Why Sunn Forgets What You Ate Last Tuesday

Sunn doesn't remember your Tuesday lunch. It remembers that you're vegetarian, that you train before work, and that deadline weeks are when dinner turns into cupboard raiding. That split is deliberate, and it's the most opinionated decision in the whole product. A coach who recalls every sandwich you've ever eaten isn't a better coach — they're a worse one, buried in detail. What you actually want remembered is the short list of things that change the advice.

Most tracking apps get this exactly backwards. They keep a perfect, permanent record of every item you ever logged, and know nothing about you. Sunn is built the other way round.

Being wrong is a feature, not a bug

Start with the part everyone worries about: an AI guessing your portions will sometimes guess wrong. Of course it will. You sent it a photo of a curry. It cannot see the ghee.

The question that actually matters isn't whether the first estimate is perfect — it's how much work it takes to fix. In most apps, correcting an entry costs more taps than creating it did: find the meal, open the item, hunt the database, re-enter the portion, save. That friction is why people quit. Not the logging. The re-logging.

On Sunn you just say what was wrong, the way you'd say it to a person:

Corrections apply at the ingredient level, not the meal level. Tell it the rice wasn't there and it removes the rice — the chicken, the sauce and the salad keep their original numbers, untouched. Tell it 250g instead of 150g and every value on that item scales by the same ratio rather than getting re-estimated from scratch. It changes what you named and nothing else, and it won't quietly invent an item you didn't mention. Then it sends back the new meal total so you can see the damage.

A tracker you can argue with in one sentence beats a more accurate one you have to fight. Perfection isn't available at dinner. Getting within twenty calories in five seconds is.

Three kinds of memory

Underneath, Sunn keeps three separate things in its head, and they behave very differently.

That third layer is the one people mean when they say an assistant "knows" them. It gets written in the background after a conversation, not during it, so nothing slows down while you're mid-sentence — and each new fact is checked against what's already on file, so "I'm vegetarian" doesn't get saved eleven times.

What it refuses to remember

Here's the part I find most interesting, because it's the opposite of how these systems usually get built. The instruction governing that third layer says, in effect: most conversations should produce nothing worth saving, and when in doubt, save nothing.

Things it's specifically told not to keep:

• What you ate in any given meal — that belongs in your food log, which is a different thing with a different job
• Daily calorie and macro totals — already recorded, and stale within a day
• What's in your fridge — true on Monday, wrong by Thursday
• Nutrition facts — it doesn't need a note reminding it that broccoli is low-carb
• Its own suggestions — a recipe it proposed is not a fact about you
• Things you were only curious about — saying you might try counting macros isn't the same as doing it, and only the doing gets recorded

To be clear about what that is and isn't: this is a design decision about usefulness, not a privacy promise. Your meals are still stored — they're your log, and the dashboard and summaries are built from them. The point is narrower and, I think, more important: the profile that shapes your coaching is kept deliberately small. Memory that grows without limit doesn't make an assistant smarter. It makes it confidently wrong about a version of you that stopped being accurate months ago.

The weekly read

One-off meals tell you almost nothing. Ten days of meals tell you plenty. So once a week Sunn reads back over your logs looking only for things that repeat: the cuisines you drift toward, the two or three proteins doing all the work, the breakfast you have most mornings, whether you grill or fry, whether dinner is at seven or ten, whether fiber is quietly missing every single week.

Single meals are explicitly ignored here, and so are exact quantities. It's looking for the shape of how you eat, not the contents of Thursday. That shape is what turns into advice worth having — and it's the difference between an app that tells you that you were 30g under on protein, and a coach that has worked out you're under on protein every weekday because breakfast is coffee, and suggests a fix that survives your actual morning.

Why any of this matters

Nearly everyone who quits tracking quits for the same two reasons: entering food is tedious, and the app never gets any easier to use the longer you stay. Month six is exactly as much work as day one. You re-tell it you don't eat pork. You re-enter the same breakfast. You re-explain your goal to something that has watched you pursue it for half a year.

Sunn is built so the work goes down over time instead of staying flat. You correct it in plain language when it's wrong, and the correction takes one sentence. It learns the handful of durable things about you and applies them without being asked again. It deliberately doesn't hoard the rest, so the picture it holds of you stays accurate and small. If you've bounced off tracking before because it never stopped feeling like data entry — or you keep ending the day hungry and can't see the pattern causing it — that's the friction this is aimed at. You can see the rest of how it works in thirty days of tracking on WhatsApp, or just start with the quiz and tell it something about yourself.