Streaming Text Parts By Hand

Sometimes you'll want to not only be able to stream custom parts, but also stream in the AI SDK's built-in parts in order to mimic them. I'm specifically thinking here of streaming manually text parts by hand.

This comes up in a few edge cases where you want something to look like an LLM call that isn't actually an LLM call, or you have some data that's been prepared in parallel that you just want to spit out all at once.

To do this, we're going to create a UIMessageStream using createUIMessageStream:

With text streaming, there's a specific sequence you need to follow. You start with a text-start event, then send multiple text-delta events for each piece of text, and finally a text-end event to complete the sequence.

Each of these message parts needs to share the same ID (textPartId in this case), which is generated using crypto.randomUUID().

The text is split into words, and each word is sent as a separate delta with a small delay to create the streaming effect:

This approach is really handy when you want to mimic a part of the AI SDK or just manually stream in some text. You can see the full example in our main.ts file.

Nice work, and I'll see you in the next one.