You're not reading numbers. You're listening to your night.

The number isn't what changes you

I had been using my own prototype for a week. The model wasn't great. The UI was worse. Every morning I'd swipe through the report, look at an AHI somewhere between 22 and 28, and feel nothing. The number didn't move me. I'd had abstract numbers told to me before by doctors and I'd kept doing the same things.

Then one morning I tapped the play button next to a flagged event at 03:47.

Eleven seconds of silence. Then a sound I'm going to call a gasp because I don't have a better word for it. It was me. Eleven seconds at 3:47 in the morning, my body just stopped breathing, and then it started again.

I sat up in bed. My wife had moved to the guest room two months earlier. I made a sleep specialist appointment that morning, before the coffee was done.

What every event captures (and why every consumer sleep app stops short)

Most apps give you a chart line. We give you a play button on every event. Tap it, hear the actual audio that the AI classified — the snore, the silence, the gasp.

• Snores — palatal, tongue-base, epiglottic. Different sounds, different anatomical sources, different things you can do about each.
• Breathing pauses — the silences. Clear stops long enough to count seconds.
• Reduced breathing events — the partial restrictions, harder to hear but acoustically detectable.
• Gasps and arousal breaths — the recovery breath that often follows a pause. The one I heard.

Most people listen to maybe 5 events from their first night, then stop listening every night. The data keeps doing its work without you pressing play. But on night one, before you trust the numbers, listening once usually changes things — for me, and for most Lab Members we've talked to about it.

Why on-device — and what that means for your audio

Your sleep audio is more intimate than most people realize. It contains things your conscious memory doesn't. It contains your partner. It contains conversations you had after going to bed. It contains the dog scratching at the door at 2 a.m.

So here's exactly what happens with it:

• The AI runs on your phone. Audio doesn't get sent anywhere by default.
• Audio is stored encrypted in the app's local storage. Delete any individual clip or all of it with one tap in Data Management.
• The only way audio leaves your phone is if you explicitly tap Export or Share on a specific clip — usually to send to a doctor, occasionally to a partner who didn't believe you.
• If your phone is lost or stolen, local encryption protects it. Biometric lock is supported on the data screen.

I designed it this way because if I were the user, I would not put my own sleep audio in someone else's cloud. That's the actual reason — not a privacy badge for marketing.

Free vs Pro on this feature

Free: 3 audio plays per report. Enough to verify a few key events on night one and decide if you want more.

Pro: Unlimited plays. Some events are worth listening to three times. The first 7 days of Pro are free; cancel through Apple or Google before day 7 and you won't be charged.

The reason for the Free limit isn't upsell — audio playback is computationally cheap but bandwidth-heavy when re-encoded for sharing. Three plays cover the "do I trust this?" check; unlimited covers the "I want to verify a pattern across 30 nights" need that emerges later.