Function calling only works in Playground not SDK

API Bugs
, , ,
1

I’m having an issue where the tools calling only works in the platform but not in the Python SDK. Given the same prompt and input, in the SDK i’m getting a response from the bot and in the platform I’m getting the correct output for function calling.

Below is my code and function definition and the message that I used to test both on Playground and SKD is “How’s my nutrition and sleep?”

EDIT: I’m using the same parameters for both SDK and Playground (model, temperature, tools=auto)

COACHING_TOOLS = [
    {
        "type": "function",
        "name": "get_nutrition_data_by_user_and_date_range",
        "description": "Retrieves nutrition data for a user within a specified date range.",
        "strict": True,
        "parameters": {
            "type": "object",
            "properties": {
                "start_timestamp": {
                    "type": "string",
                    "format": "date-time",
                    "description": "The start timestamp for the date range (ISO 8601 format). If no date specified in user message leave this field empty to retrieve data for the last 7 days."
                },
                "end_timestamp": {
                    "type": "string",
                    "format": "date-time",
                    "description": "The end timestamp for the date range (ISO 8601 format). If no date specified in user message leave this field empty to retrieve data for the last 7 days."
                }
            },
            "required": [
                "start_timestamp",
                "end_timestamp"
            ],
            "additionalProperties": False
        }
    },
{
        "type": "function",
        "name": "get_sleep_data_by_user_and_date_range",
        "description": "Retrieves sleep and recovery data for a user within a specified date range.",
        "strict": True,
        "parameters": {
            "type": "object",
            "properties": {
                "start_timestamp": {
                    "type": "string",
                    "format": "date-time",
                    "description": "The start timestamp for the date range (ISO 8601 format). If no date specified in user message leave this field empty to retrieve data for the last 7 days."
                },
                "end_timestamp": {
                    "type": "string",
                    "format": "date-time",
                    "description": "The end timestamp for the date range (ISO 8601 format). If no date specified in user message leave this field empty to retrieve data for the last 7 days."
                }
            },
            "required": [
                "start_timestamp",
                "end_timestamp"
            ],
            "additionalProperties": False
        }
    }
]

response = client.responses.create(
            model="gpt-4.1",
            temperature=0.5,
            tool_choice="auto",
            input=conversation_messages,
            tools=COACHING_TOOLS,
        )

Output from platform:

Screenshot 2025-06-29 at 17.24.12
Screenshot 2025-06-29 at 17.24.121362×1138 83 KB

Output from SDK:
[ResponseOutputMessage(id=‘msg_6861490192a481a2898aedde3206623d06adad9ab2ad6c28’, content=[ResponseOutputText(annotations=, t…in mind. How can I assist you today?', type=‘output_text’, logprobs=)], role=‘assistant’, status=‘completed’, type=‘message’)]

2

What code do you that handles the tool call?

3

If I’m setting the tool_choice=‘required’ this code below works and the model is using the tools.

 tool_calls = response.output
        if tool_calls:
            for tool_call in tool_calls:
                print("Requesting function call: " + tool_call.name)
                if tool_call.type != "function_call":
                    continue

                name = tool_call.name
                args = json.loads(tool_call.arguments)

I don’t want to use tool_choice=‘required’ because there are a lot of instances where there’s no need for tools in my use case

4

The AI is wondering how it can assist.

Either the latest user input is not asking anything targeted, or the user input is not being received within your “conversation_messages”.

Review the construction of your chat history and the addition of the speculative new input to that, by logging the value of conversation_messages right before this API call, and see if you actually have “user” asking something that “tools” will be a useful solution for the AI to employ.

5

I checked both of these and everything seems correct. The only difference is when I set the tool_choice=‘required’ it asks for the correct tool if my messages ask for either nutrition or sleep or both. If the tool_choice is auto, then no tools is asked for. My input is super obvious → “How is my nutrition?”, “How is my sleep?”

EDIT: You were right. I was not sending my message to the SDK. Now everything works fine