From the iPhone Action Button to a saved meal: a 16-second Fud AI demo

I wanted food logging to start with something you already know how to do: say what you ate.

This new 16-second Fud AI film follows one small interaction: press the iPhone Action Button, ask to log breakfast, and see the result.

The spoken exchange is deliberately short:

“Log two eggs and a latte for breakfast.” “Logged to breakfast.” Enter fullscreen mode Exit fullscreen mode Demo note: this is an animated product demo with illustrative meal data and synthetic voices, not a live screen recording. The immediate-save example has Confirm before saving turned off. It demonstrates the intended interaction; its edited timing is not a latency benchmark.

Apple documents the shortcut assignment in its Action Button guide. Fud AI also exposes the same entry point through Siri and a Control Center control on iOS 18 and later. You can always start from the orb on Home.

Confirm before saving is on by default. With it enabled, a proposed write appears as a confirmation card before changing the diary. Food saves offer Undo afterward. The animation's one-step save is an option, rather than the default everyone gets.

The integration uses a small App Intent. It opens the app and records a pending voice-session request. The app consumes that request when its UI is ready; shared App Group storage also carries requests made by the widget extension.

That handoff matters: launching the app and having the conversation UI ready are separate events. The shortcut needs to survive the gap between them.

The assistant then works through the app's tools. Confirmation and Undo remain visible parts of the interaction. I covered the shared tool contract and provider adapters in my earlier Live Voice architecture article.

For this cut, the camera moves continuously from the phone's Action Button to its voice interface. The Fud AI broccoli mark compresses into the button, the orb responds to the speech, and captions follow the spoken words. The result stays visible before the closing title.

The editable project separates the phone scene, UI, orb, captions and audio. The orb uses a WebGL adaptation of the app's native shader; deterministic browser rendering and FFmpeg produce the 4K, 60 fps export. Music ducks under the two voice lines so the request and result stay easy to hear.

The design brief was a complete, legible interaction in sixteen seconds: where it starts, what you say, and what changed.

Fud AI is open source. You can try it on iPhone, visit fud-ai.app, or inspect the source on GitHub.

For a voice logging workflow, would you keep confirmation on, or prefer immediate saving with Undo?