On-Device AI
Run conversations locally with Apple Intelligence or Gemma, online or off.
On-device AI answers conversations using a model that runs on your phone instead of Wyatt's servers. It is what makes chat work with no connection, and it can also be used while online when you would rather your messages never leave the device.
Set it up in Settings > AI > On-device AI. This is the only place it is configured — the Offline Access screen links here and shows the current state, but the download, the engine choice, and the storage all live in AI settings.
Choosing When It Runs
| Mode | Behavior |
|---|---|
| Automatic | Uses cloud AI while you're online and on-device AI when you're offline. |
| Apple Intelligence | Runs every conversation on this device using Apple Intelligence. |
| Gemma | Runs every conversation on this device using Gemma. |
Automatic is the setting most people want: full cloud quality with a working fallback in a tunnel or on a plane. Pick one of the two engines explicitly when you want everything to stay local regardless of connectivity.
Full vs. Basic
On iOS there are two tiers, and the difference matters:
- Full on-device AI (recommended) downloads a model to the device. It keeps more of the conversation in mind and can read long documents.
- Basic on-device AI is built into iOS and needs no download. It remembers less of the conversation and cannot read long documents.
Full on-device AI needs at least 6 GB of memory and a recent OS. On devices that do not meet that bar, Wyatt shows Full on-device AI isn't available on this device and offers the basic tier instead.
Setup and Storage
Full on-device AI downloads a model, which is large. Start the download on Wi-Fi and leave the app open — on iOS a Live Activity shows progress so you can track it from the Lock Screen.
The Offline Access screen mirrors the state of that setup:
| State | Meaning |
|---|---|
| Checking device support… | Wyatt is working out whether this device qualifies. |
| Not available on this device | The hardware or OS does not meet the requirements. |
| Not set up yet | Supported, but no model has been downloaded. |
| Downloading… | The model is being fetched. |
| Ready to answer chats offline | Set up and working. |
| Download failed — tap to retry | The download did not finish. |
You can remove the downloaded model from AI settings to reclaim the space. On- device AI then falls back to whatever is available — the basic tier on iOS, or cloud-only.
What to Expect
A model running on a phone is smaller than the models Wyatt uses in the cloud. Answers are shorter and less capable, especially on long documents or multi-step reasoning. That is the trade for working with no connection and keeping the conversation on the device. When you reconnect in Automatic mode, the next message goes back to the cloud model.