Vercel AI SDK
An open-source TypeScript toolkit for building AI apps — unified model API, streaming, structured output, tool calling, and UI hooks.
The Vercel AI SDK is the de facto TypeScript toolkit for AI apps: one provider-agnostic API for text, structured objects, and tool calls, first-class streaming, and framework UI hooks (React, Svelte, Vue) for building chat and generative interfaces fast.
The Vercel AI SDK is an open-source TypeScript toolkit for building AI-powered applications. It gives you one provider-agnostic API for the things every LLM app needs — generating text, generating structured objects, tool calling, and streaming — plus framework hooks for wiring those into a UI. It has become the default way to build AI features in the TypeScript/JavaScript ecosystem.
It is aimed at full-stack and frontend developers who want to ship chat and generative-UI features without gluing together a provider SDK, a streaming layer, and a state library by hand. Swap models with a one-line change, and stream tokens or structured data to the browser with first-class primitives.
Highlights
- Unified model API —
generateText,streamText, and tool calling across many providers; change models via config. - Streaming-first — stream tokens and structured output to the client with backpressure handled.
- Structured output —
generateText/streamTextwith anoutputsetting and schema validation (Zod and friends);generateObject/streamObjectare deprecated as of AI SDK 6. - UI hooks — React, Svelte, and Vue hooks (
useChat,useCompletion) for chat and generative interfaces. - Tool calling & agents — define tools the model can call, with multi-step loops.
In an AI-assisted workflow
import { streamText } from "ai";
const result = streamText({ model: "anthropic/claude", prompt });
return result.toUIMessageStreamResponse(); // stream straight to the browserTIP
The AI SDK overlaps with both structured-output libraries and gateways: it does typed output like Instructor and provider-switching like a gateway. Pair it with OpenRouter or LiteLLM when you want centralized routing/cost control behind it.
Good to know
The Vercel AI SDK is open source (Apache-2.0) and free; you pay your model provider for tokens. It's TypeScript-first — the natural choice in JS/TS apps, less so for Python backends (where Instructor/BAML fit better). It's framework-agnostic despite the Vercel name and runs anywhere Node/edge runs. The current major version is AI SDK 7 (June 2026), which includes ToolLoopAgent, WorkflowAgent for durable, resumable agent execution, tool approvals, and optional OpenTelemetry telemetry using the GenAI semantic conventions.
Frequently asked questions
- What is the Vercel AI SDK?
- The Vercel AI SDK is an open-source TypeScript toolkit for building AI-powered applications. It provides one provider-agnostic API for generating text, generating structured objects, tool calling, and streaming — generateText and streamText, with an output setting for structured results — plus React, Svelte, and Vue hooks like useChat for wiring chat and generative UI into your app.
- Is the Vercel AI SDK free?
- Yes — open source under Apache-2.0 and free; you pay your model provider for tokens. Despite the Vercel name it's framework-agnostic and runs anywhere Node or edge runtimes run.
- Vercel AI SDK vs LiteLLM?
- They overlap on provider-switching but live at different layers: the AI SDK is a TypeScript application toolkit with streaming and UI hooks, while LiteLLM is a Python library/self-hosted gateway for routing all LLM traffic. They also pair well — put OpenRouter or LiteLLM behind the AI SDK when you want centralized routing and cost control.
- Does the Vercel AI SDK support structured output?
- Yes — generateText and streamText with an output setting produce schema-validated output (Zod and friends), with streaming supported; the older generateObject and streamObject were deprecated in AI SDK 6 and will be removed in a future version. For Python backends, libraries like Instructor or BAML fill the same niche.
Filed under
typescript · streaming · structured-output · open-source · react
Related
- LiteLLMCall 100+ LLM APIs with one OpenAI-format interface — as a Python library or a self-hosted gateway/proxy.
- Add a Streaming LLM EndpointScaffold a token-streaming LLM endpoint — server-side streaming plus the client handler — so responses render incrementally instead of after a long wait.
- Calling Any Model: Unified LLM Gateways & SDKs in 2026Why teams put a unified layer in front of LLM providers — and how LiteLLM, OpenRouter, and the Vercel AI SDK compare for fallback and cost control.
- OpenRouterA hosted unified API to hundreds of models from many providers, with one key, one bill, and automatic fallbacks.
- Structured Output vs JSON Mode vs Function Calling: Which to Use in 2026The reliable ways to get typed data out of an LLM — what JSON mode, function calling, and native structured outputs each guarantee, and when to use which.
- InstructorGet structured, validated output from LLMs using plain type definitions, with automatic retries on validation failure.
- Mastra vs LangChain (2026): TypeScript Agent Frameworks ComparedMastra vs LangChain.js and LangGraph.js in 2026: license, workflows, memory, evals, observability and deployment compared, plus where the Vercel AI SDK fits.
- Vercel SandboxEphemeral Firecracker microVMs on Vercel for untrusted and AI-generated code — millisecond startup, Node and Python runtimes, persistent by default.