microsoft/autogen · error · ValueError

Run failed: {run.last_error}

Error message

Run failed: {run.last_error}

What it means

During run polling in on_messages_stream, if the Azure AI agent run reaches RunStatus.FAILED, the agent raises ValueError embedding run.last_error (the service-side error, e.g. content_filter, rate_limit, invalid tool definition). This is the service rejecting/failing the run; the message text is whatever Azure reported.

Source

Thrown at python/packages/autogen-ext/src/autogen_ext/agents/azure/_azure_ai_agent.py:719

                    thread_id=self.thread_id,
                    agent_id=self._get_agent_id,
                )
            )
        )

        # Wait for run completion by polling
        while True:
            run = await cancellation_token.link_future(
                asyncio.ensure_future(
                    self._project_client.agents.runs.get(
                        thread_id=self.thread_id,
                        run_id=run.id,
                    )
                )
            )

            if run.status == RunStatus.FAILED:
                raise ValueError(f"Run failed: {run.last_error}")

            # If the run requires action (function calls), execute tools and continue
            if run.status == RunStatus.REQUIRES_ACTION and run.required_action is not None:
                tool_calls: List[FunctionCall] = []
                submit_tool_outputs = getattr(run.required_action, "submit_tool_outputs", None)
                if submit_tool_outputs and hasattr(submit_tool_outputs, "tool_calls"):
                    for required_tool_call in submit_tool_outputs.tool_calls:
                        if required_tool_call.type == "function":
                            tool_calls.append(
                                FunctionCall(
                                    id=required_tool_call.id,
                                    name=required_tool_call.function.name,
                                    arguments=required_tool_call.function.arguments,
                                )
                            )

                # Add tool call message to inner messages
                tool_call_msg = ToolCallRequestEvent(source=self.name, content=tool_calls)

View on GitHub (pinned to 027ecf0a37)

Solutions

1. Read run.last_error in the message: content_filter → adjust prompt; rate_limit → back off and retry; invalid_request → fix the referenced tool/resource.
2. For rate limits, retry with exponential backoff and fewer parallel agents.
3. Verify the deployment exists and the project connection has access to the tools configured on the agent.
4. For content filter hits, rephrase the input or the agent instructions.

try:
    resp = await agent.on_messages(msgs, ct)
except ValueError as e:
    err = str(e)
    if "rate_limit" in err:
        await asyncio.sleep(backoff); retry = True
    elif "content_filter" in err:
        adjust_prompt(); retry = False
    else:
        raise

When it happens

Trigger: Any on_messages_stream call where the Azure run fails server-side: content policy violations, quota/rate limits, malformed tool definitions registered on the service agent, expired threads, or model deployment errors.

Common situations: Prompts tripping content filters; hitting token-per-minute quotas; a deployment that was disabled or deleted; tool definitions referencing invalid resources (e.g. bing grounding connection not configured).

Related errors

• Thread not initialized
• Agent not initialized
• Description not initialized
• Deployment name not initialized
• Instructions not initialized

AI-assisted analysis of microsoft/autogen@027ecf0a37 (2026-08-15). Data as JSON: /api/errors/ed86b6f23f4652f3. Report an issue: GitHub.