AI That Stays on Your Device

Many people imagine that an AI request must travel to a distant server. Some systems instead run a model directly on a phone or computer. Apple's technical description of its foundation models provides one example of using both on-device and server models. The important distinction is where a particular task is processed, not whether the product uses the word AI.

Consider a hypothetical note-taking app that summarizes a short note locally. That task would not need to send the note to a remote model for processing. A much larger task might require resources that the device cannot provide. Memory, processing speed, and power use limit what a local model can do, so developers face practical trade-offs.

On-device processing is therefore not a promise that every feature works offline or that an entire app never sends data anywhere. Accounts, backups, and other features may still use a network. A clearer privacy question is: what information leaves the device for this specific feature? Understanding the route of the data is more useful than relying on a broad product label.