On-device transcription, explained: why your words never have to leave your iPhone

"Private" is the most overused word in software. The difference that actually counts is mechanical, not marketing: where does the processing happen?

What "on-device" actually means

On-device transcription runs the speech-to-text model directly on your iPhone's own chip. Your spoken words become text without the audio ever leaving the phone. There's no upload, no copy sitting on a company's servers.

What the cloud sees

Cloud transcription works the opposite way: your audio is uploaded, processed on remote servers, and often retained. Your private monologue now exists as a file on infrastructure you don't control.

Why it matters for a second memory

This is more important for a second brain than for a meeting tool. A meeting transcript is semi-public. But a personal memory holds your raw, unfiltered self: worries, half-ideas, things about other people. Capturing more of your life only makes sense if capturing it stays private.

How to tell if an app is really on-device

"Private" and "secure" on a marketing page don't tell you where your audio goes. A few quick checks do:

• Try it in airplane mode. Record a short note with no connection. If it still transcribes, the model is running on your phone. If the text only appears once you're back online, it went to the cloud.
• Search the privacy policy for "upload," "store" and "retain." On-device tools have little reason to mention shipping or keeping your audio; cloud tools usually have to disclose both.
• Look for a training opt-out — or its absence. If an app needs a switch to stop training on your recordings, those recordings were leaving your device in the first place.
• Check the device requirements. On-device speech models need a recent chip, so a minimum-iPhone requirement is often a sign the work happens locally.

None of these is proof on its own, but together they tell you quickly whether your words stay with you.

How Kalpa does it

Kalpa transcribes on your device by design. Your words stay on your iPhone and aren't turned into training data. The whole point is to capture freely, knowing it isn't going anywhere you didn't choose.