Open Webui | SelectorGroupChat: No reply from team after initial message | Sending message of type GroupChatStart to SelectorGroupChatManager | Connection error - MessageHandlerException

[cma2t3r]

What happened?

Describe the bug
2 issues:

• "exception": "Connection error.", "type": "MessageHandlerException" -- during response w/o SingleThreadedAgentRuntime()
• autogen_core:Sending message of type GroupChatStart to SelectorGroupChatManager -- doesn't invoke the team at all

Using an existing/working AG[0.7.5] project to make an Open Webui plugin; I got the connection error, then the GroupChatStart issue while testing custom AgentRuntime and SelectorGroupChatManager classes during debugging.

'''

class MakeModelClient:

def __init__(
    self,
    model: str = "gpt-oss-20b-uncensored",
    model_info: dict = model_info_defaults,
    llm_url: str = None,
    llm_key: str = None,
    **kwargs,
):

    self.model = model
    self.model_info = model_info
    self.kwargs = kwargs
    self.llm_url = llm_url
    self.llm_key = llm_key

def deploy(self) -> OpenAIChatCompletionClient:

    model_client = OpenAIChatCompletionClient(
        api_key=self.llm_key,
        base_url=self.llm_url,
        model=self.model,
        temperature=0,
        seed=45,
        http_client=httpx.AsyncClient(
            timeout=120,
            verify=False,
            http2=False,
        ),
        model_info=self.model_info,
        **self.kwargs,
    )
    return model_client

class ResearchTeam:

def __init__(
    self,
    docker: DockerCommandLineCodeExecutor = None,
    model: str = "ibm/granite-4-h-tiny",
    model_research: str = None,
    model_library: str = None,
    model_science: str = None,
    model_report: str = None,
    termination_condition: str = None,
    max_selector_attempts: int = 3,
    allow_repeated_speaker: bool = True,
    selector_prompt: str = None,
    librarian_agent_name: str = None,
    researcher_agent_name: str = None,
    research_scientist_agent_name: str = None,
    code_executor_agent_name: str = None,
    llm_url: str = None,
    llm_key: str = None,
    **kwargs,
):

    self.docker = docker
    self.model = model
    self.model_research = model_research
    self.model_library = model_library
    self.model_science = model_science
    self.model_report = model_report

    self.max_selector_attempts = max_selector_attempts
    self.termination_condition = termination_condition
    self.allow_repeated_speaker = allow_repeated_speaker
    self.selector_prompt = selector_prompt
    self.librarian_agent_name = librarian_agent_name
    self.researcher_agent_name = researcher_agent_name
    self.research_scientist_agent_name = research_scientist_agent_name
    self.code_executor_agent_name = code_executor_agent_name

    self.llm_url = llm_url
    self.llm_key = llm_key
    self.kwargs = kwargs

async def deploy(
    self, history: Optional[UnboundedChatCompletionContext] = None
) -> SelectorGroupChat:

    self.model_client = MakeModelClient(
        self.model, llm_url=self.llm_url, llm_key=self.llm_key
    ).deploy()
    self.librarian_client = (
        MakeModelClient(
            self.model_library, llm_url=self.llm_url, llm_key=self.llm_key
        ).deploy()
        if self.model_library
        else self.model_client
    )
    self.researcher_client = (
        MakeModelClient(
            self.model_research, llm_url=self.llm_url, llm_key=self.llm_key
        ).deploy()
        if self.model_research
        else self.model_client
    )
    self.research_scientist_client = (
        MakeModelClient(
            self.model_science, llm_url=self.llm_url, llm_key=self.llm_key
        ).deploy()
        if self.model_science
        else self.model_client
    )
    self.report_writer_client = (
        MakeModelClient(
            self.model_report, llm_url=self.llm_url, llm_key=self.llm_key
        ).deploy()
        if self.model_report
        else self.model_client
    )

    self.orchestrator_agent = MakeOrchestratorAgent(self.model_client).deploy()
    self.librarian_agent = await MakeLibrarianAgent(self.librarian_client).deploy()

    self.researcher_agent = await MakeResearcherAgent(
        self.researcher_client
    ).deploy()

    self.research_scientist_agent = await MakeResearchScientistAgent(
        self.research_scientist_client
    ).deploy()

    team = SelectorGroupChat(
        name="Research_Team",
        max_selector_attempts=self.max_selector_attempts,
        model_client=self.model_client,
        termination_condition=self.termination_condition,
        selector_prompt=self.selector_prompt,
        allow_repeated_speaker=self.allow_repeated_speaker,
        selector_func=None,
        model_context=history,
        participants=[
            self.orchestrator_agent,
            self.research_scientist_agent,
            self.researcher_agent,
        ],
        **self.kwargs,
    )
    return team

'''

To Reproduce

.... .... later in the Pipe / Pipeline classes ... ...

self.team = await self._team.deploy()
result = await self.team.run(task=msg)
reply = result.messages[-1].content

---

issue with SingleThreadedAgentRuntime():

payload={"messages":[{"id":"e24dd84c-cbc5-49d5-abff-97e6a509f781","source":"user","models_usage":null,"metadata":{},"created_at":"2025-11-17T21:35:01.782944Z","content":"What are 5 creative things I could do with my kids' art? I don't want to throw them away, but it's also so much clutter.","role":"user","type":"TextMessage"}],"output_task_messages":true} sender=null receiver=SelectorGroupChatManager_b7bb7017-1708-43fd-8cd9-70c4a2455447/b7bb7017-1708-43fd-8cd9-70c4a2455447 kind=MessageKind.DIRECT delivery_stage=DeliveryStage.SEND type=Message

INFO:autogen_core:Sending message of type GroupChatStart to SelectorGroupChatManager_b7bb7017-1708-43fd-8cd9-70c4a2455447: {'messages': [TextMessage(id='e24dd84c-cbc5-49d5-abff-97e6a509f781', source='user', models_usage=None, metadata={}, created_at=datetime.datetime(2025, 11, 17, 21, 35, 1, 782944, tzinfo=datetime.timezone.utc), content="What are 5 creative things I could do with my kids' art? I don't want to throw them away, but it's also so much clutter.", role='user', type='TextMessage')], 'output_task_messages': True}
INFO:autogen_core:Calling message handler for SelectorGroupChatManager_b7bb7017-1708-43fd-8cd9-70c4a2455447/b7bb7017-1708-43fd-8cd9-70c4a2455447 with message type GroupChatStart sent by Unknown

---

connection error w/o SingleThreadedAgentRuntime():

INFO:openai._base_client:Retrying request to /chat/completions in 0.936063 seconds
payload={"messages":[{"id":"e24dd84c-cbc5-49d5-abff-97e6a509f781","source":"user","models_usage":null,"metadata":{},"created_at":"2025-11-17T21:35:01.782944Z","content":"What are 5 creative things I could do with my kids' art? I don't want to throw them away, but it's also so much clutter.","role":"user","type":"TextMessage"}],"output_task_messages":true} handling_agent=SelectorGroupChatManager_b7bb7017-1708-43fd-8cd9-70c4a2455447/b7bb7017-1708-43fd-8cd9-70c4a2455447 exception=Connection error. type=MessageHandlerException

Expected behavior
Invoke the team from an Open Webui Pipe / Pipeline integration; expecting a string as a return value.
Both the Pipe and Pipeline display the same behavior; either can be a good way to implement Autogen.

Screenshots

Additional context
Open Webui Pipes docs
https://docs.openwebui.com/features/plugin/functions/pipe/

Open Webui Pipelines docs
https://docs.openwebui.com/features/pipelines/

Which packages was the bug in?

Python AgentChat (autogen-agentchat>=0.4.0)

AutoGen library version.

Python 0.7.5

Other library version.

No response

Model used

ibm/granite-4-h-tiny

Model provider

Other (please specify below)

Other model provider

LMStudio / OpenAI

Python version

3.11

.NET version

None

Operating system

Ubuntu