AutoGen — 微软出品的对话式多 Agent 框架

Quick Start — AutoGen v0.4 AssistantAgent + RoundRobinGroupChat

v0.4 is async-first — note the asyncio.run(). RoundRobinGroupChat rotates speakers; SelectorGroupChat uses an LLM to pick the next speaker. Termination conditions are composable (TextMention, MaxMessage, FunctionCall, etc.). v0.2 (still widely used) has a different API — match your install version to docs carefully.

# pip install -U "autogen-agentchat>=0.4" "autogen-ext[openai]"
import asyncio
from autogen_agentchat.agents import AssistantAgent
from autogen_agentchat.teams import RoundRobinGroupChat
from autogen_agentchat.conditions import TextMentionTermination
from autogen_ext.models.openai import OpenAIChatCompletionClient

async def main():
    llm = OpenAIChatCompletionClient(model="gpt-4o-mini")

    planner = AssistantAgent(
        name="planner",
        model_client=llm,
        system_message="Break the user's goal into 2-3 concrete steps. Hand off to writer.",
    )
    writer = AssistantAgent(
        name="writer",
        model_client=llm,
        system_message="Execute the plan in clear prose. Say 'APPROVE' when done.",
    )
    critic = AssistantAgent(
        name="critic",
        model_client=llm,
        system_message="Find one concrete improvement or respond 'APPROVE' if good enough.",
    )

    team = RoundRobinGroupChat(
        participants=[planner, writer, critic],
        termination_condition=TextMentionTermination("APPROVE"),
        max_turns=8,
    )
    result = await team.run(task="Write a 150-word intro to multi-agent frameworks.")
    for msg in result.messages:
        print(f"[{msg.source}] {msg.content[:120]}")

asyncio.run(main())

核心能力

对比

实际用例

价格与许可

相关 TokRepo 资产

常见问题