Tool calling

Tool calling lets a model ask you to run a function and then use the result. The wire format is OpenAI’s tools / tool_choice.

Check the model supports it first

Tool calling is not available on every model in our catalog. There is no separate capability flag — a model supports tools exactly when tools and tool_choice appear in its supported_parameters.

If the model doesn’t declare them, your tools array is dropped before the request is forwarded. You get a normal, confident, tool-free answer and no error. This is the most common cause of “the model ignores my tools”.

Check before you build:

curl "https://api.resetdata.ai/api/v1/models/detail?slug=zai/glm-5.2" \
  -H "Authorization: Bearer $RESETDATA_API_KEY" \
  | jq '.supported_parameters | map(select(. == "tools" or . == "tool_choice"))'

Several capable text models — and all image, audio, embedding and reranker models — do not offer tool calling. The catalog in the app shows a Function-calling badge derived from the same check.

The loop

tools = [{
    "type": "function",
    "function": {
        "name": "get_weather",
        "description": "Get the current weather for a city.",
        "parameters": {
            "type": "object",
            "properties": {
                "city": {"type": "string", "description": "City name"},
            },
            "required": ["city"],
        },
    },
}]

messages = [{"role": "user", "content": "What's the weather in Sydney?"}]

resp = client.chat.completions.create(
    model="zai/glm-5.2", messages=messages, tools=tools,
)
msg = resp.choices[0].message

if msg.tool_calls:
    messages.append(msg)                      # keep the model's request
    for call in msg.tool_calls:
        args = json.loads(call.function.arguments)
        result = get_weather(**args)          # you run it
        messages.append({
            "role": "tool",
            "tool_call_id": call.id,
            "content": json.dumps(result),
        })

    resp = client.chat.completions.create(    # send results back
        model="zai/glm-5.2", messages=messages, tools=tools,
    )

print(resp.choices[0].message.content)

Four steps: you send tools, the model asks for one, you execute it and append the result with its tool_call_id, then call again so the model can answer. We never execute anything — tools run entirely in your code.

Controlling when tools are used

tool_choice	Behaviour
"auto"	Model decides. The usual choice.
"none"	Never call a tool.
{"type": "function", "function": {"name": "..."}}	Force a specific tool.

Practical notes

• Descriptions are the interface. The model chooses from your description text, so write it for a reader who cannot see your code.
• Validate arguments. Arguments are model-generated JSON. Parse defensively and never pass them unchecked into anything with side effects.
• Handle several calls at once. tool_calls is an array; a model may request multiple tools in one turn.
• Every turn costs tokens. Tool definitions are re-sent as input on each call, and the loop means several round trips per user question — see Usage and costs.
• Bound the loop. Cap iterations so a model that keeps requesting tools cannot spin indefinitely.
• Tool quality varies. Support being declared doesn’t mean every model is equally reliable at multi-step tool use. Test with your actual tools before committing.