What are the main differences between generateText, streamText, and generateObject?

generateText returns a full string plus token usage once complete without streaming; streamText returns chunked responses via toDataStreamResponse() for real-time delivery in a Next.js API route; generateObject takes a Zod schema and returns a fully-typed object matching that schema, useful for structured data extraction.

How do I implement tool calling with the Vercel AI SDK?

Define tools server-side using tool({ description, parameters: z.object(...), execute }) and pass them to streamText. Set maxSteps (e.g., maxSteps: 5) to allow the LLM to call a tool and then generate a final answer from the result. On the frontend, use useChat's toolInvocations to render intermediate tool-call state in the UI.

What providers does the Vercel AI SDK support?

The SDK abstracts away provider-specific APIs and works with OpenAI, Anthropic, Google Gemini, and Mistral through provider packages like @ai-sdk/openai and @ai-sdk/anthropic.

Why should I set maxDuration on streaming API routes?

Vercel's default timeout is 10-15 seconds, which can cut off long completions. Setting maxDuration = 30 on streaming API routes ensures long-running LLM responses complete without being interrupted.

What should I use useChat for on the frontend?

useChat manages conversational state including messages, input, submit handler, and loading state, and works against a streamText API route to handle streaming responses and tool invocations in a chat interface.

Skill

Build AI-powered React & Next.js Apps

Name: Vercel AI SDK Expert
Availability: OnlineOnly
Author: Antigravity

Production-grade Vercel AI SDK expert for building streaming AI chat, tool calling, and structured generation in Next.js/React.

Get skill

Works with vercelnextjsreact openaianthropic

Antigravity

Own this? Claim it

Spark score

out of 100

Updated 5 days ago

Version 15.8.0

Models

universal

Add to Favorites

Why it matters

Empower developers to integrate advanced AI features into their React and Next.js applications, enabling dynamic chatbots, generative UIs, and intelligent data handling.

Outcomes

What it gets done

Implement streaming LLM responses for real-time user experiences.

Integrate tool/function calling for LLM interaction with external services.

Generate structured data (JSON) from LLM outputs using Zod schemas.

Build conversational interfaces with the `useChat` hook.

Install

Add it to your toolbox

Run in your project directory:

curl -fsSL https://spark.entire.vc/get/ag-vercel-ai-sdk-expert | bash

Overview

Vercel AI SDK Expert

A Vercel AI SDK skill for building streaming AI chat, structured JSON generation, and tool-calling features in Next.js/React using a unified, provider-agnostic API. Use it for any Next.js/React feature involving LLM streaming, structured output, or tool calling; set maxDuration and maxSteps correctly to avoid the two most common failure modes (timeouts and stalled tool responses).

What it does

Vercel AI SDK Expert helps developers build AI-powered applications, chatbots, and generative UI experiences primarily in Next.js and React, covering both the ai (AI SDK Core) and @ai-sdk/react (AI SDK UI) packages - streaming, language model integration, system prompts, tool calling, and structured data generation. The SDK is a unified framework abstracting away provider-specific APIs (OpenAI, Anthropic, Google Gemini, Mistral) across two layers: AI SDK Core's server-side functions (generateText, streamText, generateObject) and AI SDK UI's frontend hooks (useChat, useCompletion).

When to use - and when NOT to

Use it when adding AI chat or text generation to a React/Next.js app, streaming LLM responses to a frontend UI, implementing tool/function calling with an LLM, returning structured JSON via generateObject, building generative UIs with streaming React components, migrating from direct OpenAI/Anthropic API calls to the unified SDK, or troubleshooting useChat/streamText streaming issues.

Inputs and outputs

generateText returns a full string plus token usage once complete (no streaming); streamText in a Next.js API route returns result.toDataStreamResponse() for chunked delivery. generateObject takes a Zod schema and returns a fully-typed object matching it - useful for structured extraction like parsing a receipt into store name, total, line items, and date. On the frontend, useChat manages conversational state (messages, input, submit handler, loading state) against a streamText API route. Tool calling is defined server-side with tool({ description, parameters: z.object(...), execute }), and maxSteps (e.g. maxSteps: 5) lets the LLM call a tool and then generate a final answer from the result rather than stopping after the tool call - useChat's toolInvocations can render intermediate tool-call state in the UI.

const result = streamText({
  model: openai('gpt-4o'),
  messages,
  tools: { getWeather: tool({ description: '...', parameters: z.object({ location: z.string() }), execute: async ({ location }) => `...` }) },
  maxSteps: 5,
});
return result.toDataStreamResponse();

Integrations

Works with provider packages like @ai-sdk/openai and @ai-sdk/anthropic for model selection. Best practices baked in: set maxDuration = 30 on streaming API routes since Vercel's default 10-15s timeout cuts off long completions; always give tool parameters comprehensive Zod description tags since the LLM relies entirely on those strings to decide when to call a tool; never skip maxSteps when tools are provided, or the LLM can't reply after seeing a tool result; and always wrap generateObject output in try/catch despite Zod's shape guarantee.

Who it's for

Next.js/React developers building AI chat, generative UI, or tool-using LLM features who want a unified, provider-agnostic API instead of hand-rolling streaming and function-calling against each provider separately.

FAQ

Common questions

Discussion