Tool use
Letting a model call functions you define — search, read a file, send a request — and read back what they return.
Also written: function calling, tool calling, tools.
You describe the functions available, with their parameters. The model replies not with prose but with a structured request to call one. Your code runs it and hands back the result, and the model continues.
This is the mechanism underneath essentially every useful AI product. The model stays a text predictor; everything it can actually do comes from the tools you gave it, which means the tool descriptions are part of your prompt and deserve the same care.