microsoft/semantic-kernel · error · AgentInvokeException

Failed to get a response from the agent. Please consider inc

Error message

Failed to get a response from the agent. Please consider increasing the auto invoke attempts.

What it means

Raised by get_response after the loop over maximum_auto_invoke_attempts exits without producing a final non-RETURN_CONTROL response. Each iteration where the agent requests function calls (RETURN_CONTROL events) consumes one attempt; if the agent keeps requesting functions on every turn without ever emitting a final chunk, all attempts are exhausted.

Source

Thrown at python/semantic_kernel/agents/bedrock/bedrock_agent.py:355

                        file_items = self._handle_files_event(event)
                    elif BedrockAgentEventType.TRACE in event:
                        trace_metadata = self._handle_trace_event(event)

                if not chat_message_content or not chat_message_content.content:
                    raise AgentInvokeException("Chat message content is expected but not found in the response.")

                if file_items:
                    chat_message_content.items.extend(file_items)
                if trace_metadata:
                    chat_message_content.metadata.update({"trace": trace_metadata})

                if not chat_message_content:
                    raise AgentInvokeException("No response from the agent.")

                chat_message_content.metadata["thread_id"] = thread.id
                return AgentResponseItem(message=chat_message_content, thread=thread)

        raise AgentInvokeException(
            "Failed to get a response from the agent. Please consider increasing the auto invoke attempts."
        )

    @trace_agent_invocation
    @override
    async def invoke(
        self,
        messages: str | ChatMessageContent | list[str | ChatMessageContent] | None = None,
        *,
        thread: AgentThread | None = None,
        on_new_message: Callable[[ChatMessageContent], Awaitable[None]] | None = None,
        agent_alias: str | None = None,
        arguments: KernelArguments | None = None,
        kernel: "Kernel | None" = None,
        **kwargs,
    ) -> AsyncIterable[AgentResponseItem[ChatMessageContent]]:
        """Invoke an agent.

View on GitHub (pinned to c028a0c7dc)

Solutions

Increase maximum_auto_invoke_attempts in FunctionChoiceBehavior.Auto(maximum_auto_invoke_attempts=N) to allow more tool-use rounds.
Review the agent's instructions to ensure it knows when to stop calling functions and produce a final answer.
Check that plugin functions return well-formed results that satisfy the agent's information need (avoid errors that cause retries).
Add a stop condition or restructure the workflow so the agent converges on a text response.

Example fix

// before
agent = BedrockAgent(model, function_choice_behavior=FunctionChoiceBehavior.Auto())
# default attempts exhausted

// after
agent = BedrockAgent(
    model,
    function_choice_behavior=FunctionChoiceBehavior.Auto(maximum_auto_invoke_attempts=10),
)

Defensive patterns

Strategy: retry

Validate before calling

from semantic_kernel.connectors.ai.function_choice_behavior import FunctionChoiceBehavior

fcb = FunctionChoiceBehavior.Auto(maximum_auto_invoke_attempts=10)  # raise the ceiling
agent = BedrockAgent(model, function_choice_behavior=fcb)

Try / catch

from semantic_kernel.exceptions.agent_exceptions import AgentInvokeException

for attempts in [5, 10, 15]:
    agent.function_choice_behavior.maximum_auto_invoke_attempts = attempts
    try:
        resp = await agent.get_response(message=msg, thread=thread)
        break
    except AgentInvokeException as e:
        if "increasing the auto invoke attempts" not in str(e):
            raise
else:
    raise RuntimeError("Agent never converged on a final response")

Prevention

Set maximum_auto_invoke_attempts high enough for your tool-use depth.
Write agent instructions that direct it to produce a final answer after gathering information.
Ensure plugin functions return successful, complete results to avoid retry loops.
Log each RETURN_CONTROL round to detect circular tool use early.

When it happens

Trigger: Triggered when every iteration of range(self.function_choice_behavior.maximum_auto_invoke_attempts) hits the RETURN_CONTROL branch — the agent repeatedly asks for function calls and never produces a terminal text response.

Common situations: A plugin function returns results that always prompt the agent to call another function (circular tool use); maximum_auto_invoke_attempts is too low (default); the agent's instructions overly favor tool use over answering; a function throws and the agent retries indefinitely; the model is stuck in a function-calling loop.

Related errors

AI-assisted analysis of microsoft/semantic-kernel@c028a0c7dc (2026-08-13). Data as JSON: /api/errors/ffa68ae0c8cd54bd. Report an issue: GitHub.