AI SDK (Vercel)

The AI SDK is a TypeScript toolkit designed to help developers build AI-powered applications and agents, with support for React, Next.js, Vue, Svelte, Node.js and other JavaScript environments ^[1]. It is published by Vercel, and its source lives in the vercel/ai repository on GitHub.

A single interface for many models

Model providers typically each ship their own client library, request format and conventions for streaming a response back to the browser. An application that wants to use more than one provider — or to switch providers later — has to absorb those differences somewhere. The AI SDK standardizes how models are integrated across the providers it supports: OpenAI, Anthropic, Google Generative AI, Azure, Amazon Bedrock, Groq, Mistral, DeepSeek, Together.ai and Cohere, plus Vercel's own AI Gateway ^[1]. Application code is written against the SDK's interface, and the provider-specific details sit behind it.

Origin

The vercel/ai repository was created on 23 May 2023 and has roughly 28,000 stars ^[2]. GitHub reports its license as NOASSERTION, meaning no standard license identifier is recorded ^[2].

Three surfaces

The SDK is organized around three primary surfaces ^[1].

AI SDK Core is a unified API for generating text, structured objects and tool calls, and for building agents ^[1]. Its streamText function handles multiple tool steps automatically ^[1]: when a model asks to call a tool, the result is passed back and the model continues, and that loop is managed by the function rather than written out by hand at each call site.

AI SDK UI provides hooks for chat and generative interfaces that are not tied to a single front-end framework ^[1]. The hooks include useChat and useObject ^[1]. useChat covers the conversational case — sending messages and receiving streamed replies — while useObject covers generating structured objects that a UI can render directly.

AI SDK Harnesses offer a uniform API for running established agent harnesses, such as Claude Code, Codex and Pi, through something called HarnessAgent ^[1]. Where Core wraps model calls, this surface wraps already-built agent runtimes behind one calling convention.

Starter templates

Alongside the library, Vercel ships starter templates built on the SDK, including a chatbot starter, a RAG knowledge base and a multi-modal chat ^[1]. These are working starting points rather than documentation samples: a developer can begin from a functioning chat or retrieval application and modify it.

Where it fits

The SDK sits at the point where a web application meets a model provider. Three choices define it. It is written for TypeScript and the JavaScript ecosystem rather than for a single framework, listing React, Next.js, Vue, Svelte and Node.js among its targets ^[1]. It treats provider choice as a configuration detail by standardizing integration across a long list of providers and its own gateway ^[1]. And it covers the whole path from a single text generation call up through structured output, tool calls, agents and pre-built agent harnesses ^[1]. For developers, the practical effect is that the parts of an AI application most likely to change — which model is behind it, which framework renders it, which agent runtime executes it — are addressed by the same toolkit.