The Market Scout
A week of dinners, decided in about a minute, and turned into one shopping list, ordered the way the shop is actually laid out.
The constraint isn't cooking. It's time.
Deciding what to eat five nights a week is a small tax that gets paid over and over. The recipe apps that exist solve discovery: they show you a thousand dishes you'll never make. The actual bottleneck is narrower: cooking for two, on a weeknight, with a fixed amount of attention to spend.
So Ichikawa doesn't rank recipes by how good they look. It ranks them by how much of your evening they take, and it splits that into hands-on minutes and waiting minutes, because those are not the same cost. Forty minutes with thirty of them in the oven is a different evening than twenty-five minutes standing at the hob.
Library, week, list, route
A library that's actually yours
141 recipes, curated from dishes already cooked and liked, not a stock catalogue. Search by keyword, filter by "under N minutes", remove what stopped earning its place.
Add by pasting a link
Drop in any recipe URL and it reads the page's structured recipe data: title, servings, ingredients, steps, image. Phone-friendly, because that's where links get found.
Surprise me
One tap fills Monday to Sunday from the library, with one dish deliberately carried over to the next day. Leftovers are a feature, not an accident.
One shopping list, done properly
Quantities scale to the number of eaters, then merge across the whole week. Unit-aware, so 200 g and 300 g of the same thing becomes one line, not two.
Walk the shop once
The list regroups itself by aisle and draws a walking route through the store, so you cross the floor once instead of doubling back for the yoghurt.
Photo inbox
Snap a page from a cookbook and it's stored for later conversion. Deliberately not a live AI call; the transcription happens offline, at no per-photo cost.
Every step knows its own minutes
The active/passive split only works if the data underneath it exists. Every active recipe in the library was enriched step by step (each step tagged with its minutes and whether it's hands-on or waiting), plus a per-recipe hint on what to start in parallel while something else simmers.
That enrichment pass was itself run by a fleet of agents rather than by hand, which is the point of keeping a crew: the tedious, wide, mechanical job is exactly the one to hand off.
The week, photographed off the phone
Five screens from the app while it is running, at phone size: the week, the library, the list sorted into the order the shop is actually walked, that same route drawn on the plan of the shop floor, and tonight’s dish with its minutes on it.
Kawaii bento
Ichikawa is the one member of the crew that isn't dark and ominous. It's a domestic tool used at the kitchen counter and in a supermarket aisle, and it's dressed for that: soft sakura pink, rounded type, a panda-with-ramen mark, generous tap targets.
What's built, what isn't
Shipped: the library, search and time filter, add-from-URL, the weekly randomiser, the unit-aware shopping list, aisle grouping and the walking route, the mobile recipe editor, per-step timing across the whole active library, and the photo inbox.
- Store floorplan geometry is schematic, not a faithful survey of the building.
- Price comparison across shops, promotions and a pantry model are deferred: researched, not built.
- Photo-to-recipe conversion is a manual trigger, by choice: a live vision call per photo would cost money on every use, and the offline route costs nothing.