microsoft/semantic-kernel · error · ServiceInvalidExecutionSettingsError

If response_format has type 'json_schema', 'json_schema' mus

Error message

If response_format has type 'json_schema', 'json_schema' must be a valid dictionary.

What it means

Raised by the Azure AI Inference prompt-execution settings validator when response_format is a dict whose 'type' is 'json_schema' but the nested 'json_schema' value is not a dict. The connector needs the schema as a proper JSON-schema dictionary to enable structured output; a string, None, list, or other type cannot be sent as a valid schema definition.

Source

Thrown at python/semantic_kernel/connectors/ai/azure_ai_inference/azure_ai_inference_prompt_execution_settings.py:73

    @model_validator(mode="before")
    def validate_response_format_and_set_flag(cls, values: Any) -> Any:
        """Validate the response_format and set structured_json_response accordingly."""
        if not isinstance(values, dict):
            return values
        response_format = values.get("response_format", None)

        if response_format is None:
            return values

        if isinstance(response_format, dict):
            if response_format.get("type") == "json_object":
                return values
            if response_format.get("type") == "json_schema":
                json_schema = response_format.get("json_schema")
                if isinstance(json_schema, dict):
                    values["structured_json_response"] = True
                    return values
                raise ServiceInvalidExecutionSettingsError(
                    "If response_format has type 'json_schema', 'json_schema' must be a valid dictionary."
                )
        if isinstance(response_format, type):
            if issubclass(response_format, BaseModel):
                values["structured_json_response"] = True
            else:
                values["structured_json_response"] = True
        else:
            raise ServiceInvalidExecutionSettingsError(
                "response_format must be a dictionary, a subclass of BaseModel, a Python class/type, or None"
            )

        return values


@experimental
class AzureAIInferenceEmbeddingPromptExecutionSettings(PromptExecutionSettings):
    """Azure AI Inference Embedding Prompt Execution Settings.

View on GitHub (pinned to c028a0c7dc)

Solutions

Ensure response_format['json_schema'] is a dict — pass a parsed JSON-schema object, e.g. {'type':'json_schema','json_schema': {'name':'X','schema': {...}}} per the Azure AI Inference shape.
If you have a Pydantic model, either pass the class/type directly (the validator accepts BaseModel subclasses) or call MyModel.model_json_schema() to get the dict.
Double-check the key name is 'json_schema' (not 'schema') and the value is not a string.

Example fix

# before
settings.response_format = {"type": "json_schema", "json_schema": '{"type":"object"}'}

# after
settings.response_format = {
    "type": "json_schema",
    "json_schema": {"name": "MyObj", "schema": {"type": "object", "properties": {}}},
}

Defensive patterns

Strategy: validation

Validate before calling

def validate_json_schema_response_format(rf):
    if isinstance(rf, dict) and rf.get("type") == "json_schema":
        assert isinstance(rf.get("json_schema"), dict), \
            "json_schema must be a dict when type is 'json_schema'"
    return rf

settings.response_format = validate_json_schema_response_format(settings.response_format)

Type guard

def is_valid_json_schema_dict(rf) -> bool:
    return (
        isinstance(rf, dict)
        and rf.get("type") == "json_schema"
        and isinstance(rf.get("json_schema"), dict)
    )

Try / catch

from semantic_kernel.exceptions import ServiceInvalidExecutionSettingsError
try:
    settings.response_format = rf
except ServiceInvalidExecutionSettingsError as e:
    if "json_schema" in str(e):
        settings.response_format = {"type": "json_schema", "json_schema": parsed_schema_dict}

Prevention

Always pass the inner schema as a dict, never a JSON string.
Use the key name 'json_schema' (not 'schema').
For Pydantic models, pass the class directly or via model_json_schema().

When it happens

Trigger: Passing response_format={'type':'json_schema','json_schema': '<json string>'} or omitting/mis-typing the inner 'json_schema' key (e.g. using 'schema' instead of 'json_schema', or passing a Pydantic class instead of its .model_json_schema() dict).

Common situations: Confusing OpenAI's response_format shape with a custom one; passing a serialized JSON string instead of a parsed dict; using the wrong key name ('schema' vs 'json_schema'); pasting an example that omits the inner dict.

Related errors

AI-assisted analysis of microsoft/semantic-kernel@c028a0c7dc (2026-08-13). Data as JSON: /api/errors/c47d2bb98a73df5d. Report an issue: GitHub.