from crewai import Agent, Task, Crew, Process

# 定义Agent角色
researcher = Agent(
    role="资深市场研究员",
    goal="深入分析目标市场的竞争格局和用户需求",
    backstory="你是一位有10年经验的市场分析师，擅长数据挖掘和趋势预测",
    tools=[search_tool, scrape_tool],
    llm="gpt-4.1",
    verbose=True
)

analyst = Agent(
    role="数据分析师",
    goal="将研究数据转化为可执行的洞察",
    backstory="你是一位擅长数据可视化和统计分析的专家",
    tools=[python_tool, chart_tool],
    llm="gpt-4.1"
)

writer = Agent(
    role="商业报告撰写人",
    goal="撰写结构清晰、论据充分的商业分析报告",
    backstory="你是一位资深商业咨询顾问，擅长将复杂数据转化为管理层可理解的语言",
    llm="gpt-4.1"
)

# 定义任务
research_task = Task(
    description="调研2026年中国企业AI部署市场的现状，包括市场规模、主要玩家、技术趋势",
    expected_output="一份包含市场规模数据、Top 10厂商列表、3-5个关键趋势的结构化报告",
    agent=researcher
)

analysis_task = Task(
    description="基于研究结果，分析市场机会和风险，识别目标客户画像",
    expected_output="SWOT分析矩阵和目标客户画像表",
    agent=analyst,
    context=[research_task]  # 依赖研究任务的输出
)

report_task = Task(
    description="撰写最终的商业分析报告，包含执行摘要、详细分析和行动建议",
    expected_output="一份专业的商业分析报告（约3000字）",
    agent=writer,
    context=[research_task, analysis_task]
)

# 组建团队并执行
crew = Crew(
    agents=[researcher, analyst, writer],
    tasks=[research_task, analysis_task, report_task],
    process=Process.sequential,
    verbose=True
)

result = crew.kickoff()

from autogen import AssistantAgent, UserProxyAgent, GroupChat, GroupChatManager

# 配置LLM
llm_config = {
    "model": "gpt-4.1",
    "api_key": "your-api-key",
    "temperature": 0.7
}

# 定义Agent
planner = AssistantAgent(
    name="Planner",
    system_message="""你是一个项目规划师。你的职责是：
    1. 分析用户需求，拆解为可执行的子任务
    2. 分配任务给合适的团队成员
    3. 整合各成员的输出，形成最终方案
    当你认为任务已经完成时，回复'TERMINATE'。""",
    llm_config=llm_config
)

coder = AssistantAgent(
    name="Coder",
    system_message="""你是一个高级Python开发者。你的职责是：
    1. 根据Planner的分配编写高质量代码
    2. 确保代码有完整的错误处理和文档
    3. 只编写代码，不做分析""",
    llm_config=llm_config
)

reviewer = AssistantAgent(
    name="Reviewer",
    system_message="""你是一个代码审查专家。你的职责是：
    1. 审查Coder编写的代码质量
    2. 检查是否有bug、安全漏洞、性能问题
    3. 提供改进建议""",
    llm_config=llm_config
)

# UserProxy用于执行代码
executor = UserProxyAgent(
    name="Executor",
    human_input_mode="NEVER",
    code_execution_config={"work_dir": "workspace", "use_docker": True}
)

# 创建群聊
group_chat = GroupChat(
    agents=[planner, coder, reviewer, executor],
    messages=[],
    max_round=20,
    speaker_selection_method="auto"  # 自动决定谁发言
)

manager = GroupChatManager(groupchat=group_chat, llm_config=llm_config)

# 启动对话
executor.initiate_chat(
    manager,
    message="请帮我开发一个自动化数据清洗工具，支持CSV和Excel格式，包含缺失值处理、异常值检测和数据标准化功能。"
)

from langgraph.graph import StateGraph, START, END
from langgraph.prebuilt import ToolNode
from typing import TypedDict, Annotated
import operator

# 定义状态
class ResearchState(TypedDict):
    topic: str
    research_results: Annotated[list, operator.add]  # 可追加的列表
    analysis: str
    report: str
    quality_score: float
    needs_revision: bool

# 定义节点函数
def research_node(state: ResearchState) -> dict:
    """研究节点：搜集信息"""
    topic = state["topic"]
    results = search_and_scrape(topic)
    return {"research_results": results}

def analysis_node(state: ResearchState) -> dict:
    """分析节点：深度分析研究结果"""
    analysis = llm.invoke(
        f"基于以下研究结果进行深度分析：\n{state['research_results']}"
    )
    return {"analysis": analysis.content}

def report_node(state: ResearchState) -> dict:
    """报告节点：生成最终报告"""
    report = llm.invoke(
        f"基于分析结果撰写报告：\n{state['analysis']}"
    )
    return {"report": report.content}

def quality_check_node(state: ResearchState) -> dict:
    """质量检查节点"""
    score = evaluate_quality(state["report"])
    return {
        "quality_score": score,
        "needs_revision": score < 0.8
    }

def should_revise(state: ResearchState) -> str:
    """条件路由：是否需要修订"""
    if state["needs_revision"]:
        return "revise"
    return "finish"

# 构建状态图
graph = StateGraph(ResearchState)

# 添加节点
graph.add_node("researcher", research_node)
graph.add_node("analyzer", analysis_node)
graph.add_node("writer", report_node)
graph.add_node("quality_checker", quality_check_node)

# 添加边
graph.add_edge(START, "researcher")
graph.add_edge("researcher", "analyzer")
graph.add_edge("analyzer", "writer")
graph.add_edge("writer", "quality_checker")
graph.add_conditional_edges(
    "quality_checker",
    should_revise,
    {"revise": "analyzer", "finish": END}  # 质量不达标则回到分析节点
)

# 编译并运行
app = graph.compile()
result = app.invoke({"topic": "2026年企业AI部署趋势"})

# LangGraph的Checkpoint机制：支持暂停和恢复
from langgraph.checkpoint.memory import MemorySaver

checkpointer = MemorySaver()
app = graph.compile(checkpointer=checkpointer)

# 运行并在每个节点保存状态
config = {"configurable": {"thread_id": "session-001"}}
result = app.invoke({"topic": "2026年AI趋势"}, config)

# 查看任意节点的状态
history = list(app.get_state_history(config))
for snapshot in history:
    print(f"节点: {snapshot.metadata}, 状态: {snapshot.values}")

# 混合使用示例：LangGraph编排 + CrewAI执行子任务
from langgraph.graph import StateGraph, START, END
from crewai import Crew, Agent, Task

def research_phase(state):
    """用CrewAI处理研究阶段"""
    crew = Crew(
        agents=[researcher_agent, fact_checker_agent],
        tasks=[research_task, verification_task],
        process=Process.sequential
    )
    results = crew.kickoff(inputs={"topic": state["topic"]})
    return {"research": results}

def analysis_phase(state):
    """用LangGraph原生节点处理分析"""
    analysis = llm.invoke(f"分析：{state['research']}")
    return {"analysis": analysis}

# LangGraph负责整体编排
graph = StateGraph(State)
graph.add_node("research", research_phase)
graph.add_node("analysis", analysis_phase)
graph.add_edge(START, "research")
graph.add_edge("research", "analysis")
graph.add_edge("analysis", END)