from langgraph.graph import StateGraph, END
from typing import TypedDict

class AgentState(TypedDict):
    messages: list
    next_action: str
    retry_count: int

def router(state: AgentState) -> str:
    if state["retry_count"] >= 3:
        return "escalate"
    return state["next_action"]

graph = StateGraph(AgentState)
graph.add_node("think", think_node)
graph.add_node("act", act_node)
graph.add_node("escalate", escalate_node)
graph.add_conditional_edges("think", router, {
    "act": "act", "escalate": "escalate"
})
graph.add_edge("act", "think")  # 循环回退
app = graph.compile(checkpointer=SqliteSaver(conn))

from crewai import Agent, Task, Crew, Process

researcher = Agent(
    role="Senior Research Analyst",
    goal="Uncover cutting-edge developments in AI",
    backstory="You're a veteran analyst at a top tech fund.",
    tools=[search_tool, scrape_tool],
    llm="gpt-4o",
    verbose=True
)
writer = Agent(
    role="Tech Content Strategist",
    goal="Transform analysis into compelling narratives",
    backstory="Former journalist turned AI storyteller.",
    llm="claude-sonnet-4-6",
    verbose=True
)
task1 = Task(description="Research latest multi-agent architectures",
             expected_output="5-page analysis report",
             agent=researcher)
task2 = Task(description="Write a viral LinkedIn post from the analysis",
             expected_output="LinkedIn post draft",
             agent=writer)
crew = Crew(agents=[researcher, writer],
            tasks=[task1, task2],
            process=Process.sequential)
result = crew.kickoff()

from agents import Agent, Runner, function_tool, RunContext

@function_tool
def lookup_order(ctx: RunContext, order_id: str) -> str:
    return f"Order {order_id}: shipped, ETA 2 days"

triage = Agent(name="Triage", instructions="Route to sales or support.")
sales = Agent(name="Sales", instructions="Handle purchase inquiries.",
              tools=[lookup_order])
support = Agent(name="Support", instructions="Resolve technical issues.")

triage.handoffs = [sales, support]
result = await Runner.run(triage, messages=[...])

import autogen

config_list = [{"model": "gpt-4o", "api_key": "..."}]
planner = autogen.AssistantAgent(
    "Planner", system_message="Create step-by-step plans.",
    llm_config={"config_list": config_list})
coder = autogen.AssistantAgent(
    "Coder", system_message="Write Python code to solve problems.",
    llm_config={"config_list": config_list})
reviewer = autogen.AssistantAgent(
    "Reviewer", system_message="Review code for bugs and improvements.",
    llm_config={"config_list": config_list})
groupchat = autogen.GroupChat(
    agents=[planner, coder, reviewer], messages=[], max_round=12)
manager = autogen.GroupChatManager(
    groupchat=groupchat, llm_config={"config_list": config_list})
planner.initiate_chat(manager, message="Build a web scraper")

from pydantic_ai import Agent, RunContext
from pydantic import BaseModel

class AnalysisResult(BaseModel):
    summary: str
    confidence: float
    sources: list[str]

class AppDeps:
    db: Database
    api_client: HttpClient

agent = Agent(
    'openai:gpt-4o',
    deps_type=AppDeps,
    result_type=AnalysisResult,
    system_prompt="Analyze data and return structured results."
)

@agent.tool
async def fetch_records(ctx: RunContext[AppDeps],
                        query: str) -> list[dict]:
    return await ctx.deps.db.query(query)

result = await agent.run(
    "Analyze Q1 sales trends",
    deps=AppD(db=db, api_client=client)
)
# result.data 是严格类型化的 AnalysisResult

from llama_index.core.workflow import Workflow, StartEvent, StopEvent, step

class QueryEvent(Event):
    query: str
    context: dict

class RAGWorkflow(Workflow):
    @step
    async def retrieve(self, ev: StartEvent) -> QueryEvent:
        docs = await self.retriever.retrieve(ev.query)
        return QueryEvent(query=ev.query, context={"docs": docs})

    @step
    async def synthesize(self, ev: QueryEvent) -> StopEvent:
        response = await self.synthesizer.synthesize(
            query=ev.query, context=ev.context)
        return StopEvent(result=response)

wf = RAGWorkflow(timeout=60)
result = await wf.run(query="What are the latest multi-agent patterns?")

graph TB
    subgraph "范式A：状态机图（LangGraph）"
        A1[节点：Think] -->|condition: retry<3| A2[节点：Act]
        A1 -->|condition: retry>=3| A3[节点：Escalate]
        A2 -->|loop| A1
        A3 --> A4[END]
        style A1 fill:#4A90D9,color:#fff
        style A2 fill:#7ED321,color:#fff
        style A3 fill:#D0021B,color:#fff
    end

    subgraph "范式B：角色Crew（CrewAI）"
        B1[Agent: 研究员<br/>Role: 分析师] -->|Sequential| B2[Agent: 作家<br/>Role: 内容策略]
        B2 -->|Output| B3[最终报告]
        style B1 fill:#F5A623,color:#fff
        style B2 fill:#BD10E0,color:#fff
        style B3 fill:#50E3C2,color:#fff
    end

    subgraph "范式C：Handoff链（OpenAI SDK）"
        C1[Triage Agent] -->|handoff| C2[Sales Agent]
        C1 -->|handoff| C3[Support Agent]
        C2 -->|complete| C4[END]
        C3 -->|complete| C4
        style C1 fill:#4A90D9,color:#fff
        style C2 fill:#7ED321,color:#fff
        style C3 fill:#F5A623,color:#fff
    end

quadrantChart
    title Agent Framework Strategic Positioning 2026
    x-axis Low Complexity --> High Complexity
    y-axis Low Control --> High Control
    quadrant-1 Production Heavyweight
    quadrant-2 Enterprise Suite
    quadrant-3 Quick Start
    quadrant-4 Vertical Specialist
    LangGraph: [0.85, 0.9]
    CrewAI: [0.5, 0.6]
    OpenAI SDK: [0.3, 0.4]
    AutoGen: [0.6, 0.5]
    PydanticAI: [0.4, 0.7]
    LlamaIndex: [0.7, 0.65]
    Claude SDK: [0.35, 0.75]

flowchart TD
    Start[选择Agent框架] --> Q1{需要多Agent协作?}
    Q1 -->|Yes| Q2{协作模式?}
    Q1 -->|No| Q3{核心需求?}
    Q2 -->|角色分工| CrewAI
    Q2 -->|对话协商| AutoGen
    Q2 -->|复杂状态流| Q4{需要精确控制?}
    Q4 -->|Yes| LangGraph
    Q4 -->|No| CrewAI
    Q3 -->|类型安全| PydanticAI
    Q3 -->|知识密集| LlamaIndex
    Q3 -->|锁定OpenAI| OpenAI_SDK
    Q3 -->|锁定Claude| Claude_SDK
    Q3 -->|最大灵活性| LangGraph

    style Start fill:#4A90D9,color:#fff
    style LangGraph fill:#7ED321,color:#fff
    style CrewAI fill:#F5A623,color:#fff
    style OpenAI_SDK fill:#BD10E0,color:#fff
    style AutoGen fill:#D0021B,color:#fff
    style PydanticAI fill:#50E3C2,color:#fff
    style LlamaIndex fill:#4A90D9,color:#fff
    style Claude_SDK fill:#7ED321,color:#fff