Your skin is constantly emitting acetone, ammonia, aldehydes and CO₂. Researchers have measured these compounds one lab study at a time, on a few dozen people each. There is no open baseline dataset. SkinSenseAir is a wrist-worn instrument that measures your skin's emissions alongside your heart rate and your environment — and 1,000 of us are going to build that baseline together.
We would rather show you the instrument working than tell you it does.
The campaign film. Two and a half minutes on what your skin emits and why nobody has mapped it.
Out of the box to a live trace on your wrist, unedited and in real time.
Route, skin gas and heart rate recorded together — then we ask Claude what it makes of them.
The five sensors, the sealed chamber, the calibration routine, and what each one can and cannot see.
Every gesture on the device, in real time — paging, rotating, sleeping, and the vent prompt firing mid-run.
This is the honest reason to back this project. A single reading from a single wrist tells you almost nothing — there is no published range to compare it against. But 1,000 instruments running the same sensors, the same firmware and the same 42-column schema, in different climates and on different bodies, produces something that does not currently exist: an open, pooled baseline of human skin emissions with environmental and biometric context attached.
Seal the chamber against your forearm, pick an activity, press start. Skin gas, chamber climate, heart rate, HRV, SpO₂, GPS and altitude are logged together every few seconds.
Contribution is opt-in and per-session. You pick each run individually. Location is coarsened to a 10 km grid before it leaves the device, and there is no account to create.
Shared runs are aggregated into a public dataset, released under CC BY-SA. Anyone can download it. We are not selling it, and we are not the only ones who get to use it.
Once there is a pool, your dashboard can finally answer "compared to what?" — showing where your run sits against the distribution instead of floating without reference.
Every photograph below is the actual hardware. Nothing here is a visualisation.
Worn position, strap fitted, screen live.
SHT40, SGP40, STCC4, BME690 and the MiCS-6814 array, open and vented.
Silicone cup against the forearm, the part that makes a flux reading possible.
COLMI R02 on the finger, SkinSenseAir on the wrist, paired.
Waypoints coloured by CO₂, with heart rate at each point.
42 columns, on the card, yours to take. No cloud lock-in.
Analogue hands on the 1.47″ panel. Faces are JSON on the card.
The same unit, four button presses apart. Both laid out separately.
The prompt that stops a saturated chamber quietly ruining a run.
The dashboard is the deep end. But most of the time you are on a walk, in a lab coat, or halfway through a session with one hand full — so the instrument has to be readable and operable on its own. Seventeen pages on a 1.47″ panel, driven entirely by one button.
An analogue dial with real hands, plus the date. The device has no RTC battery and, in field mode, no route to a time server — so the browser sets the clock the moment you open the dashboard, then corrects the drift every six hours.
The four numbers worth glancing at, in 48 pt type you can read at arm's length. The eight MiCS channels are split across two pages rather than crammed onto one.
Two large arc gauges with the live figure beside them, and the chamber flux in ppm per minute — the one reading here that does not depend on how much ambient CO₂ was trapped when you sealed the cup.
Skin temperature, chamber humidity, dew point and absolute humidity — the numbers you need to interpret everything else, because humidity is the dominant confounder in any metal-oxide reading. Then the five derived indices as bars.
A fifteen-minute trace of any channel, and the ring's vitals on the device itself — heart rate, SpO₂, HRV and body temperature, without reaching for a phone.
Local conditions and your position, both sourced from the phone and pushed down to the unit — so the device shows them even when it is running as its own access point with no internet of its own.
Charge state on a proper lithium discharge curve rather than a straight line — because a straight line reads 35 % on a nearly flat pack and then sits at 100 % for the first half hour of a session.
Files on the card, network and IP, free memory, the run label, and the AI analyst's summary — all reachable without opening a browser.
There is one button on the unit. Gloves, a lab coat pocket, a wrist strap under a sleeve — the fewer things to find, the better. So it counts.
Landscape or portrait, laid out separately rather than merely rotated — the type sizes and the page structure both change to suit the shape.
Both power states are previewed on screen while you hold, and nothing happens until you let go — so if you overshoot, keep holding and read what it says.
A sealed chamber saturates, and a saturated chamber reads as a falling metabolic rate rather than as a full box. This is the setting most likely to save a run.
A capture that stopped itself at minute ten would look like it had worked, which is worse than not running at all. So it never interrupts one.
The device stands alone. But for the analysis that needs a bigger screen, the dashboard is served by the unit itself and opens in any browser — nothing to install, no account, and it keeps working if we disappear. Here is every screen and what it does.
The screen you watch while a session runs. A four-pen strip-chart recorder scrolling the last 15 minutes, three analogue gauges with damped needles, and the sealed-chamber flux reading.
All eight MiCS-6814 gas equivalents at once, plus the raw sensor resistances underneath — which are the numbers that actually matter for machine learning.
History for any channel over 5 minutes, 15 minutes, an hour, or the whole session, on a real clock axis rather than a sample count.
Your COLMI R02, read directly over standard Bluetooth profiles. Heart rate, HRV computed from RR intervals on-device, SpO₂, skin temperature, steps and ring battery.
Optional smartwatch support for people who already own one. Vitals, blood pressure estimate, respiration, stress index, and a single-lead ECG trace on proper ECG paper.
Scan, identify and connect wearables. Devices are fingerprinted by name and by the Bluetooth services they advertise, so you can see what a device can actually provide before pairing it.
Position from your phone's GNSS, with an honest readout of which positioning source is in use — because that determines whether you get altitude and speed at all.
Your walk, coloured by any channel you choose. Tap any waypoint for every reading recorded at that moment — gas, biometrics and position together.
Everything is written to the SD card as plain CSV. This screen manages it: browse by day, preview any file as a chart, download it, or delete it.
The correlation matrix. Six biometrics against six gas channels, Pearson r, computed only where at least 30 samples line up within six seconds of each other.
A conversation with Claude about your own data. It receives your window statistics, your correlations, your route and a built-in table of published skin-gas research.
Network, appearance, power and calibration. Join a router or run as a standalone access point — rename it, or lock it with a password — and set the run label that becomes your machine-learning target.
The protocol, written into the device. Burn-in, calibration, chamber materials, and a plain explanation of what each reading can and cannot tell you — including the parts that are estimates.
Skin emissions move with perfusion, exertion and skin temperature. Without heart rate and HRV recorded at the same moment, you cannot tell a metabolic signal from someone simply walking faster. That is why the ring is bundled rather than sold as an upsell — the dataset is substantially weaker without it.
The first 1,000 units ship with a COLMI R02, chosen for one unglamorous reason: it speaks the standard Bluetooth heart-rate and pulse-oximeter profiles, so we can read it directly instead of going through a vendor cloud.
Every SkinSenseAir has a built-in analyst powered by Claude. You can ask it anything about your session in plain language and it answers with your actual numbers — the window statistics, the correlations between your ring and your skin gas, your route, and a reference table of published skin-gas research that ships with the device.
Here is the part we are most proud of. The analyst is instructed to look for the boring explanation first. Humidity, movement, skin temperature and chamber accumulation will produce impressive-looking correlations all day long. The prompt requires it to name the confounder before it names a mechanism, and to say plainly when the hardware cannot support a conclusion. A tool that tells you what you want to hear is worse than no tool.
Real output from the shipping firmware. Your API key, your account, your data — we never see it.
Both tiers include the ring while the first 1,000 last. Both ship with the full CSV export, the open dashboard, and the STEP and OpenSCAD files for the case.
We would rather lose a pledge than overstate what this instrument does. That standard is the same one built into the AI Analyst.
No. Contribution is opt-in per session, and the device is fully functional if you never share a single run. Everything logs to the card either way and the CSV is yours.
The sensor columns, your activity label, and a location coarsened to a 10 km grid square. Not your name, not your address, not a precise track. There is no account.
If it exposes the standard Bluetooth heart-rate and pulse-oximeter profiles, yes — the firmware reads those directly. Oura is the notable exception: its Bluetooth link is encrypted and bound to their app, so it needs their cloud API instead.
No. Seventeen pages on the device cover every reading, and one button starts a log, captures a baseline, arms a flux run and changes the ML label. The phone dashboard adds the deep analysis — correlation matrix, route map, AI analyst — plus two things the unit cannot get on its own: the clock, and local weather. Open the dashboard once and both are set.
Because it is the difference between data and confident nonsense. A sealed chamber saturates — gases accumulate against the emission gradient until the reading stops tracking what your skin is emitting and starts tracking what the cup has already collected. The flux flattens toward zero and looks exactly like a falling metabolic rate. Set an interval and the device prompts you to open the chamber; acknowledging it also disarms the flux run, because venting destroys the baseline it was measuring against.
Yes, and it takes about a minute. A face is a small JSON file on the SD card describing colours, tick style and hand geometry — every key optional. Write it in the dashboard's text box and it saves straight to the card, or drop a file on the card directly. Four are included to start from. We chose a style file over image faces deliberately: images would need a decoder, a filesystem driver and a conversion step, for artwork most people would not change.
The battery page shows charge on a real lithium discharge curve, not a straight line, and estimates remaining runtime from the drop it has actually observed. Be aware there is no current shunt on this board — the estimate is extrapolation, so it shows dashes for the first ten minutes rather than inventing a figure, and it is wrong right after a load change. You can also set an auto-sleep timer; it never interrupts a logging run.
Concentration in a sealed cup depends on how long it has been sealed. Emission rate does not. The flux figure is the one that means something physiologically.
The dashboard, firmware and case files are published. The AI Analyst uses your own Anthropic API key, stored on your device — we never see your data or your key.
The hardware works today; the risk is manufacturing, not invention. The realistic risks are tooling delays and sensor lead times, both of which we have quoted and budgeted for. Kickstarter is all-or-nothing, so you are not charged unless we fund.