Back to Products
AI-Powered On-Premise Offline

Vision Search

Offline Visual Search for the Factory Floor

Your warehouse carries hundreds of carton SKUs. When workers find an unlabeled box, identifying it used to mean guessing or hunting through binders. Now they snap a photo on their phone and the system instantly returns the top-16 matching cartons — material number, dimensions, and similarity score — all without internet access.

Vision Search returning the top matching cartons for a photo of an unlabeled box
~700
Carton SKUs supported
Top-16
Results per photo query
100%
Offline / LAN operation
Zero retraining
Just reindex on inventory change
Core Features

Built for the Factory Floor

Six capabilities that go from unlabeled box to printed SKU label in under a minute — no AI expertise required.

Photo-Based AI Recognition

Point your phone camera at any unlabeled carton and tap to capture. The CLIP vision model compares your photo against the full image library and returns the top-16 most similar cartons, ranked by cosine similarity score.

Size Filtering

Enter the carton's measured length, width, and height plus a tolerance range before running the AI search. The system narrows candidates to boxes that physically match first — dramatically improving accuracy for similar-looking SKUs.

Label Printing

Click any result card and the system generates a print-ready A4 SKU label with material number, dimensions, and a barcode. Send it straight to the nearest printer — no copy-pasting or manual entry required.

Phone Camera + QR Access

A QR code posted near the workstation lets any worker open the app instantly on their personal phone. The interface automatically activates the rear camera — no app store downloads, no accounts, no training needed.

Offline / LAN Deployment

Runs entirely on a Windows PC in the workshop — no internet required, no cloud subscription, no data leaving your facility. Workers on the same LAN reach it over Wi-Fi. The CLIP model (~581 MB) runs locally at full speed.

Zero-Training Maintenance

When your carton inventory changes, just add or remove photos and trigger a one-click reindex. No model retraining, no data science involvement, no downtime. The system adapts in minutes and is ready immediately.

How It Works

From Unknown Box to Printed Label

Four steps, under a minute. No expertise or training required from the worker.

01

Take a Photo

Scan the QR code at your workstation to open the app on your phone. Point the rear camera at the unlabeled carton and tap to capture. Optionally enter measured dimensions to enable size filtering.

02

System Matches

The CLIP model encodes your photo into a feature vector and performs cosine similarity search across the full carton image library — all processed locally on the workshop PC in seconds.

03

Review Results

The app displays the top-16 matching cartons with thumbnail images, material numbers, dimensions, and similarity scores. Scroll to confirm the correct SKU visually.

04

Print the Label

Tap the correct result card and send an A4 SKU label directly to the nearest printer. The carton is labeled and back in the workflow — no manual lookups, no transcription errors.

See Vision Search Working on Your Cartons

We'll set up a live demo using your actual SKU photos — see top-16 results on your own unlabeled boxes before you commit to anything.

Request a Demo