Room photos

Room is a capture mode for filling Kept quickly: take up to five photos of a room, a cupboard or a shelf, and AI lists every object it sees. Kept merges the same object seen in two photos, leaves out what is already in Kept, and puts the rest in the Inbox as drafts already in that room, each with its own picture cut from the photo.

Before you start

Room mode reads photos with AI, so it needs what other AI capture needs:

• AI capture on in the location (an owner or admin turns it on, in Location settings → What to track), and an AI provider that reads photos (AI providers).
• You can edit things in the location: members, admins and owners. Viewers don’t see the mode.

When Room can’t run, the mode says why instead of hiding: AI capture is off in the location, no provider that reads photos is set up, or the AI budget is used up for now.

Take the photos

1. Open Capture, then Room on the mode strip. From a place’s page, Fill from photos opens it on that place; on the location map, a room’s card has Photograph this room.
2. Check where the photos go: the place chip at the top. A room capture goes into a room, a spot or a box, never the Unplaced area.
3. Take 1 to 5 photos, or pick them with Gallery. Stand in the doorway and take one photo per wall. Each photo waits in the tray, where you can remove it.
4. Send. Offline, the photos wait on the phone and go when it is back online, like any capture.

Kept labels in the photos are read on the phone before they are sent. A thing whose label shows is known for certain, wherever it is in the location.

Review the room

The Inbox shows one Room card at the head of the capture’s batch, with the drafts it made under it:

• New things: each a draft in the room, with its own picture, name, brand, model, colour and type filled in where the photo showed them. Accept, edit or discard them as any draft, or Accept names for the whole batch. A draft that looks like a thing you already have is flagged as a likely duplicate.
• Furniture that holds things (shelves, bookcases, cabinets, wardrobes, chests of drawers): Add as a place makes each a place inside the room, so things can go in it. Other furniture (a sofa, a bed, a table) is a draft like any thing.
• Already in Kept: things the photos showed that the room already has, by the same label, the same brand and model, or the same name. Tick them and Mark seen. A thing whose label was read starts ticked.
• Each photo says what it gave. “Couldn’t read photo 3” means that photo failed and the others still count; “This photo had more; take a closer one” means it showed more than one photo can list (60 objects).

People, pets and the room itself (walls, floors, windows, doors, fixtures) are never listed, and no picture is ever cut of a person.

Done closes the card. Discard drops the photos. When no photo could be read, Retry reads them again.

On the map

When the room is drawn on the location map, each accepted thing is put on the room’s floor with a shape guessed from its type and name, in a free spot. The next time you Arrange, the map lists what Kept placed, so you can move each where it really stands. Kept placing things never interrupts someone arranging the map at the same time.

What it costs

Each photo is one AI read, and a capture of more than one photo adds one short text-only call that merges what the photos have in common. Before you send, the tray shows an upper bound, such as “About 3 photos · up to ~31,000 tokens · Groq”: the photo, the instructions and the longest answer Kept allows, for each photo; a shorter answer costs less. A model that thinks before it answers bills its thinking on top.

Room reading counts against the same budgets and caps as other photo reading. When a cap is reached, the card says “Waiting for the AI budget”, and the photos are read when AI resumes. Each call shows in AI usage as Room photo or Room photos merged.

Privacy

• What the provider receives is the same as for any capture: a copy of each photo with location and camera data removed, and the fixed instructions.
• The full room photos are deleted when you close the room card (Done or Discard). Each draft keeps only its own cut-out picture.
• While the room card is open, its photos are private to the people who review it: owners, admins and members of the location. Viewers don’t see them on the room’s or the box’s page, in search or on their phone, and neither does a read-only API token, a read-only MCP connection or a share link. While it is under review, a room photo is never a box’s cover photo.
• To keep the full photos instead, an owner or admin turns on Keep room photos after review in Location settings → General. After Done they stay on the room’s page as ordinary photos, visible to everyone who sees the location.

In Try Kept

Try Kept has no camera and runs no AI: Room mode offers Use a sample room photo, an illustrated room, and every room photo gives the same sample results, so you can see the Room card, the drafts and the furniture without a provider. The tray says “No AI runs in this demo” where a real install shows the cost.