三个主流智能体框架（AutoGen、CrewAI、LangGraph）

贺公子之数据科学与艺术

发布于 2026-01-20 14:02:35

660

三个主流智能体框架（AutoGen、CrewAI、LangGraph）的代码实现案例，基于最新文档和社区实践整理：

AutoGen 案例：多智能体协作求解数学问题

from autogen import AssistantAgent, UserProxyAgent, config_list_from_json

# 加载LLM配置（例如OpenAI API）
config_list = config_list_from_json(env_or_file="OAI_CONFIG_LIST")

# 创建两个智能体
assistant = AssistantAgent("assistant", llm_config={"config_list": config_list})
user_proxy = UserProxyAgent("user_proxy", code_execution_config={"work_dir": "coding"})

# 注册自动回复逻辑
def ask_question(recipient, messages, sender, config):
    return True, "解方程 x^2 + 5x + 6 = 0"

user_proxy.register_reply(AssistantAgent, ask_question)

# 启动对话
user_proxy.initiate_chat(assistant, message="请解决这个数学问题")

CrewAI 案例：新闻调研任务链

from crewai import Agent, Task, Crew

# 定义智能体
researcher = Agent(
    role="资深研究员",
    goal="发现人工智能领域的最新突破",
    backstory="专注科技新闻分析的专家",
    verbose=True
)

writer = Agent(
    role="技术作家",
    goal="撰写吸引眼球的科技文章",
    backstory="擅长将复杂技术转化为通俗内容的作家"
)

# 创建任务
research_task = Task(
    description="找出2024年AI领域最重要的3项进展",
    agent=researcher
)

write_task = Task(
    description="根据研究结果撰写300字的博客文章",
    agent=writer
)

# 组建任务流
crew = Crew(
    agents=[researcher, writer],
    tasks=[research_task, write_task],
    verbose=2
)

# 执行
result = crew.kickoff()
print(result)

LangGraph 案例：循环审批工作流

from langgraph.graph import Graph
from langchain_core.messages import HumanMessage

# 定义节点
def review_document(state):
    return {"reviewer_feedback": "文档需要修改第3部分"}

def approve_document(state):
    if "approved" not in state:
        return {"decision": "reject"}
    return {"decision": "approve"}

# 构建图
workflow = Graph()
workflow.add_node("review", review_document)
workflow.add_node("decision", approve_document)
workflow.add_edge("review", "decision")

# 设置条件分支
workflow.add_conditional_edges(
    "decision",
    lambda x: "end" if x["decision"] == "approve" else "review"
)

# 编译执行
app = workflow.compile()
result = app.invoke({"document": "项目计划书.pdf"})