On-device AI apps that keep data on the phone

We build apps where the AI runs on the device itself. No server sees the user's voice or data, the feature works offline, and there is no per-request cloud cost. Misbar Quran is our working example.

Runs: On the phone, offline
Privacy: Audio and data stay on the device
Shipped: Misbar Quran (Android, iOS)

Why on-device

Privacy by design. When recognition runs on the phone, the recording never leaves it. There is nothing to store, leak or delete on a server.
Works offline. The feature is available on a plane, in a basement or with no data plan.
Instant feedback. No round trip to a server, so results follow the user as they speak.
No cloud bill per use. Cost does not grow with every request a user makes.

What we have built

Misbar Quran listens while a user recites from memory and follows them word by word with speech recognition running on the phone. It flags missing, swapped or skipped words, and it can find a verse from a few spoken words. The app is free, has no ads, and the voice never leaves the device.

How we approach an on-device AI feature

1. Measure before we promise

We test the model on real recordings, including the mistakes real users make, and check the result on mid-range phones, not only the newest ones.

2. Design for being wrong

Any model makes mistakes. The interface says so where it matters. Misbar marks its tajweed notes as possibly wrong rather than presenting them as final.

3. Compare old and new on the same data

When we change the model or the matching logic, the new version is compared against the old one on the same set of recordings before it ships.

Have a feature that should run on the phone? Email intelgets@gmail.com.