What an AI health coach is, and what Biokode does with your data

What it sounds like when the data cross

A morning message and a question of yours, with the kind of data Kodi actually handles:

Kodi

(In the morning, opening the app.)

“Morning 🐼 Last night 6 h 10 min, HRV a bit under your average. Over your last few months, the nights after a late dinner come out about 8% shorter: you have strength at 18:00 today, so dinner at 20:30 works. Target: 2,150 kcal and 140 g of protein.”

Is my afternoon coffee hurting my sleep?

“With what I have, I cannot say: you have 4 afternoons with coffee logged and I need a few more to compare them against the rest of your nights. If you tap it when you have one, in two or three weeks I will tell you with numbers instead of an opinion.”

Counter, chatbot or coach: not the same thing

All three live on a lot of phones and do different jobs. It is worth knowing which one you need before paying for any of them.

• A calorie counter adds up what you log and compares it with a target. Useful and honest, but it does not know you slept five hours or trained twice as much this week.
• A general-purpose chatbot answers one-off nutrition questions very well. It does not have your log, does not know what you weigh today and only speaks when you speak to it.
• An AI health coach works on your data, has memory and takes the initiative: it proposes, remembers context and adjusts the plan when your numbers call for it. That is the category Biokode is in.

Which data goes into Biokode

A coach is only as good as what it sees. Biokode brings four sources together and stores them with the same rule: the user’s local day, values with their provenance, and no estimate dressed up as a measurement.

• Food: every entry becomes calories, macros and 27 micronutrients compared with EFSA dietary reference values for your sex and age. Portions estimated from a photo are flagged as estimated and can be corrected.
• Sleep and recovery: through Apple Health or Health Connect come sleep duration, stage breakdown, resting heart rate and overnight HRV. Biokode does not measure them; it reads them from your watch or ring.
• Training and movement: sessions, the time they end, and steps. The time matters: a workout finishing late counts as late training in the sleep analysis.
• How you feel: mood and energy for the day on a 1 to 5 scale, plus life events you note yourself (alcohol, sauna, meditation, a trip, a very late meal). One tap, no forms.
• Optional clinical context: you can upload a blood test as a PDF and Biokode extracts up to 41 markers so Kodi keeps them in mind as context. It does not interpret them the way a doctor would.

How it finds patterns without making them up

This is the part we care most about, because it is where cheating is easiest. Biokode looks at a rolling window of several months and, for each factor (late dinner, alcohol, late training, sauna, meditation, an event you noted), compares the nights following days with that factor against the nights without it: duration, share of deep and REM sleep, resting heart rate and HRV.

A factor does not even show up until there is a minimum number of nights with it and of baseline nights. The strength of each association comes from a statistical test per metric: the difference is measured against how much that metric swings from night to night in your case, not against a generic table. An HRV that bounces a lot every night does not hand out labels on a handful of data points; a stable resting heart rate earns them with less.

Two more brakes: an effect floor (a difference that is too small is not labelled however consistent it is) and a sample ceiling (a strong label needs more nights, and a very strong one more still). And days when you logged nothing do not count as “no late dinner”: unknown is not the same as absent.

The same engine looks further: what happens to your intake the day after a short night, how your averages shift by weekday, what the weeks you train more have in common, and how what you eat relates to your weight one or two weeks later. Anything short of its data minimum is shown as “observing, N of X days”: a pattern in the making stays visible instead of vanishing.

A calorie plan that learns from your scale

The plan starts with the standard: basal expenditure by Mifflin-St Jeor, an activity factor, a pace you choose (gentle, moderate or intensive, which becomes a percentage below your expenditure to lose fat or above it to build muscle, with prudent caps) and protein in a grams-per-kilo band taken from the ISSN.

From there, every week the engine looks at your recent weeks: your weight trend (not your last weigh-in), with its margin of uncertainty, and your average intake. That yields an observation of your real expenditure, which is combined with the current estimate giving more weight to the more reliable data: few weigh-ins or a lot of scatter, small step; dense weigh-ins, convergence in a few weeks.

No automatic adjustment moves more than a bounded step from your current target, protein stays in its band instead of scaling with calories, and the change is proposed by Kodi in its own words for you to confirm. If data is missing (too few weigh-ins, or too many days without food logged), the engine says so and waits instead of adjusting blind.

Kodi: a coach that opens and closes your day

Kodi writes on its own twice a day: in the morning, with your plan, your streak and what is worth keeping in mind; and at night, with the balance and what is up tomorrow. In between it answers when you write, with your data in front of it. No pings at random hours.

It remembers what you tell it (your schedule, your usual dishes, the knee that cannot take impact) and every night it goes over the day so it has something useful to say in the morning. You pick the tone, and it adapts to how you write.

If you want no mascot and no messages, data-only mode switches Kodi off with a toggle and leaves the engine intact: Bio Score, micronutrients against EFSA, patterns and adaptive plan, with the sober interface of a metrics dashboard.

Bio Score: the whole picture in one number, rules in plain sight

The Bio Score is a daily score of what you actually logged: food, water, sleep and movement. It does not claim to measure your health; it summarises your day with explicit criteria so that, when it goes up or down, you know exactly why. It is the number Kodi uses to talk about the whole without throwing four charts at you.

Who an AI health coach is for

• For anyone already wearing a watch or ring who suspects that what they eat, drink or train affects their rest, but has never been able to check it against their own data.
• For anyone who wants to lose fat or build muscle with a plan that adjusts to their real expenditure, not a formula’s.
• For anyone who logs food often and wants logging to take seconds and to be worth more than a number.
• Not for anyone looking for a diagnosis, a treatment or a substitute for a consultation. Biokode works with healthy adults on habits, food and general training, and refers you to a professional when something falls outside that.

Where your data lives

Health data is encrypted in transit and at rest and hosted in the European Union under GDPR. AI processing happens in the United States under standard contractual clauses. There are no ads and no data sales, and you can delete your account and everything in it whenever you want. The methodology page details the sources (EFSA, WHO, ISSN, ACSM, Mifflin-St Jeor, USDA, CIQUAL) and the measured results.

Frequently asked questions

Keep going