AI Agent 编排框架深度对比：LangGraph vs AutoGen vs CrewAI vs TaskMatrix

from langgraph.graph import StateGraph, END

# 定义状态
class AgentState(TypedDict):
    messages: list
    next_action: str

# 构建图
graph = StateGraph(AgentState)

# 添加节点
graph.add_node("planner", planner_agent)
graph.add_node("executor", executor_agent)
graph.add_node("critic", critic_agent)

# 添加边（DAG）
graph.add_edge("planner", "executor")
graph.add_edge("executor", "critic")
graph.add_edge("critic", END)  # 或回到 planner

# 编译
app = graph.compile()

from langgraph.graph import StateGraph, END
from langchain_openai import ChatOpenAI
from langchain.agents import load_tools

llm = ChatOpenAI(model="gpt-4o")

# Planner Agent
planner_prompt = "你是一个任务规划专家，将用户需求拆解成步骤。"
planner = {"messages": lambda state: state["messages"]} | planner_prompt | llm

# Executor Agent
executor_prompt = "你负责执行具体任务，调用必要的工具。"
executor = {"messages": lambda state: state["messages"]} | executor_prompt | llm

# 构建图
graph = StateGraph(AgentState)
graph.add_node("planner", planner)
graph.add_node("executor", executor)
graph.add_edge("planner", "executor")
graph.add_edge("executor", END)

app = graph.compile()
result = app.invoke({"messages": ["帮我写一个博客网站"]})

import autogen

# 定义 Assistant Agent
assistant = autogen.AssistantAgent(
    name="assistant",
    llm_config={"model": "gpt-4o"}
)

# 定义 User Proxy（模拟用户/执行者）
user_proxy = autogen.UserProxyAgent(
    name="user_proxy",
    human_input_mode="NEVER",  # 自动执行
    max_consecutive_auto_reply=10
)

# 发起对话
user_proxy.initiate_chat(
    assistant,
    message="帮我分析过去一周的销售数据并生成报告"
)

import autogen

# 正方 Agent
pro_agent = autogen.AssistantAgent(
    name="pro_debater",
    system_message="你是辩论赛正方，论点清晰，论据充分。",
    llm_config={"model": "gpt-4o"}
)

# 反方 Agent
con_agent = autogen.AssistantAgent(
    name="con_debater",
    system_message="你是辩论赛反方，逻辑严密，善于反驳。",
    llm_config={"model": "gpt-4o"}
)

# 裁判 Agent
judge = autogen.AssistantAgent(
    name="judge",
    system_message="你是裁判，总结双方论点并给出裁决。",
    llm_config={"model": "gpt-4o"}
)

# 辩论流程
user_proxy = autogen.UserProxyAgent(name="moderator", human_input_mode="NEVER")

user_proxy.initiate_chats([
    {"sender": user_proxy, "message": "辩题：AI是否会取代程序员", "recipient": pro_agent},
    {"sender": pro_agent, "message": "请回应反方观点", "recipient": con_agent},
    {"sender": con_agent, "message": "最后请裁判总结", "recipient": judge}
])

from crewai import Agent, Task, Crew

# 定义 Agent（角色）
researcher = Agent(
    role="市场调研员",
    goal="收集竞品信息并总结市场趋势",
    backstory="你是有10年经验的市场分析师，擅长数据洞察"
)

analyst = Agent(
    role="数据分析师",
    goal="基于调研报告提取关键洞察",
    backstory="你擅长将复杂数据转化为可执行的建议"
)

writer = Agent(
    role="内容撰写员",
    goal="将分析结果写成清晰的报告",
    backstory="你是有经验的技术写作者，文笔简洁有力"
)

# 定义任务
research_task = Task(
    description="调研 AI Agent 市场，包含主要玩家和产品",
    agent=researcher
)

analysis_task = Task(
    description="分析调研结果，给出三个核心洞察",
    agent=analyst,
    context=[research_task]  # 依赖 research_task
)

writing_task = Task(
    description="将洞察写成一份 500 字的市场报告",
    agent=writer,
    context=[analysis_task]
)

# 创建 Crew 并启动
crew = Crew(agents=[researcher, analyst, writer], tasks=[research_task, analysis_task, writing_task])
result = crew.kickoff()

from taskmatrix import Agent, TaskGraph, Tool

# 定义工具
search_tool = Tool(name="web_search", func=web_search)
code_tool = Tool(name="execute_code", func=execute_code)
doc_tool = Tool(name="read_doc", func=read_doc)

# 定义 Agent
agent = Agent(
    name="研究助手",
    tools=[search_tool, code_tool, doc_tool],
    llm_config={"model": "gpt-4o"}
)

# 定义任务图
graph = TaskGraph()
graph.add_task("问题分析", agent.analyze)
graph.add_task("信息收集", agent.search, deps=["问题分析"])
graph.add_task("代码验证", agent.code, deps=["信息收集"])
graph.add_task("生成报告", agent.report, deps=["代码验证"])

# 执行
result = graph.execute("研究 LLM 在金融风控中的应用")

crew = Crew(agents=[researcher, analyst, writer], tasks=tasks)
result = crew.kickoff()

user_proxy.initiate_chat(assistant, message="写一份AI Agent市场调研报告")

app = graph.compile()
result = app.invoke({"messages": ["写一份AI Agent市场调研报告"]})

graph = TaskGraph().add_tasks([...]).execute("AI Agent市场调研")