Vercel AI SDK

1 · What it is

The Vercel AI SDK is a toolkit for building AI apps and agents in TypeScript, a programming language. You can use it from Node.js on a server, or in front-end frameworks such as Svelte, Vue, Next.js and React. The makers of Next.js publish the AI SDK as open-source code that anyone can use at no cost. It has two main parts. AI SDK Core asks models for text, structured data and tool calls. AI SDK UI gives you ready-made hooks for chat screens. A hook is a small reusable piece of front-end code.

The main problem it solves is that every AI company, or provider, has its own way of calling its models. The SDK puts one shared interface in front of many providers. Switching providers means changing the model you name, not rewriting the calls. Supported providers include OpenAI, Anthropic, Google, Mistral and Groq, and outside developers have written more.

Large language models, the AI behind chatbots, can be slow when they write long answers. Picture a homework-help chat page. On the server, the streamText function sends the reply in pieces as the model writes them. In the page, the useChat hook shows each piece as it arrives, so the student reads along instead of watching a spinner.

Tools are things the model can call to do a specific task, such as looking up the weather. Each tool comes with a schema. A schema is a description of the inputs the tool accepts. The model reads it. The SDK uses the same schema to check the model’s tool request. By default, a call gets one answer from the model. To allow more steps, you set a stop rule. With a stop rule, the SDK feeds each tool result back to the model until there are no more tool calls or the rule says stop. The ToolLoopAgent class wraps that loop for you. In the docs’ weather example, the agent’s first step is a weather lookup that returns a Fahrenheit reading.