microsoft/autogen · error · ValueError

Specified model_client doesn't support structured output mod

Error message

Specified model_client doesn't support structured output mode.

What it means

CodeExecutorAgent requires that when retries are enabled the model must support structured (JSON) output, checked via model_client.model_info["structured_output"]. Retries depend on re-requesting the output in a structured format, so an unsupported model cannot be used with max_retries_on_error > 0.

Source

Thrown at python/packages/autogen-agentchat/src/autogen_agentchat/agents/_code_executor_agent.py:495

        if model_client is not None:
            self._model_client = model_client

        if model_context is not None:
            self._model_context = model_context
        else:
            self._model_context = UnboundedChatCompletionContext()

        self._system_messaages: List[SystemMessage] = []
        if system_message is None:
            self._system_messages = []
        else:
            self._system_messages = [SystemMessage(content=system_message)]

        if self._max_retries_on_error > 0:
            if not self._model_client or not self._model_client.model_info:
                raise ValueError("model_client.model_info must be provided when max_retries_on_error > 0")
            if not self._model_client.model_info["structured_output"]:
                raise ValueError("Specified model_client doesn't support structured output mode.")

    @property
    def produced_message_types(self) -> Sequence[type[BaseChatMessage]]:
        """The types of messages that the code executor agent produces."""
        return (TextMessage,)

    @property
    def model_context(self) -> ChatCompletionContext:
        """
        The model context in use by the agent.
        """
        return self._model_context

    async def on_messages(self, messages: Sequence[BaseChatMessage], cancellation_token: CancellationToken) -> Response:
        async for message in self.on_messages_stream(messages, cancellation_token):
            if isinstance(message, Response):
                return message
        raise AssertionError("The stream should have returned the final result.")

View on GitHub (pinned to 027ecf0a37)

Solutions

1. Set max_retries_on_error=0 (default) when the model lacks structured output support.
2. Switch to a model/client that declares structured_output=True in model_info.
3. If you control the model_info dict and the backend actually supports JSON mode (e.g. vLLM guided decoding), correct the flag.

Example fix

// before
agent = CodeExecutorAgent(
    name="coder",
    code_executor=executor,
    model_client=local_client,  # structured_output: False
    max_retries_on_error=3,
)

// after
agent = CodeExecutorAgent(
    name="coder",
    code_executor=executor,
    model_client=local_client,
    max_retries_on_error=0,
)

info = getattr(model_client, "model_info", None) or {}
if max_retries_on_error > 0 and not info.get("structured_output", False):
    max_retries_on_error = 0  # model can't support structured retry

agent = CodeExecutorAgent(name="coder", code_executor=exec_,
                          model_client=model_client,
                          max_retries_on_error=max_retries_on_error)

def supports_structured_output(client) -> bool:
    info = getattr(client, "model_info", None) or {}
    return bool(info.get("structured_output", False))

When it happens

Trigger: CodeExecutorAgent(max_retries_on_error=N) with a model whose model_info has structured_output=False (e.g., many open-weight/local models served via clients that declare no JSON-mode support).

Common situations: Using local models (llama.cpp, vLLM without guided JSON) or custom clients with hand-written model_info that sets structured_output=False; enabling retries by copy-paste.

Related errors

• Unsupported config type {config.GetType()}
• Messages should not be provided in options
• The agent name must be a valid Python identifier.
• model_client.model_info must be provided when max_retries_on
• At least one of max_total_token, max_prompt_token, or max_co

AI-assisted analysis of microsoft/autogen@027ecf0a37 (2026-08-15). Data as JSON: /api/errors/a53fe1cc750f1fb4. Report an issue: GitHub.