Skip to content

[Bug]: 多智能体在msghub 聊天 ,智能体记忆会包含其他智能体的思考过程 #1327

@toohandsome

Description

@toohandsome

AgentScope is an open-source project. To involve a broader community, we recommend asking your questions in English.

Describe the bug

https://doc.agentscope.io/zh_CN/tutorial/workflow_conversation.html
根据这个文档中 消息共享 章节 的示例代码

多智能体在msghub 聊天 ,智能体记忆会包含其他智能体的思考过程,正常来说 某个智能体,但是不是稳定复现,有时候包含其他智能体的思考,有时候不包含

Image

To Reproduce
Steps to reproduce the behavior:

import asyncio
import json
import os

import agentscope
from agentscope.agent import ReActAgent
from agentscope.formatter import DashScopeMultiAgentFormatter, DeepSeekChatFormatter, DeepSeekMultiAgentFormatter, \
    OpenAIMultiAgentFormatter
from agentscope.message import Msg
from agentscope.model import DashScopeChatModel, OpenAIChatModel, AnthropicChatModel
from agentscope.pipeline import MsgHub

# agentscope.init(
#     # ... other configurations like model_configs
#     studio_url="http://localhost:3000",
# )
print("xx")

Qwen3 = OpenAIChatModel(
    client_kwargs={"base_url": "https://api.scnet.cn/api/llm/v1"},
    model_name="Qwen3-235B-A22B",
    api_key="sk-xxx",
)
MiniMax = OpenAIChatModel(
    client_kwargs={"base_url": "https://api.scnet.cn/api/llm/v1"},
    model_name="MiniMax-M2.5",
    api_key="sk-xxx",
)
DeepSeek = OpenAIChatModel(
    client_kwargs={"base_url": "https://api.scnet.cn/api/llm/v1"},
    model_name="DeepSeek-R1-0528",
    api_key="sk-xxx",
)

kimi2 = OpenAIChatModel(
    client_kwargs={"base_url": "https://apis.iflow.cn/v1"},
    model_name="kimi-k2-0905",
    api_key="sk-xxx",
)

formatter = DashScopeMultiAgentFormatter()
deepseekFormatter = DeepSeekMultiAgentFormatter()
openaiFormatter = OpenAIMultiAgentFormatter()


kimi2 = ReActAgent(
    name="kimi2",
    sys_prompt="你是一个名为 kimi2 的学生。",
    model=kimi2,
    formatter=openaiFormatter,
)

alice = ReActAgent(
    name="Alice",
    sys_prompt="你是一个名为 Alice 的学生。",
    model=Qwen3,
    formatter=openaiFormatter,
)

bob = ReActAgent(
    name="Bob",
    sys_prompt="你是一个名为 Bob 的学生。",
    model=DeepSeek,
    formatter=deepseekFormatter,
)

charlie = ReActAgent(
    name="Charlie",
    sys_prompt="你是一个名为 Charlie 的学生。",
    model=MiniMax,
    formatter=openaiFormatter,
)


async def example_msghub() -> None:
    """使用 MsgHub 进行多智能体对话的示例。"""
    async with MsgHub(
            [
                kimi2,
                alice,
                bob,
                charlie
            ],
            # 进入 MsgHub 时的公告消息
            announcement=Msg(
                "system",
                "现在大家互相认识一下,简单自我介绍。",
                "system",
            ),
    ):
        await kimi2()
        print("=======================================================================================================")
        await alice()
        print("=======================================================================================================")
        await bob()
        print("=======================================================================================================")
        await charlie()


asyncio.run(example_msghub())

async def example_memory() -> None:
    """打印 Alice 的记忆。"""
    print("Alice 的记忆:")
    # alice.memory.
    for msg in await alice.memory.get_memory():
        print(
            f"{msg.name}: {json.dumps(msg.content, indent=4, ensure_ascii=False)}",
        )


asyncio.run(example_memory())

Expected behavior
A clear and concise description of what you expected to happen.

Environment (please complete the following information):

  • AgentScope Version: 1.0.17
  • Python Version: 3.11
  • OS: windows 11

Metadata

Metadata

Assignees

Labels

No labels
No labels

Type

No type

Projects

No projects

Milestone

No milestone

Relationships

None yet

Development

No branches or pull requests

Issue actions