ai-sdk_skill

Overview

This skill provides a concise reference for the Vercel AI SDK, a TypeScript toolkit for building AI-powered frontends and server integrations. It covers text and structured generation, streaming, tool definitions, agents, hooks for React/Vue/Svelte, and provider integrations. Use it to implement production-ready chat UIs, tool-calling flows, and agent workflows with streaming support.

How this skill works

The SDK exposes core functions like generateText/streamText and generateObject/streamObject for synchronous and streaming generation. It enables defining tools with Zod schemas and executing them inside model runs, and it supplies a ToolLoopAgent for orchestrating multi-step agent workflows. UI hooks such as useChat, useCompletion, and useObject connect frontend components to streaming model responses and transport layers.

When to use it

• Building chat interfaces with streaming responses (useChat).
• Generating free-form text or structured JSON from models (generateText/generateObject).
• Defining callable tools with typed inputs and executing them from LLMs.
• Implementing agents that loop over tools and decisions (ToolLoopAgent).
• Integrating with external providers (OpenAI, Anthropic, Google, Mistral).

Best practices

• Use Zod schemas for tool input/output to validate and type structured data.
• Prefer streaming APIs (streamText/streamObject) for responsive UIs and progressive rendering.
• Isolate provider configuration so you can swap models without changing app logic.
• Limit agent step counts and use stopWhen/prepareStep controls to avoid runaway loops.
• Convert UI messages to model messages with provided helpers for consistent prompts.

Example use cases

• React chat widget that streams model replies with useChat and a custom transport.
• Server route that converts UI messages and streams model output to the client.
• Tool-enabled assistant that calls a weather or search tool via typed Zod inputs.
• Agent that performs multi-step research and calculations using ToolLoopAgent with stopWhen.
• Streaming JSON generator for forms or data ingestion using generateObject/streamObject.