Chapter 10 of 36 · ~1 min

Tool Calling

A model that can only produce text is limited to what it absorbed in training. Give it a way to ask for things (a search, a database lookup, a calculator, an API, a business system) and the pattern changes. Instead of generating an answer, the model generates a request: which tool, with what arguments. The application runs the tool and hands the result back as more text. The model reads it and continues.

This is the hinge on which the rest of the course turns. From here on the model is not just answering, it is deciding what to do. And the moment a model decides what to do, a designer has to decide how much of that deciding to allow.

A tool call, step by stepModelApplicationToolrequest: search("…")run the toolresultresult, as textfinal answerThe model never runs anything. It writes a request; the application decides whether to honour it.
A tool call is a request the model writes. The application runs it and hands the result back as text.

Experiment

Live model

Ask a question that needs a price. Watch the model write a tool request, the tool answer, and the model finish.

Runs against a live model through Chatterfly's server. Your text is sent to the model provider and not stored. The tool is a fake price list on the server; nothing real is looked up.

Big question

When the model chooses the action, who is responsible for the outcome?