JC JC Mobile App Studio
Subscribe JC

On device AI , Thursday June 5, 2026

What Apple Intelligence can and cannot do without the cloud.

Apple Intelligence is not one thing. It is three layers stacked together, and they send your data to very different places. Here is which is which.

Apple Intelligence gets talked about as if it were a single feature, but under the hood it is a layered system that decides, request by request, where the work should happen. For anyone who cares about privacy, the useful skill is knowing which layer a given feature uses, because that determines whether your data stays on your phone. There are roughly three layers.

Most of the everyday features run directly on your phone using Apple's on device models. Summarizing text, the Writing Tools that rewrite and proofread, smart replies, prioritizing notifications, and many of the language tasks apps build with Apple's Foundation Models. These work without a network, your content does not leave the device, and they keep going in airplane mode. This is the layer to prefer for anything sensitive.

When a request is too big for the phone, Apple can send it to what it calls Private Cloud Compute, its own servers running on Apple silicon, designed so that your data is used only to answer the request and is not stored or made readable to Apple afterward. It is a genuinely stronger privacy design than typical cloud AI. It is still important to be clear eyed that this is the cloud, your request did leave the device, and you are trusting Apple's architecture and its independent verification that the design holds.

For the broadest, open ended questions, Apple can hand off to an external provider, with ChatGPT being the headline integration. This is the layer to treat with the most care. Apple asks for your permission before sending anything out, and the request leaves Apple's world entirely and lands on a third party's servers under that company's policies. When you see a prompt asking whether to use ChatGPT, that is the system telling you, plainly, that this one is leaving the building.

Sponsored

You do not need to track the plumbing on every tap, but a simple heuristic helps. Focused tasks over text you already have, summarize this, rewrite this, tend to be on device. Big general knowledge questions and anything that visibly asks for your consent to use an outside assistant are leaving the device. And the simplest test of all still works: if a feature keeps functioning in airplane mode, it is running locally.

Apple Intelligence does more on device than most AI systems, and its cloud layer is designed more privately than most. But on device is not the same as the whole thing is on device, and the marketing umbrella covers all three layers at once. The good news is the system is mostly built to tell you when you are crossing a line, especially the external one, if you know to watch for the prompt.

The apps from this studio lean on layer one, on device AI with Apple's Foundation Models, and are upfront about anything that would go further. To see what that looks like in practice, the full lineup is at jcmobileappstudio.com.

— JC Mobile App Studio

Sponsored

More from the blog

Plain-language writing on workers' rights, investing, and on-device AI.

Read the blog

Contact

Get in touch.

Beta access, app ideas, bug reports, or partnership questions, the inbox is open.

Support available in English and Español.