Causal
Devices, your data, and what causes what
Measure the room, the body and the day.Find out what is causing what.
Causal is a family of devices that measure the health of your life, starting with sense, a monitor for the room you sleep in. An app puts them beside what your phone and wearables already record, and what you tell it you feel, then works out which of it is connected: causes, with their sizes, not a chart that moves together.
A worked example
Some mornings you wake up foggy.Nothing you own can tell you why.
The watch says you slept. The room has no opinion. The fog is only in your head, which is where nobody is measuring. Causal measures all three, on one clock.
sense · Bedroom
CO₂ in the evening
1,340 ppm
windows shut since the cold snap
Watch
Deep sleep that night
−19 min
against your usual
You, next morning
Brain fog, logged at 8:10
6 / 10
on 7 of the last 9 such mornings
Causal
The one thing worth changing
Air the room
before bed, then check in a fortnight
What a dashboard reports
“Your sleep quality and your indoor air are related this month.” Both went the wrong way, so both get shown. Which one to change is left to you.
What Causal estimates
Evenings with CO₂ above 1,000 ppm were followed by about 19 fewer minutes of deep sleep, and the foggy mornings follow from the lost deep sleep, not from the air itself. So the one change worth making is airing the room before bed, and the model will tell you in a fortnight whether it worked.
Illustrative figures. Effects are estimated inside your own history, so the numbers on your model will be your own, including the inputs that turn out to do nothing.
The system
Hardware that measures your life,and one model that reads all of it.
Devices
sense is first: a wall-mounted monitor that measures the air, the light, the sound and the walls of a room. A bedside device and a ceiling device follow. Each one is another set of variables that can rule a cause in or out.
Your phone and wearables
Sleep, heart rate variability, resting heart rate, respiration, steps, temperature: Causal reads what Apple Health already holds, from the watch or ring you already wear. Nothing new to keep.
The app
Where the room, the body and what you feel sit on one timeline. Log a symptom in two taps. Read what the model found, with the numbers and the doubt beside them.
The model
Causal inference over the lot: which inputs changed which outcomes, by how much, and how sure we are. Estimated inside your own history, never from a population average.
The first device
sense measures the roomthe way a building physicist would.
A monitor the size of a light switch, on the wall of the room you sleep in. It measures everything a room does to you, from the particles in the air to the damp inside the wall behind it, and sends the lot to your Causal model every minute.
Air
- Particulates PM1 · PM2.5 · PM4 · PM10
- CO₂
- Volatile organic compounds (index)
- Nitrogen oxides (index)
- Formaldehyde
Climate
- Temperature
- Humidity
- Barometric pressure
- Door and window events from pressure
Light
- Brightness
- Colour temperature
- Flicker
- Daylight and evening light exposure
Sound
- Noise level, dB(A)
- Night-time noise against WHO guidance
- Alarm and appliance patterns
Walls
- Surface temperature, coldest patch (thermal array)
- Moisture inside the wall (electrodes)
- A cabled probe for the damp corner
- Vibration and knocks
What it can name
Not a number on a dial.A problem, with a name.
Readings become diagnoses of the room: what is wrong, since when, and how sure sense is. The confidence words are the honest ones from our own testing, not marketing.
Condensation risk and cold bridges
The coldest patch of wall against the room's humidity gives the surface humidity and a mould index, per hour.
Under-ventilation, and how many people
How fast CO₂ rises and decays after the door closes.
Cooking, smoke or a heater
The size mix of the particles together with CO₂ and the gas indices.
Noise that reaches your sleep
Night-time levels against the 30 dB(A) WHO guideline, and what pattern the noise has.
Flicker, and too little daylight
Light sampled fast enough to see the mains, and daily exposure against the circadian guidance.
A solvent, paint or new-furniture source
The VOC and NOx indices with formaldehyde. It can say “solvent-like”; it cannot name the compound.
Rising damp against condensation
A shallow and a deep electrode pair behind the back plate, read against an install baseline.
sense
Wall-mounted. Air, climate, light, sound and the walls. Wi-Fi 6 and Thread, no hub, no subscription for any reading.
Bedside
Breathing and movement through the night from the nightstand, with the room's own numbers beside them.
Ceiling
Who is in the room, where, and how still, from above, without a camera. One unit per room.
The app
Four places and a button.Nothing to keep up with.
Black and white, one serif for the names of things, and the numbers in plain words. The same design language as the rest of what we make.
Today
The room now, your body last night, what you logged, and the one effect most worth knowing about this week.
Room
Every reading from sense with a plain-words grade, the last 24 hours as a line, and the problems it can name for this room.
Causes
What the model found: each effect as a sentence, its size, its interval, and the evidence behind it. Tap one to see both series aligned.
Log a symptom
Two taps from anywhere: pick from 259 symptoms in plain words, slide the severity. Symptoms are the outcomes the model works back from.
You
Which sources are connected, what each one shares, and the buttons to export everything or delete it all.
Put sense on the wall, connect your phone
sense joins your Wi-Fi from the app in a minute. Apple Health is one permission. From then on the room and your body are measured without you doing anything.
Log what you feel, when you feel it
A headache, a short night, a foggy morning. Symptoms are what the whole thing is for: they are the outcomes, in your words, that the model works back from.
Everything on one clock
A reading every minute, a sleep score once a night, a symptom when it happens. Causal aligns them onto a shared timeline so they can be compared without quietly inventing agreement.
Change one thing, then check
Causal suggests the single input most worth testing and watches what follows. Each test sharpens your model. It is an n-of-1 trial that runs itself.
The science
Correlation is cheap.Causation is the work.
Four problems stand between a chart that moves together and a claim you can act on. Causal is built around them.
Direction
A stuffy room and a bad night arrive together. One of them is upstream. Causal uses the order in which things actually change — minutes for CO₂, a night for sleep, a morning for the fog — to work out which way the arrow points.
Confounders
A cold snap shuts your windows and changes your sleep at once. Blame the air and you have treated a symptom. Causal looks for the common cause before crediting the visible one.
Natural experiments
A week away, a new purifier, a guest in the spare room, the heating coming on. Your life is already running interventions on itself; those breaks in routine are where the causal information lives.
Honest uncertainty
Every effect comes with an interval and a grade. A few effects are large, most are small, and some cannot be identified from the data you have. Causal says nothing at all when the data will not support a claim.
Correlation tells you what happened together.Causation tells you what to change.
How symptoms connect
When we get sick, what we notice firstare symptoms.
These are the 259 symptoms Causal lets you log, and how medical research links them: two sit close when the literature describes them in the same conditions, across millions of papers. Tap one to see what it travels with, and why. It is a map of the literature, not a diagnosis.
Data: Zhou et al., Human symptoms–disease network, Nature Communications 2014 (CC BY 4.0). Open the full map.
Privacy
Causal asks for a lot of your life.It stays yours.
Read the privacy policy and our security practices.
Per-source control
Every input is connected deliberately and can be disconnected the same way. Removing a source removes its history from your model.
Yours to take
One button exports everything Causal holds about you as a single file. Another deletes it all, at once, for real.
Never sold, no ads
Your readings, your health data and your symptoms are not sold, brokered, or used to target you with anything. Encrypted in transit and at rest.
Inspectable conclusions
You can see which inputs produced an estimate, how many days it rests on, and how confident it is. An explanation you cannot open is not an explanation.
sense is being built now.Be in the first rooms it goes into.
We are opening Causal slowly, to people who want to know what their home and their habits are doing to them, and to researchers who want to argue with us about identification.