Tool use

On its own a model can only emit text. Tool use is what gives it hands. You define an action with a name, a description, and a schema for its inputs; the model decides when to call it and with what arguments; your code runs the real thing and hands the result back. Fetch an order, run a query, hit an API: all of it happens through tools.

Defining a tool

With the Vercel AI SDK a tool is a description plus an input schema plus the function that does the work. Pass it to the model and it can call it:

import { generateText, tool, stepCountIs } from 'ai'
import { openai } from '@ai-sdk/openai'
import { z } from 'zod'

const getWeather = tool({
  description: 'Get the current weather for a city',
  inputSchema: z.object({ city: z.string() }),
  execute: async ({ city }) => ({ city, tempC: 21 }),
})

const { text } = await generateText({
  model: openai('gpt-5-mini'),
  tools: { getWeather },
  stopWhen: stepCountIs(5),
  prompt: 'What is the weather in Lisbon?',
})

The model reads the description and schema, decides to call getWeather, the SDK runs it, and the result flows back into the answer.

What makes tools work

• The description is the interface. The model chooses a tool based on its name and description, so write them for a reader who has never seen your code.
• Fewer, sharper tools beat a big pile. Too many options and the model picks wrong or hallucinates arguments. Start with two or three.
• Validate the arguments. The model chooses the inputs, and it can choose badly. A typed schema catches the obvious mistakes before you execute.