Photograph the label. Numbers done.
No typing, no searching a database that has never heard of your yoghurt. Cox reads what is printed on the packet — and does the rest of the maths itself.
The model reads. The code does the maths.
There is a nutrition table on the photo — Cox reads it off and notes which column it read. Converting to 100 grams, multiplying by your amount, summing it up: that is the code, not the language model.
That is the difference between a figure that was read and one that was guessed. If there is no serving size printed, Cox says so instead of inventing a plausible one. And if kilojoules and kilocalories are swapped, the row is rejected.
From photo to meal. Sample data.
No label? Then an estimate — marked as one
A plate in a restaurant has no table. Cox then estimates item by item with grams and macros, adds it up itself and enters it straight away — no confirmation step, but visibly an estimate, with how confident it is.
What the model originally saw stays stored. If you correct it, your figure replaces its own — and later you can read back how far off it was.
The day with its balance. Sample data.
The second time, the barcode is enough
The scanner runs in your browser — the camera image never leaves your device. The first scan of a product comes up empty; then you photograph the label once, and the barcode travels into your own food list with it.
From the second scan on, the meal is there without a single model call. Plus favourites, serving presets and “same as last time”.
Four ways to a meal. Sample data.
So the coach understands why today's session didn't go anywhere.
Food is not a separate diary next to your training. It is one of the quantities Cox uses to explain what happened.
A balance that keeps up
Your expenditure is estimated from your own data and kept updated, not taken from a formula for everybody. Today is projected forward while it is still running.
Protein across the day
Not just the daily total but the distribution: how many occasions, how big the largest gap. Both count towards what you build in the gym.
Carbs around the session
Before, during and after the hard session — computed against your actual start times, not an assumed daily routine.
Measured or estimated — the page says which. Sample data.
Honest beats complete: Cox separates measured rows from estimated ones and shows both — by row and by calorie. That is exactly where the two numbers drift apart.
And on the phone, where the eating happens.
Logging, paging back, checking where the expenditure came from — the same numbers as in the browser, just where you actually are.
The day, and what is in it
Calories and all four macros against your target, and under them every meal on its own with its photo. What is estimated says so; what you correct replaces the estimate.
Paging back works on a day with nothing in it too — and then the screen says so rather than sitting empty.
The day and its meals. Example data.
Week and month, with expenditure marked across
One column per day for what you ate, and expenditure as a mark across it — two bars would be sixty-two strokes on a palm-sized screen over a month.
Tapping a bar shows what it is made of. The average leaves the running day out: a projection is not a measurement and has no business inside a mean.
The week, intake against expenditure. Example data.
And you can see where the expenditure comes from
Which source the number came from, and what it is made of — resting and active split apart. If several devices deliver, you choose which one counts; connected ones sit at the top, and both states are named ("delivering data" and "nothing delivered yet").
If you wear no device, you set a fixed daily expenditure. It is a fallback, not a measurement: a real reading always beats it, and it never lands in a day's row.
Where the expenditure comes from. Example data.
Cox passes no verdict on your food.
There are no good and no forbidden foods, no points, no traffic light and no comment on your character. Cox tells you how much protein is still missing and whether the carbs fit tomorrow's session. That is all.
- No judgement of a meal as good or bad
- No assessment of your eating behaviour — that is not a job for a training app
- No supplements, no dosages, no named preparations
- Blood pressure readings are displayed, not interpreted
Cox is a wellness companion for your training — not medical advice and not a medical device.
What happens to the images.
- Every image is re-encoded when stored — location and device data are stripped in the process. Otherwise a photo from the gym gives away where you train.
- Images are private and belong to your account. Delete the account and they go with it.
- To read them, the image goes to an AI model. Which one and at which provider is set out in the privacy policy (German).
- One photo per meal; the picture does not replace a label you may want to correct later.
Nutrition is a module of its own.
You can add it — or leave it and just train strength. Everything is free during the closed beta — this module included.