集成模式

所有框架的集成都遵循以下核心模式:

from ppio_sandbox.agent_runtime import AgentRuntimeApp

# 1. 创建 Agent Runtime 应用实例
app = AgentRuntimeApp()

# 2. 初始化你的 Agent 框架

# 3. 使用装饰器定义入口点
@app.entrypoint
def agent_invocation(request: dict) -> dict:
    """
    Args:
        request: 请求数据,通常包含 prompt 等字段
    Returns:
        响应数据字典
    """
    prompt = request.get("prompt", "")
    
    # 调用你的 Agent 框架
    result = your_agent.run(prompt)
    
    return {"result": result}

# 4. 运行应用
if __name__ == "__main__":
    app.run()

LangGraph

LangGraph 是 LangChain 官方推出的用于构建有状态、多步骤 Agent 应用的框架。

示例代码

完整示例项目请参考这里

from langchain_community.chat_models import ChatOpenAI
from typing_extensions import TypedDict
from typing import Annotated
from langgraph.graph import StateGraph, START
from langgraph.graph.message import add_messages
from langgraph.prebuilt import ToolNode, tools_condition

# 导入 PPIO Agent Runtime
from ppio_sandbox.agent_runtime import AgentRuntimeApp

app = AgentRuntimeApp()

# 定义状态
class State(TypedDict):
    messages: Annotated[list, add_messages]

# 初始化 LLM
llm = ChatOpenAI(model="gpt-4")

# 定义工具
def get_weather(location: str) -> str:
    """获取指定位置的天气信息"""
    return f"{location} 的天气是晴天,温度 23°C"

tools = [get_weather]
llm_with_tools = llm.bind_tools(tools)

# 定义节点函数
def chatbot(state: State):
    return {"messages": [llm_with_tools.invoke(state["messages"])]}

# 构建图
graph_builder = StateGraph(State)
graph_builder.add_node("chatbot", chatbot)
graph_builder.add_node("tools", ToolNode(tools=tools))
graph_builder.add_conditional_edges(
    "chatbot",
    tools_condition,
)
graph_builder.add_edge("tools", "chatbot")
graph_builder.add_edge(START, "chatbot")

graph = graph_builder.compile()

# 定义入口点
@app.entrypoint
def agent_invocation(request: dict) -> dict:
    """LangGraph Agent 入口点"""
    prompt = request.get("prompt", "你好!")
    
    # 调用 LangGraph
    result = graph.invoke({
        "messages": [{"role": "user", "content": prompt}]
    })
    
    # 提取最后一条消息
    final_message = result['messages'][-1].content
    
    return {"result": final_message}

if __name__ == "__main__":
    app.run()

OpenAI Agents SDK

OpenAI Agents SDK 是 OpenAI 官方提供的构建 AI Agent 的工具包。

示例代码

完整示例项目请参考这里

import os
from ppio_sandbox.agent_runtime import AgentRuntimeApp
from openai import AsyncOpenAI

# 创建应用
app = AgentRuntimeApp()

# 定义工具函数
def get_weather(city: str) -> str:
    """获取城市天气（模拟）"""
    weather_data = {
        "北京": "晴天，15°C",
        "上海": "多云，20°C",
        "深圳": "小雨，25°C",
    }
    return weather_data.get(city, f"{city}：晴天，23°C")

# 工具定义
TOOLS = [
    {
        "type": "function",
        "function": {
            "name": "get_weather",
            "description": "获取指定城市的天气信息",
            "parameters": {
                "type": "object",
                "properties": {
                    "city": {
                        "type": "string",
                        "description": "城市名称，如 '北京', '上海'"
                    }
                },
                "required": ["city"]
            }
        }
    }
]

# Agent 核心逻辑
async def run_agent(query: str) -> str:
    """运行 OpenAI Agent（支持函数调用）"""
    client = AsyncOpenAI(
        base_url=os.getenv("OPENAI_API_BASE"),
        api_key=os.getenv("PPIO_API_KEY"),
    )
    
    messages = [
        {"role": "system", "content": "你是一个有用的 AI 助手，可以查询天气。"},
        {"role": "user", "content": query}
    ]
    
    # 第一次调用
    response = await client.chat.completions.create(
        model=os.getenv("MODEL_NAME", "deepseek/deepseek-v3.1-terminus"),
        messages=messages,
        tools=TOOLS,
        tool_choice="auto"
    )
    
    response_message = response.choices[0].message
    
    # 如果需要调用工具
    if response_message.tool_calls:
        messages.append(response_message)
        
        for tool_call in response_message.tool_calls:
            function_args = eval(tool_call.function.arguments)
            function_response = get_weather(**function_args)
            
            messages.append({
                "tool_call_id": tool_call.id,
                "role": "tool",
                "name": "get_weather",
                "content": function_response
            })
        
        # 第二次调用获取最终响应
        final_response = await client.chat.completions.create(
            model=os.getenv("MODEL_NAME", "deepseek/deepseek-v3.1-terminus"),
            messages=messages
        )
        return final_response.choices[0].message.content
    
    return response_message.content

# PPIO Agent Runtime 入口点
@app.entrypoint
async def agent_invocation(request: dict) -> dict:
    """入口点函数"""
    prompt = request.get("prompt", "你好！")
    result = await run_agent(prompt)
    return {"result": result}

# 启动应用
if __name__ == "__main__":
    app.run()

Microsoft AutoGen

AutoGen 是微软开发的多 Agent 对话框架，支持多 Agent 协作和对话，可以独立运行也可以与人类一起工作。

示例代码

完整示例项目请参考这里

import os
from ppio_sandbox.agent_runtime import AgentRuntimeApp
from autogen_agentchat.agents import AssistantAgent
from autogen_agentchat.messages import TextMessage
from autogen_ext.models.openai import OpenAIChatCompletionClient
from autogen_core.models import ModelFamily, ModelInfo

# 创建应用
app = AgentRuntimeApp()

# 定义工具函数
async def get_weather(city: str) -> str:
    """获取指定城市的天气"""
    weather_data = {
        "北京": "晴天，15°C",
        "上海": "多云，20°C",
        "深圳": "小雨，25°C",
    }
    return weather_data.get(city, f"{city}：晴天，23°C")

# 创建 AutoGen Agent
def create_agent():
    """创建 AutoGen Agent"""
    model_client = OpenAIChatCompletionClient(
        base_url=os.getenv("OPENAI_BASE_URL", "https://api.ppinfra.com/v3/openai"),
        model=os.getenv("MODEL_NAME", "deepseek/deepseek-v3.1-terminus"),
        api_key=os.getenv("OPENAI_API_KEY"),
        model_info=ModelInfo(
            vision=False,
            function_calling=True,
            json_output=True,
            family=ModelFamily.UNKNOWN,
        ),
    )
    
    agent = AssistantAgent(
        name="assistant",
        model_client=model_client,
        tools=[get_weather],
        system_message="你是一个有用的 AI 助手，可以查询天气信息。",
        reflect_on_tool_use=True,
    )
    
    return agent

# 运行 Agent
async def run_agent(prompt: str) -> str:
    """运行 AutoGen Agent"""
    agent = create_agent()
    
    # 创建用户消息
    message = TextMessage(content=prompt, source="user")
    
    # 运行 Agent
    response_message = await agent.on_messages([message])
    
    # 提取响应内容
    if response_message and hasattr(response_message, 'chat_message'):
        return response_message.chat_message.content
    
    return str(response_message)

# PPIO Agent Runtime 入口点
@app.entrypoint
async def agent_invocation(request: dict):
    """入口点函数"""
    prompt = request.get("prompt", "你好！")
    result = await run_agent(prompt)
    return {"result": result}

# 启动应用
if __name__ == "__main__":
    app.run()

Google ADK

Google Agent Development Kit (ADK) 是 Google 提供的 Agent 开发工具包。

完整示例

完整示例项目请参考这里

import os
import uuid
import asyncio
from google.adk.agents import LlmAgent
from google.adk.runners import Runner
from google.adk.sessions import InMemorySessionService
from google.adk.tools import google_search
from google.genai import types
from ppio_sandbox.agent_runtime import AgentRuntimeApp

# 创建应用
app = AgentRuntimeApp()

APP_NAME = "google_search_agent"

# 创建 Google ADK Agent
root_agent = LlmAgent(
    model=os.getenv("GEMINI_MODEL", "gemini-2.0-flash"), 
    name=APP_NAME,
    instruction="我可以通过搜索互联网回答你的问题。尽管问我！",
    tools=[google_search]
)

# 创建 Session 服务和 Runner
session_service = InMemorySessionService()
runner = Runner(
    agent=root_agent, 
    app_name=APP_NAME, 
    session_service=session_service
)

# 运行 Agent
async def run_agent(query: str) -> str:
    """运行 Google ADK Agent"""
    user_id = "user_default"
    session_id = str(uuid.uuid4())
    
    # 创建 Session
    await session_service.create_session(
        app_name=APP_NAME, 
        user_id=user_id, 
        session_id=session_id
    )
    
    # 创建用户消息
    user_content = types.Content(
        role='user', 
        parts=[types.Part(text=query)]
    )
    
    # 运行 Agent
    result = ""
    async for event in runner.run_async(
        user_id=user_id, 
        session_id=session_id, 
        new_message=user_content
    ):
        if event.is_final_response() and event.content and event.content.parts:
            result = event.content.parts[0].text
    
    return result

# PPIO Agent Runtime 入口点
@app.entrypoint
def agent_invocation(request: dict):
    """入口点函数"""
    prompt = request.get("prompt", "你好！")
    result = asyncio.run(run_agent(prompt))
    return {"result": result}

# 启动应用
if __name__ == "__main__":
    app.run()

产品简介

开始使用

入门教程

已接入工具

核心集成模式

LangGraph

示例代码

OpenAI Agents SDK

示例代码

Microsoft AutoGen

示例代码

Google ADK

完整示例

产品简介

开始使用

入门教程

已接入工具

​核心集成模式

​LangGraph

​示例代码

​OpenAI Agents SDK

​示例代码

​Microsoft AutoGen

​示例代码

​Google ADK

​完整示例

核心集成模式

LangGraph

示例代码

OpenAI Agents SDK

示例代码

Microsoft AutoGen

示例代码

Google ADK

完整示例