A light and a voice on your nightstand. It fires when you open a watched app past your bedtime, and it wakes you with the thirty-minute dawn that actually has a trial behind it.
Almost everything else on a nightstand measures the hours in between — the part you have the least control over. This works on the two edges, where an intervention can still change the outcome.
23:41
Bedtime breach
The light goes amber and one line is spoken out loud, into the room. There is no notification to swipe away, no app to close, no streak to protect. Your phone reports that the app opened; the device decides on its own whether the hour matters — because a rule that lives in the phone stops matching at midnight.
07:15
Dawn
Deep red while your eyes are still shut, arriving at cyan as they open. At the wake time the light holds and the day's calendar is read aloud — or a clip you recorded in your own voice. Ten minutes later, what you had planned for that weekday.
You set one number. Everything else is computed from it, so the parts cannot drift out of step with each other.
| Time | Job | What happens |
|---|---|---|
| W−30 | dawn_ramp | Deep red climbs toward 480 nm cyan while you are still asleep |
| W | wake_hold | The light holds. The calendar brief speaks, or a clip you uploaded yourself |
| W+10 | plan brief | What you had set out to do on that weekday, read back |
| W+20 | off | The light goes out |
Van De Werken and colleagues ran artificial dawn across the final thirty minutes of sleep in sixteen adults who reported difficulty waking, and measured less subjective sleepiness and more subjective alertness on waking. So the ramp ends at the alarm. A light that begins at the alarm is a different intervention with a different evidence base, and we are not going to quietly claim the first study for the second design.
What that trial found was self-report. Performance, core body temperature and the cortisol response were all null results. The honest sentence is less groggy, by self-report, and nothing past it.
A closed eyelid attenuates short wavelengths far more than long ones, so thirty minutes out, blue is mostly stopped at the skin and red is the part that still arrives. Melanopsin — the retinal pigment that carries the circadian signal — peaks near 480 nm, which is where the ramp finishes, as the eyes open.
The path runs red, violet, blue, cyan, and is routed around the neutral point deliberately. On a ring with a hard power budget, white spends most of its output in the parts of the spectrum the circadian system barely looks at.
This is a cue, not a lightbox. It is not going to move your body clock on its own, and we would rather say that here than in a footnote.
A 60 GHz radar module reads presence and breathing from across the nightstand. Nothing to wear, nothing under the mattress, nothing to charge. It resolves a body in a bed — it cannot resolve a face, and there is no camera and no microphone anywhere in the device.
If the radar is confident and current that the bed is empty, the morning routine stands down. If the reading is stale, ambiguous, or the sensor is not there at all, the device behaves exactly as it would without one. Absence of evidence never changes what it does — which is the only version of this that is safe to sleep next to.
The wake alarm is held on the device, not in the phone and not on a server. A dead battery, a dropped network, or an app the operating system decided to stop running cannot silently un-set your alarm. The phone refreshes the wake time; it does not own it.
When you were in bed, and for how long, every night — that is the most personal file this device produces, and it is written to the device's own storage. There is no account to create and no subscription. It is priced once, and then it is yours.
Sleep debt comes back as accumulated shortfall against a need computed from your own answers — you are three hours down over the fortnight. Not a score, not a stage breakdown, no chart dividing the night into REM and deep. A score is a compression, and the thing it compresses away is the arithmetic you would need in order to disagree with it.
The first-run questionnaire derives your bedtime and your alarm from mid-sleep on free days, following the Munich ChronoType Questionnaire, and it asks about work days and free days separately — a work-day answer measures your alarm clock rather than you. It is arithmetic on what you report. It is not a diagnosis, and it is not screening for anything.
Each of these is a plausible-sounding sentence about a device like this one. None of them is supported by the research behind it, so none of them appears anywhere else on this page.
| Part | Detail |
|---|---|
| Compute | ESP32-S3, dual core. Every decision is made on the device |
| Presence | 60 GHz mmWave radar. Contactless. No camera, no microphone |
| Light | 24 high-CRI addressable LEDs, aimed up into a diffusing dome rather than out through it |
| Sound | 2–3 W full-range driver in a sealed chamber. It speaks, so it is not a beeper |
| Clock | Real-time clock with its own cell, so the alarm survives a power cut |
| Setup | Bluetooth LE pairing from the phone. No account |
| Size | Roughly 150 mm across and 130 mm tall, with a flat top |
| Power | External certified 5 V supply, with headroom for full output |
Target price is $150–250, paid once, with no subscription of any kind. That figure comes from the current bill of materials and the usual multiple for hardware at low volume — it is a target, not a price anyone is being charged.
Every physiological statement on this page traces to one of these. The second half of each entry is the part that usually gets left off.
The bench build works and is running nightly. The version that arrives in a box is next. If you want to hear when there is something to put on a nightstand, leave an address — that is the only thing this list is for.