Open the App Store and half the “AI” apps do the same quiet thing. You type something personal, and it is shipped off to a company’s server to be processed, and often kept. The intelligence lives somewhere you can’t see, on hardware you don’t own.

Our AI apps work the other way around. The model runs on your device. What you polish in Better Writing, what you ask Second Brain, and the phrases you practice in Pocket Lingo do not go to a Studio AM AI server. There is no Studio AM account or processing log with your words in it. Storage is a separate choice: Better Writing’s saved writing and correction-review progress, and Second Brain notes, stay local unless you enable each app’s optional private iCloud sync, which Studio AM cannot read.

Why we made it the hard way

On-device is harder. The model Apple builds into the iPhone is small (far smaller than the giant models in the cloud) and getting reliable work out of it takes real engineering. The easy path was to wire everything to a big cloud model and get smarter answers on day one.

We didn’t, for one reason: the data these apps touch is about as personal as data gets. What you are trying to say. What you want to remember. What you spent and where. The words you are embarrassed to mispronounce. We would rather not be able to see that than ask you to trust a promise that we won’t look.

There is a quieter benefit too. A server that answers every request costs money for every request, which is why so many AI apps need a subscription just to cover their bill, and why “your data” starts to look like a way to pay it. When the AI runs on hardware you already own, that whole incentive disappears. We are not renting you access to a server, and we are not in the data business.

What “on your phone” actually buys you

  • AI processing is private by construction. Your text is processed on your hardware rather than sent to our server. Each app’s privacy page separately explains local storage and optional Apple services.
  • It works on a plane after setup. Once a compatible model is ready, the intelligence is already in your pocket. The first model or voice download, a purchase, optional sync, or an external link still needs a connection.
  • It does not depend on a Studio AM inference service. There is no per-request AI server to keep alive. App access and archive rights are separate product decisions, which each current product page explains.

The honest trade-off

On-device intelligence needs compatible hardware and iOS or iPadOS 26. Apple’s built-in model may be available immediately; some Studio AM apps can instead offer a downloaded local model on other supported devices. The exact download and memory floor vary by app, so the product page and in-app setup, not a rough phone-generation list, are the source of truth.

For Better Writing specifically, the lighter download is about 0.9 GB and requires at least 3.5 GB of physical memory; the enhanced model is about 2.6 GB and requires at least 5.5 GB. Devices below the downloaded-model floor need Apple’s compatible built-in runtime. Editing works offline once one of those compatible runtimes is ready. Optional private iCloud sync, purchases, model and natural-voice downloads, exports and Share actions, and a user-approved DuckDuckGo handoff remain explicit network paths.

There is a second trade we make on purpose. The small model does not get a blank check. We ground each workflow in supplied content and add checks around its output: Better Writing’s Rewrite mode aims to reject unsupported details, while its Draft and Reply tools work from notes or a confirmed transcript you provide. Second Brain can decline when it cannot ground an answer in saved notes. These safeguards reduce risk; AI output can still be wrong and should be reviewed. Smaller and careful beats bigger and loose, especially when the subject is you.

This is the feature, not the compromise

We don’t treat on-device as a limitation we are tolerating until the cloud gets cheaper. It is the reason the apps are worth trusting in the first place. The best thing we can do with your most personal data is never hold it.

Small studio. Private by design. That part isn’t going to change.

Studio AM