Which note apps actually use Apple Intelligence on-device?

Since Apple opened its on-device foundation models to third-party apps, “powered by Apple Intelligence” has become a checkbox claim. It’s worth knowing what that claim can hide. An app can use Apple’s on-device dictation to capture your words, then ship the text to GPT-class cloud models for everything intelligent, and still gesture at Apple Intelligence in its marketing. The capture is on-device; the thinking about you isn’t.

The distinction that matters is which parts of the AI pipeline run on the phone:

• Speech-to-text: Apple’s on-device speech engines transcribe on the neural engine. Many apps genuinely use this part.
• The language model: the summaries, questions, and drafts. This is where most apps quietly go to the cloud, because server models are bigger and the integration is easier.
• Speech output: if the app talks back, that synthesis can also be local or cloud.

Xtraktr runs all three on the device. Apple’s on-device foundation model powers the interview — the AI that asks you questions about your idea — and the drafting that turns your answers into usable text, with on-device speech recognition in front and on-device voices behind. That’s why the whole app passes the airplane-mode test with nothing degraded.

A phone-sized model is not a data-center model, and apps that are truly on-device should be designed around that. Xtraktr’s design leans into it. Interviewing you and structuring your words is exactly the kind of work a local model does well, as opposed to generating essays from nothing. The constraint shapes the privacy story: your words never touch our servers because the architecture never needs them to.

If you’re evaluating any app in this category, skip the marketing page and run the test: airplane mode, then try every AI feature. The on-device claim survives contact with the toggle or it doesn’t.