What causes "No content in the last message" in microsoft/autogen?

Retry the turn — often transient when the service is still committing messages; add a small delay before on_messages when polling showed a just-completed run.

microsoft/autogen · error · ValueError

No content in the last message

Error message

No content in the last message

What it means

The assistant's last message was found but its content attribute is empty/None, so there is no text to turn into a chat message. The agent refuses to return an empty response and raises rather than emitting a blank TextMessage.

Source

Thrown at python/packages/autogen-ext/src/autogen_ext/agents/azure/_azure_ai_agent.py:810

        ):
            if cancellation_token.is_cancelled():
                trace_logger.debug("Message retrieval cancelled by token.")
                break
            agent_messages.append(msg)
            if len(agent_messages) >= message_limit:
                break
        if not agent_messages:
            raise ValueError("No messages received from assistant")

        # Get the last message from the agent (role=AGENT)
        last_message: Optional[ThreadMessage] = next(
            (m for m in agent_messages if getattr(m, "role", None) == "agent"), None
        )
        if not last_message:
            trace_logger.debug("No message with AGENT role found, falling back to first message")
            last_message = agent_messages[0]  # Fallback to first message
        if not getattr(last_message, "content", None):
            raise ValueError("No content in the last message")

        # Extract text content
        message_text = ""
        for text_message in last_message.text_messages:
            message_text += text_message.text.value

        # Extract citations
        citations: list[Any] = []

        # Try accessing annotations directly

        annotations = getattr(last_message, "annotations", [])

        if isinstance(annotations, list) and annotations:
            annotations = cast(List[MessageTextUrlCitationAnnotation], annotations)

            trace_logger.debug(f"Found {len(annotations)} annotations")
            for annotation in annotations:

View on GitHub (pinned to 027ecf0a37)

Solutions

Retry the turn — often transient when the service is still committing messages; add a small delay before on_messages when polling showed a just-completed run.
Inspect the thread via the raw SDK to see the actual message shapes and confirm the run truly produced text.
If the run only performed tool calls, send a follow-up turn so the model produces a textual reply.
Upgrade autogen-ext; last-message selection logic has evolved.

Defensive patterns

Strategy: retry

Try / catch

try:
    resp = await agent.on_messages(msgs, ct)
except ValueError as e:
    if "No content in the last message" in str(e):
        await asyncio.sleep(1)
        resp = await agent.on_messages(msgs, ct)  # one retry
    else:
        raise

Prevention

Add a small delay between run completion and message retrieval.
If the run only made tool calls, send a follow-up turn to elicit a textual reply.
Inspect thread messages with the raw SDK when this recurs.

When it happens

Trigger: A thread message whose content is an empty list/None — e.g. a message that only carried tool_call metadata, or a message partially deleted; the role-based fallback picks agent_messages[0] which may also be contentless.

Common situations: Runs that end immediately after tool calls (the tool-call message has no content); message truncation on the service side; racing with deletion/cleanup of thread messages.

Related errors

AI-assisted analysis of microsoft/autogen@027ecf0a37 (2026-08-15). Data as JSON: /api/errors/f611a1bb5b7f66ae. Report an issue: GitHub.