Scan, Don't Type: Adding Items to AllKeep With Your Camera

I'll save you the suspense: the reason most home-inventory apps die unused is manual entry. Specifically, the third item.

By item three, you're typing "Logitech MX Master 3S Wireless Mouse" into a phone keyboard with one hand, holding the mouse in the other, second-guessing whether you should also type the serial, and wondering if life isn't too short for this. By item ten, you've quit.

We spent more time fixing this than any other part of the app. Here's the result.

The two-tap rule

Adding a new item should be two taps, plus a photo.

1. Open camera.
2. Snap.
3. Confirm or correct.

That's it. Anything more, and the app failed at its job. Anything less, and we don't have enough information to make the entry useful later.

What the scanner actually does

Three layers, in order of speed:

1. Barcode lookup (fastest, ~200ms)

If the item has a visible barcode, AllKeep recognizes it on the fly and pulls product data from open product databases. You don't press a button — point the camera, the entry appears in the preview. If it's right, tap "Add". If it's wrong (or there's no match), fall through to layer 2.

Works well for: packaged goods, electronics with retail boxes, books with ISBNs, cosmetics, supplements. Basically anything that ever lived on a store shelf.

2. On-device photo hints (fast, ~500ms)

For items without barcodes — which is most actual stuff in your home — the app uses on-device image analysis to suggest a category and tags. This runs locally on your phone. No network, no privacy tradeoff.

It's not perfect. It's a "guess" engine, not a labeling one. What it gets:

• Broad category (kitchen appliance, electronics, clothing item, tool, kid's toy) — about 80% accuracy
• Approximate brand if it's visible — works for things with visible logos, hit-and-miss otherwise
• Color and material hints — useful for searching later

What it doesn't get: serial numbers, model numbers, or any detailed specs. For that, you fall through to layer 3 or just skip it.

3. Server-side AI (optional, ~2 seconds)

If on-device hints aren't enough, you can opt-in to server-side AI categorization per workspace. This is the "this is clearly an 8-quart Lodge enameled cast iron Dutch oven, model EC8D33" level of accuracy.

It's opt-in for two reasons. First, privacy — server-side AI means the photo briefly leaves your phone. Second, it costs money — we use AI credits for it, included in the premium tier on the web.

If you turn it off, the app works fully offline, just with less clever auto-categorization.

A real walkthrough: 30 kitchen items in 10 minutes

I tested this on my own kitchen last weekend. Counter, drawers, cabinet under the sink. 30 items.

• Total time, start to finish: 10 minutes 14 seconds
• Items where the barcode lookup worked: 11 (mostly packaged cookware boxes I'd kept)
• Items where on-device hints were correct: 14 (visible brand or obvious category)
• Items I had to type a name for manually: 5 (random old pots, one ambiguous gadget)

The 5 manual entries were unavoidable — at some point you have a random kitchen tool that nothing identifies. But 5 of 30, not 30 of 30, is the whole game.

Edge cases (what doesn't work yet)

The honest list:

• Handmade goods — pottery, hand-knitted things, kid's drawings. Nothing identifies them. Use a generic tag like "handmade" and a description.
• Vintage / pre-2000 items — not in barcode databases. Photo
  • manual category is fine; AI will miss the era.
• Books in non-Latin scripts — barcode (ISBN) works. OCR for the title is hit-and-miss; we're working on it.
• Sets vs. individuals — if you photograph 4 wine glasses, the app doesn't auto-detect "this is 4 of the same thing". We're prototyping a "quantity from photo" mode; not shipped yet.

Privacy

For the on-device path: photos do not leave your phone. The analysis runs locally; we don't see them. They sync to your account when you're online, encrypted in transit and at rest.

For the server-side AI path: photos go to our backend, are processed by the AI provider we use, and are deleted from intermediate storage after categorization. Only the resulting metadata is stored long-term. You can audit and delete this in Settings → AI.

Off by default in some workspaces; check Workspace → Settings → AI categorization.

Try it

The mobile app is the place where the camera flow actually shines. Open it, point at a single shelf, tap your way through it.

Install Inventory by AllKeep on Google Play →

If you've been doing inventory by spreadsheet, the difference is hard to overstate. The web app supports manual entry for the items where you genuinely need to type something — but on mobile, the camera should be doing 90% of the work.

---

AllKeep is a software company in Tokyo building tools that actually work.