Streaming Text

We saw how easy it is to get some text from an LLM, but our users are used to read a stream of text that is getting written as the LLM respond.

That’s why streamText is so powerful, it handles all the streaming for you

import { google } from '@ai-sdk/google';
import { streamText } from 'ai'; // <- This is what you're looking for!

const model = google('gemini-2.5-flash-lite');
const prompt = 'What is the capital of France?';

const stream = streamText({ model, prompt }); 

for await (const chunk of stream.textStream) {
  process.stdout.write(chunk);
}

This time, instead of leveraging the generateText function that we saw in the previous lesson that just prints text on the screen, we let the LLM write down the answer to our prompt.

One little difference with our previous exercise has been the way we print out text.

For the previous lesson, we’ve used console.log to have the generated text in our terminal. And that was great for what we had to do, but with streamText ↗ (opens in a new tab) we need to have something different because with console.log we get a newline character at the end of each call, and that would’ve messed our display.

That’s why we use process.stdout.write, with it we can write to the standard output stream without adding any extra characters or formatting