Tool use (function calling)

With tool use, you describe functions to the model: a name, a description, and a schema for the arguments. When the model decides a tool would help, it responds with a tool call instead of a final answer. Your code runs the function, sends the result back, and the model continues with real data in hand. The model never executes anything itself.

This is how AI features reach beyond text. Tools let a model look up an order, search a knowledge base, check inventory, create a ticket, or query analytics, which fixes the knowledge-cutoff problem and grounds answers in live data. Tools are also a clean way to get structured choices, like which UI component to show.

The design work is in the tools themselves. Clear names and descriptions matter more than clever prompts. Scope every tool to the current user's permissions in your code, never trusting the model to pass the right account ID. Require confirmation for actions with consequences, validate arguments, return useful errors, and cap the number of steps.

Start with a few well-defined tools. Dozens of overlapping ones confuse the model, cost tokens on every call, and widen what an injected instruction could trigger.