vercel-ai-sdk-expert

Use streamText from the ai package in an App Router API route and return result.toDataStreamResponse(). On the client, the useChat hook from @ai-sdk/react consumes the stream and updates messages automatically.

Define tools in streamText using the tool() helper with a Zod parameters schema and an execute function. Set maxSteps to 2 or higher so the LLM can see the tool result and produce a final text response.

Use generateObject with a Zod schema describing the expected shape. The returned object is fully typed according to the schema, and you should still wrap the call in try/catch to handle validation failures.

The serverless function is hitting its default timeout. Add export const maxDuration = 30 to the Next.js API route file so streamText has enough time to finish streaming the response.

Yes, the SDK abstracts provider-specific APIs. Install the relevant provider package such as @ai-sdk/openai or the Anthropic equivalent, then pass models like openai('gpt-4o') or anthropic('claude-3-5-sonnet-20240620') to the same core functions.