Structured Output and Tool Calling · Tool Calling · lesson 4 of 6

Defining a tool

about 16 minutes · free · runs in your browser

A function, described in JSON

Tool calling inverts the conversation. Instead of asking the model for text, you hand it a list of functions it may call, and it answers with which one, and with what arguments. Your code runs it. The model never touches your data — it only asks.

A tool is a JSON description of a function:

{
    "type": "function",
    "function": {
        "name": "get_weather",
        "description": "Look up the current weather for a city",
        "parameters": {
            "type": "object",
            "properties": {"city": {"type": "string"}},
            "required": ["city"],
        },
    },
}

The description is not documentation, it is the prompt. It is the only thing the model reads when deciding whether this tool is the right one, and a vague description is the whole reason a model "ignores" a tool it should have used.

Your turn: write tool_spec(name, description, params) that builds this structure, where params maps a parameter name to a JSON type string. Every parameter is required.

You start from this, and edit it in the browser:

import fake_llm


def tool_spec(name, description, params):
    """Build a tool definition. params maps a name to a JSON type string."""
    return {}