OpenAI Responses API: Function Calling Loop

OpenAI tool calls are output items

With the Responses API, function calling is a loop over response output items. You provide JSON Schema-backed function definitions. The model emits function_call items. Your application executes those calls and sends function_call_output items back with the matching call_id.

The modern detail that matters in real systems: preserve the response output items that are still part of the run. Do not throw away tool-call state and then wonder why the second call feels amnesiac. Tiny detail, huge bug farm.

Strict schemas reduce ambiguity

Use strict: true and additionalProperties: false when the tool has a known contract. This narrows the model's output space and makes executor validation simpler. Parameter descriptions are not decorative; they are part of the action surface the model sees.

Model choice is part of the run contract

Current OpenAI docs recommend gpt-5.5 as the default starting point for complex reasoning and coding, with smaller variants such as gpt-5.4-mini or gpt-5.4-nano when latency and cost dominate. Do not cargo-cult a model string from an old example. Write down why the run uses that model.

Outputs should be machine-readable

Return small JSON strings with stable fields. Avoid raw API dumps. If the model needs more detail, include a next-step hint like read_full_record or fetch_url instead of stuffing everything into one result.

Code

Responses API function calling shape·python

from openai import OpenAI
import json

client = OpenAI()

tools = [{
    "type": "function",
    "name": "lookup_order",
    "description": "Look up one order by id. Use when the user asks about order status.",
    "parameters": {
        "type": "object",
        "properties": {
            "order_id": {"type": "string", "description": "Public order id, e.g. ORD-1042"}
        },
        "required": ["order_id"],
        "additionalProperties": False,
    },
    "strict": True,
}]

input_items = [{"role": "user", "content": "Where is ORD-1042?"}]

while True:
    response = client.responses.create(
        model="gpt-5.5",
        input=input_items,
        tools=tools,
    )

    calls = [item for item in response.output if item.type == "function_call"]
    if not calls:
        print(response.output_text)
        break

    input_items += response.output
    for call in calls:
        args = json.loads(call.arguments)
        result = {"order_id": args["order_id"], "status": "shipped"}
        input_items.append({
            "type": "function_call_output",
            "call_id": call.call_id,
            "output": json.dumps(result),
        })

OpenAI Responses API: Function Calling Loop

OpenAI tool calls are output items

Strict schemas reduce ambiguity

Model choice is part of the run contract

Outputs should be machine-readable

Code

External links

Exercise

Progress

Comments 0