Tools & JSON responses

Describe the functions you're ready to run and pass them in the request.

{
  "model": "gpt-5.6-sol",
  "messages": [{ "role": "user", "content": "What's the weather in Kaliningrad?" }],
  "tools": [
    {
      "type": "function",
      "function": {
        "name": "get_weather",
        "description": "Current weather in a city",
        "parameters": {
          "type": "object",
          "properties": { "city": { "type": "string" } },
          "required": ["city"]
        }
      }
    }
  ]
}

Then it's a two-step flow: the model asks for a call — you supply the result.

// 1. The model asks to call a function
const first = await client.chat.completions.create({ model, messages, tools });
const call = first.choices[0].message.tool_calls?.[0];

if (call) {
  const args = JSON.parse(call.function.arguments);
  const result = await getWeather(args.city); // your code

  // 2. Return the result and get the final response
  messages.push(first.choices[0].message);
  messages.push({ role: "tool", tool_call_id: call.id, content: JSON.stringify(result) });

  const final = await client.chat.completions.create({ model, messages, tools });
  console.log(final.choices[0].message.content);
}

If the response feeds into your code, set a schema — the model returns an object with the exact structure, no text parsing required.

{
  "model": "gpt-5.6-sol",
  "messages": [{ "role": "user", "content": "Parse this address: 12 Lenin St, Kaliningrad" }],
  "response_format": {
    "type": "json_schema",
    "json_schema": {
      "name": "address",
      "schema": {
        "type": "object",
        "properties": {
          "street": { "type": "string" },
          "house": { "type": "string" },
          "city": { "type": "string" }
        },
        "required": ["street", "house", "city"],
        "additionalProperties": false
      }
    }
  }
}

The gateway itself doesn't access the internet. Add a tool like web_search, run the search with your own service, and feed the results back into the conversation — the model will answer using fresh data.