microsoft/semantic-kernel · error · ServiceInvalidExecutionSettingsError

Auto invocation of tool calls may only be used with a single

Error message

Auto invocation of tool calls may only be used with a single completion.

What it means

Raised by _verify_function_choice_settings when auto-invocation of tool calls is enabled and extra_parameters contains 'n' greater than 1. Auto-invocation needs a single deterministic completion to feed the tool result back into the conversation loop; requesting multiple completions (n>1) makes the result ambiguous and the loop cannot proceed.

Source

Thrown at python/semantic_kernel/connectors/ai/azure_ai_inference/services/azure_ai_inference_chat_completion.py:207

            if len(chunk.choices) == 0:
                continue
            chunk_metadata = self._get_metadata_from_response(chunk)
            yield [
                self._create_streaming_chat_message_content(chunk, choice, chunk_metadata, function_invoke_attempt)
                for choice in chunk.choices
            ]

    @override
    def _verify_function_choice_settings(self, settings: "PromptExecutionSettings") -> None:
        if not isinstance(settings, AzureAIInferenceChatPromptExecutionSettings):
            raise ServiceInvalidExecutionSettingsError(
                "The settings must be an AzureAIInferenceChatPromptExecutionSettings."
            )
        if settings.extra_parameters is not None and settings.extra_parameters.get("n", 1) > 1:
            # Currently only OpenAI models allow multiple completions but the Azure AI Inference service
            # does not expose the functionality directly. If users want to have more than 1 responses, they
            # need to configure `extra_parameters` with a key of "n" and a value greater than 1.
            raise ServiceInvalidExecutionSettingsError(
                "Auto invocation of tool calls may only be used with a single completion."
            )

    @override
    def _update_function_choice_settings_callback(
        self,
    ) -> Callable[["FunctionCallChoiceConfiguration", "PromptExecutionSettings", FunctionChoiceType], None]:
        return update_settings_from_function_call_configuration

    @override
    def _reset_function_choice_settings(self, settings: "PromptExecutionSettings") -> None:
        if hasattr(settings, "tool_choice"):
            settings.tool_choice = None
        if hasattr(settings, "tools"):
            settings.tools = None

    @override
    def _prepare_chat_history_for_request(

View on GitHub (pinned to c028a0c7dc)

Solutions

1. If you need auto tool-invocation, set extra_parameters n to 1 (or remove the 'n' key).
2. If you need multiple completions (n>1), disable auto-invocation (function_choice_behavior=None) and handle tool calls manually.
3. Split into two separate calls: one for multiple completions, one for tool-augmented single completion.

Example fix

# before
settings.extra_parameters = {"n": 3}
settings.function_choice_behavior = "auto"

# after  (auto-invocation path)
settings.extra_parameters = {"n": 1}
settings.function_choice_behavior = "auto"

def ensure_single_completion_for_auto_invoke(settings):
    if settings.function_choice_behavior and settings.function_choice_behavior.auto_invoke:
        n = (settings.extra_parameters or {}).get("n", 1)
        assert n == 1, "Auto-invocation requires n == 1"
    return settings

def auto_invoke_with_single_completion(settings) -> bool:
    ep = settings.extra_parameters or {}
    return ep.get("n", 1) == 1 or not (
        settings.function_choice_behavior and settings.function_choice_behavior.auto_invoke
    )

from semantic_kernel.exceptions import ServiceInvalidExecutionSettingsError
try:
    service._verify_function_choice_settings(settings)
except ServiceInvalidExecutionSettingsError as e:
    if "single completion" in str(e):
        settings.extra_parameters = {"n": 1}

When it happens

Trigger: Enabling function_choice_behavior='auto' (or required) while also setting extra_parameters={'n': N} with N>1 on AzureAIInferenceChatPromptExecutionSettings. Azure AI Inference does not expose n directly, so users configure it via extra_parameters.

Common situations: Wanting multiple candidate responses (n>1) and tool use in the same call; leaving extra_parameters from a prior config that set n for non-tool calls; misunderstanding that auto-invocation implies n=1.

Related errors

• The settings must be an AzureAIInferenceChatPromptExecutionS
• No FunctionResultContent found in the message items
• Tool message found without a preceding message.
• Tool message found after a user or system message.
• If response_format has type 'json_schema', 'json_schema' mus

AI-assisted analysis of microsoft/semantic-kernel@c028a0c7dc (2026-08-13). Data as JSON: /api/errors/382d26b464b3025d. Report an issue: GitHub.