Stream tokens in the UI

Create a LanguageModelSession and stream the reply. The framework yields snapshots of the text so far. Update the screen on each snapshot.

import FoundationModels

let session = LanguageModelSession()
let stream = session.streamResponse {
    "Write a two-sentence packing list for a day hike."
}

var visible = ""
for try await partial in stream {
    visible = partial.content
    // Bind `visible` to a SwiftUI Text view.
}

That sketch follows the public streamResponse path. Confirm the partial property name in the Xcode 27 SDK. Some builds expose content on the snapshot. Do not invent a second streaming API.

Show a typing cursor or a quiet status until the first token. Then replace it with the growing text. Do not block the main actor on the whole stream. The async sequence is the wait.

If you need structured output, Apple also documents streamResponse overloads that take a generable type or a schema. The clock work in the next lesson is the same: start before the call, mark the first yield, mark the last yield.

Keep instruction text in a registry, not in the view. Version prompts like product copy shows how.

Next, put numbers on that stream.