Zod schema for CoreMessage

[davidbarratt]

Feature Description

In the documentation, this example is provided:

import { openai } from '@ai-sdk/openai';
import { streamText } from 'ai';

// Allow streaming responses up to 30 seconds
export const maxDuration = 30;

export async function POST(req: Request) {
  const { messages } = await req.json();

  const result = await streamText({
    model: openai('gpt-4-turbo'),
    messages,
  });

  return result.toDataStreamResponse();
}

However, req.json() returns Promise<any>. There are no guarantees here that the request contains an array of CoreMessage objects, let alone an object with a messages key.

Since there is no runtime type checking here, we have to make a request to the model and wait for it to fail before we get an error message.

I think it would be helpful if this library provided a Zod schema for CoreMessage, that way the input can be verified at runtime without having to make a request to the model.

Use Case

• Verify the request at runtime which reduces the number of bogus requests to the AI model.
• Mutate the messages array (i.e. removing tools that may only be called once in a chat session).

Additional context

Writing the CoreMessage in a Zod schema, should allow you to also generate the same types from that schema by using type CoreMessage = z.infer<typeof CoreMessageSchema> which should reduce the overhead in maintaining the schema and types.

Related microsoft/TypeScript#26188

[regexyl]

Ran into the same need on my end - have submitted a PR!

[lgrammel]

messages are in fact not core messages. they are messages from useChat, let's call them ui messages for now. so the validation with the core msg schema would fail / be the wrong approach. streamText automatically converts ui messages to core messages fwiw

[lgrammel]

Schemas are exposed in 4.1.3