Smart Scan

Smart Scan sends a photo to a vision model and gets back a list of what it can see. It is the difference between an inventory you build and one you photograph, and it is the reason people finish the garage instead of abandoning it.

Two modes#

Smart Scan runs in two modes. You do not choose between them from a menu. The app picks based on what you are doing.

Whole-photo is the one people mean by Smart Scan. Photograph an open box or a shelf, and it returns every distinct thing it can identify as a separate item, each with a name, a description and suggested tags. You get a checklist to review before anything is saved.

Single-subject consolidates the photo into exactly one result. This is what runs when you photograph a bin (naming the container rather than cataloguing its contents) and when you add one item and want it named. It is the mode behind Auto Smart Scan.

Auto Smart Scan#

Found under Settings → AI, and on by default. With it enabled, taking a photo while adding a bin or an item quietly suggests a name and description instead of leaving you a blank field.

It needs a connection, because the vision model runs on the server. Item photos fall back to on-device suggestions when you are offline; bin photos do not.

If you would rather name things yourself, turn it off. Nothing else changes. Manual Smart Scan from a bin remains available either way.

Reviewing before it saves#

Nothing is written until you say so. A scan produces a checklist: every row starts ticked, every name is editable, and you untick anything the model invented or duplicated.

This is not a formality. Vision models are good at obvious objects in good light and unreliable about things that are partly hidden, unusually lit, or ambiguous at a glance. The review step is where the inventory becomes accurate, and it is much faster than typing the list; you are correcting rather than composing.

Treat the output as a first draft written by someone who has never seen your house. It is right often enough to save you most of the typing and wrong often enough that you should read it.

Getting better results#

• Open the container and shoot from above. A closed box scans as a box.
• Spread things out if it is quick to. Overlapping objects get merged or missed, and this is the single biggest factor in a good result.
• Light it. Phone torch, room light, daylight. Underexposed photos produce vague names like "an object" more often than they produce wrong ones.
• Take two. A bin holds three photos: half-full and full gets you the things at the bottom, which is exactly what you will search for later.
• Do not expect model numbers. It reads shapes, not spec sheets. Anything precise (a serial, a part number) you type or scan as a barcode.

Usage limits#

Smart Scan calls a paid vision API, so it is metered. When an account is on a metered plan, a usage entry appears in Settings → AI showing where you are. Run out and scanning stops with a message saying so; everything else in the app carries on working normally.

On a self-hosted server, the meter is yours: Smart Scan uses an API key you supply, and the cost and the limit are between you and the provider.

When not to bother#

Smart Scan earns its keep on containers full of miscellaneous things. It is a waste of time on a bin holding twelve identical items, where typing the name once and setting a quantity is faster and more accurate than any model will be.