Tools & Structured Output

Models that advertise capabilities.tools: true can choose custom functions. Your application executes the function and sends the result back; the platform does not run customer code.

Function calling with Responses

const tools = [
  {
    type: "function",
    name: "get_order_status",
    description: "Look up an order by its public id",
    parameters: {
      type: "object",
      properties: {
        order_id: { type: "string" },
      },
      required: ["order_id"],
      additionalProperties: false,
    },
    strict: true,
  },
];

const first = await client.responses.create({
  model: "your-tool-capable-model",
  input: "Where is order A-104?",
  tools,
});

const call = first.output.find((item) => item.type === "function_call");
if (call) {
  const result = await getOrderStatus(JSON.parse(call.arguments));
  const second = await client.responses.create({
    model: "your-tool-capable-model",
    tools,
    input: [
      ...first.output,
      {
        type: "function_call_output",
        call_id: call.call_id,
        output: JSON.stringify(result),
      },
    ],
  });
  console.log(second.output_text);
}

Function calling with Chat Completions

Use the standard tools, tool_choice, parallel_tool_calls, assistant tool_calls, and role: "tool" message fields. Both streaming and non-streaming tool calls are supported on compatible model routes.

Strict JSON Schema

const response = await client.responses.create({
  model: "your-structured-output-model",
  input: "Extract the invoice number and total: Invoice A-52, total $90.",
  text: {
    format: {
      type: "json_schema",
      name: "invoice",
      strict: true,
      schema: {
        type: "object",
        properties: {
          invoice_number: { type: "string" },
          total: { type: "number" },
        },
        required: ["invoice_number", "total"],
        additionalProperties: false,
      },
    },
  },
});

const invoice = JSON.parse(response.output_text);

For Chat Completions, send the equivalent schema under response_format: { type: "json_schema", json_schema: ... }.

Structured output is provider- and model-dependent. Enable it only when model discovery advertises structuredOutput: true. A provider rejection is returned as an OpenAI-shaped error rather than silently falling back to unstructured text.