Video logging and a faster AI coach
A photo works when there is one plate in front of you. A table with four dishes was always the harder case — so this release replaces the single frame with a short clip.
Three seconds instead of one frame
Record a short pan across the table and the model sees each dish from several angles. Depth cues that a single photo throws away — how high the rice sits in the bowl, how thick the slice is — come back, and the portion estimate stops leaning on averages.
Everything lands as separate entries. A bowl of pasta, a side salad and a glass of juice arrive as three items with their own weights, so correcting one does not blow up the rest of the meal.
The coach answers while you are still looking at the plate
The daily plan used to refresh on a schedule. It now recomputes as soon as an entry lands, which is the moment the advice is actually useful: you can still add a side, skip the dessert, or move dinner an hour earlier.
Under the hood the day is summarised once and reused, instead of being rebuilt for every question. In practice a suggestion comes back in about a second on a recent iPhone.
Also in this release
- Barcode scanning keeps the packet's serving size, so 30 g of cereal no longer becomes 100 g.
- Weekly reports separate weekdays from weekends — the split most people never notice on their own.
- Voice notes accept several dishes in one sentence in all nine languages.
Nothing about your history changes: past days keep their entries, and the new estimates only apply to what you log from here.
