OpenAI Structured Output Allowed types

API
,
1

It seems like Structured Output does not allow objects of arbitrary types. For example, if I have the following Pydantic model:

class ToolCall(pydantic.BaseModel):
    """Represents a single tool call with its arguments.

    This model is used to structure and validate tool calls made by the agent.
    """

    tool: str = pydantic.Field(..., description="The name of the tool to be called")
    tool_running_message: str = pydantic.Field(
        default="", description="The message to be displayed when the tool is running"
    )
    tool_completed_message: str = pydantic.Field(
        default="", description="The message to be displayed when the tool is completed"
    )
    tool_failed_message: str = pydantic.Field(
        default="", description="The message to be displayed when the tool fails"
    )
    args: Dict[str, Any] = pydantic.Field(
        default_factory=dict,
        description="The arguments to be passed to the tool. Can be left empty if the tool does not require any arguments.",
    )

class ResponseFormat(pydantic.BaseModel):
    reasoning: str = pydantic.Field(
        ...,
        description="Analyse the conversation history and the current user query to provide a detailed reasoning process.",
    )
    tool_calls: typing.List[ToolCall] = pydantic.Field(
        ..., description="The tool calls to be made", min_length=1
    )

I get a 400 Bad Request Error:

BadRequestError: Error code: 400 - {'error': {'message': "Invalid schema for response_format 'ResponseFormat': In context=(), 'required' is required to be supplied and to be an array including every key in properties. Extra required key 'args' supplied.", 'type': 'invalid_request_error', 'param': 'response_format', 'code': None}}

Am I doing something wrong here?