Vercel AI SDK Documentation

What the SDK is

The AI SDK is an open source TypeScript library from Vercel, currently at major version 7. Its repository has about 27k GitHub stars as of September 2026 and is updated daily. It covers the parts of AI features that are tedious to write yourself, in three main areas:

• AI SDK Core: one API for generating text and structured objects, calling tools, embeddings and reranking, and building agents, across providers such as OpenAI, Anthropic, Google, xAI, Azure, Amazon Bedrock, Groq, Mistral, Cohere and DeepSeek.
• AI SDK UI: framework-agnostic hooks for chat and generative interfaces, with streaming to the browser.
• Agent support: tool calling loops, memory, workflow patterns, loop control, approval policies, subagents and MCP (Model Context Protocol) support, plus telemetry and developer tools.

It works with React, Next.js, Vue, Svelte, Nuxt, Node.js, Expo and TanStack Start. You do not need to host on Vercel to use it.

Where it earns its place

Streaming and structured output. Both are fiddly to implement correctly, almost every AI feature needs them, and getting them from a maintained library is a clear win. Structured output uses a schema (commonly Zod), so the model's answer arrives as typed data.

Provider switching. Changing from one model provider to another is mostly a configuration change, which makes comparing models on your own task easy.

Agents without a heavy framework. Tool calling and multi-step loops are built in, so simple agents need no extra library.

Before you start

• Solid JavaScript or TypeScript, including async/await and promises.
• Basic React or another frontend framework for the UI parts.
• An API key from at least one provider.

How to read the docs

1. Start with the introduction and the getting started guide for your framework.
2. Read the Core sections on generating text, streaming and structured data. Everything else builds on these.
3. Then tool calling and agents, once you have a working chat.
4. Use the provider pages only when you need a specific model's options.
5. Check the migration guides if you follow an older tutorial, because APIs changed between major versions.

A good first project

Build a small app that answers questions about your own documents: stream responses into a chat UI, add one tool (such as a search function), and return a structured summary using a schema. That exercises the most important parts of the SDK in a weekend.

Where it falls short

The SDK moves fast, so blog posts and videos go out of date quickly; trust the official docs over tutorials. It is a library, not a full platform: evaluation, prompt management and observability need additional tools. And it is TypeScript only; Python developers need a different library.

Common questions

Is it free? Yes, it is open source. You pay only for the model providers you call.

Do I have to use Next.js or Vercel? No, it supports many frameworks and runs anywhere Node.js does.

Is it a replacement for LangChain? For many TypeScript web apps, yes, since it covers generation, tools and agents with less abstraction.

Best for: TypeScript and React developers adding AI features to an existing product.