An AI nutrition tracker for calories, macros and meals

The reason most food logs get abandoned is not the maths — it is the twelve taps it takes to record a sandwich. So there are five ways to get a meal in, and you use whichever is fastest at that moment: describe it in plain language (in English or German), photograph the plate, scan the product barcode, search the food database by name, or type the numbers in yourself.

Typing and searching lead to the same place: your own saved foods and the food databases, with personal, generic and branded results kept apart so a named product is never quietly swapped for a generic one. A barcode goes straight to the product. Those carry the source's own stated values, and the source and its preparation — raw or cooked rice are not the same row — stay visible.

A photo, or an AI estimate you ask for by name, is a good guess and nothing more — so it arrives in editable fields, with every number, the portion multiplier and the name open for you to correct before you save it. The tracker never blurs the difference, and nothing is estimated on your behalf because you typed something.

After saving, a logged item is still yours: each row shows a badge for where it came from — database, photo, estimate or entered by hand — and can be edited or deleted, with a window to undo the deletion. An estimate-heavy day therefore looks like an estimate-heavy day rather than passing itself off as measured.

Most people eat the same twenty things. So once a meal is right, name it and save it with its ingredients — and from then on it is two taps, on the web app or the Android app, whichever you have in your hand.

Reuse keeps the review step. Pick the items you want from a saved meal or from any earlier day, choose the day and slot, adjust the amounts that were different this time, and check the totals before you confirm. Confirming twice does not log it twice.

A logged meal is a snapshot of itself: editing or deleting the saved version later leaves every day you already recorded exactly as it was. And saving a meal records nothing — it counts when you log it, not before.

Calories are one number out of several that decide whether a day supported your goal. So the tracker gives you a daily calorie target and a daily protein target — plus carbohydrate and fat — computed from your body data and the goal you chose, and shows what is left of each as the day goes on.

At the end of the day you get one Coach Verdict out of 100. It weighs five things, not one: how your calories sat against your goal, your protein, the quality of the food you ate, whether you fuelled around training, and your fibre, fruit and vegetables — with an honest deduction when the day's salt ran over the daily reference. Every logged item also carries an automatically estimated NOVA 1 to 4 food-processing group, so “400 kcal of oats” and “400 kcal of biscuits” are not filed as the same day.

What comes back is coaching, not a leaderboard: up to two things that actually went well, and exactly one prioritised thing to change tomorrow. Protein that is already where it should be does not get raised again, and a genuinely good day is allowed to be a good day.

Every tracker can tell you that you have 700 calories and 60 grams of protein left. The hard part is the next sentence, and that is the one this tracker answers: it turns the remainder into two to four concrete meals, each with its macros and a one-line reason it fits — and you log one with a tap, adjust the portion first, or ignore all of them.

Ask it to plan the rest of the day and it lays out the remaining meals as a coherent set rather than three unrelated suggestions. Over a week, it can anchor a handful of meals against the shortfalls that keep recurring and hand you the shopping list for them, with the rest of the days left deliberately free.

All of it comes from curated meal templates and our own recipe library, so a suggestion is always real, cookable food — never an invented dish, and never a rigid menu. Quantities on the shopping list are merged and counted, never summed out of free text, because a list that guesses at grams is worse than one that says “rolled oats ×3”.

The same log serves all three, because what changes is the target and not the tracker. You pick the goal — fat loss, lean gain, maintenance, performance, or a deliberate mass gain — and your calorie target moves with it while your protein target is set from the band that goal needs.

If you want the reasoning behind the two goals people search for most, each has its own page: how a deficit is sized and how to read the scale when you are losing weight, and why energy decides whether you gain but training decides what the gain is made of when you are going the other way.

You can change the goal whenever your plans change, and a goal you have stated yourself is never quietly overwritten later — including by a training plan you add afterwards. A cut alongside a strength block is a coherent decision, and it stays your decision.

Nutrition-only is a complete path here, not a trial version. Enter your body basics, pick a goal, and you have targets, the day's guidance, logging and your coach's read — with no training plan, and none created for you in the background. That last part is enforced in the code rather than promised in marketing copy: nothing on the nutrition side is allowed to reach the plan generator, and a test fails the build if that ever changes.

If you do train with a plan here, the targets get more informed rather than more complicated: the energy your sessions demand is added on top, so a hard training day and a rest day stop carrying the same number, and your coach can read how you ate before the sessions you finished — reported as something observed alongside each other, never as cause and effect.

Honestly: an AI estimate from a description or a photo is an approximation of a portion it cannot weigh. It is usually close enough to steer a week, and it will sometimes be wrong by a meaningful margin — most often on portion size, on oil and dressings, and on composite home-cooked dishes. Where the number matters, scan the barcode or pick the product from the database and the value is the product's own. Where it does not, a fast approximation you actually log beats a precise one you skip.

The tracker is also honest about your log rather than your food. Only the days you actually tracked count towards a read, today is left out until it is over, and behind every read sits a “Why?” that shows what it was based on — how many days, which values, which period. When there is not enough to say something solid, it says that instead of producing a number that merely looks precise. A partly logged day is read as a partly logged day, not as a disciplined one.

And it stops where medicine starts. Micronutrients are answered one at a time, as whether your week's food covers them — never as a deficiency, and never as a lab result. Supplements are kept as a routine you planned, with generic types and their sources rather than brands. The tracker does not diagnose anything and is not a substitute for a doctor or a registered dietitian, particularly if you are pregnant, managing a condition such as diabetes or an eating disorder, taking medication affected by what you eat, or under 18.

On your data: we do not sell it. Product analytics only runs if you accept it in the cookie banner, our events are limited to category-level properties and never carry your food entries, free text or health notes, and you can download a copy of your data or delete your account and everything in it yourself from your account settings. Estimating a meal from text or a photo sends it to the large-language-model provider named in our privacy policy; a barcode lookup goes to the Open Food Facts database.