How the AI
sees your kitchen.
ScanMyPantry combines on-device Core ML computer vision, barcode scanning, and GPT-4o to identify food, track expiry dates, and generate recipes. Here's how each piece works — in plain English.
Point your camera
Open ScanMyPantry and point your iPhone at your fridge, pantry, or kitchen shelves. The live camera view shows real-time detection boxes as the AI identifies items. No special lighting or angles needed — it works in typical kitchen conditions.
AI identifies everything
The on-device Core ML model processes each frame, drawing bounding boxes around detected food items and classifying them into 100+ categories — produce, dairy, packaged goods, spices, beverages, and more. Confidence scores determine whether to trust on-device results or escalate to cloud AI (GPT-4o) for uncertain items. Average detection time: under 3 seconds.
Inventory builds itself
Detected items are added to your inventory automatically. Each item gets an expiry date (detected from packaging via OCR or estimated from category), a storage location (fridge/pantry/freezer), and a freshness status. Your inventory is always sorted with expiring-soonest items first.
Recipes from what you have
Tap "Generate Recipes" and GPT-4o creates personalized meals from your actual inventory. The AI knows your dietary preferences, what's about to expire, and what flavor pairings work. Missing ingredients get added to your shopping list — orderable via Instacart in one tap.
The technology stack
Core ML Vision
Custom YOLO-based object detector trained on 500,000+ labeled food images. Runs entirely on-device — private, offline-capable, sub-3-second inference. Recognizes 100+ food categories in real time.
Barcode Engine
AVFoundation barcode detection for packaged goods. Queries Open Food Facts database for product info, nutrition data, and typical shelf life. Fallback for items the visual model is uncertain about.
GPT-4o (Cloud AI)
Used for recipe generation and uncertain item identification. The AI receives your inventory list (not images) and dietary preferences. No food photos are stored or sent to servers.
iCloud Sync
SwiftData + CloudKit for household sharing. Up to 5 family members see the same real-time inventory. End-to-end encrypted. Optional — works fully offline without an account.
Technology FAQ
What technology powers ScanMyPantry?
ScanMyPantry uses a combination of on-device Core ML computer vision (trained on 500,000+ labeled food images), barcode scanning (for packaged goods), and cloud AI (GPT-4o for recipe generation and uncertain item identification). The Core ML model runs entirely on your iPhone — no internet needed for basic scanning.
How does the AI identify food?
The Core ML model is a custom YOLO-based object detector trained specifically on food items in kitchen environments. When you point your camera, the model processes each frame in real time, drawing bounding boxes around detected food items and classifying them into 100+ categories. Confidence scores determine whether to use on-device results or fall back to cloud AI.
Does scanning work without internet?
Yes. The Core ML model runs entirely on your device. You can scan your entire kitchen offline — on an airplane, in a basement, anywhere. Barcode scanning and AI recipe generation need internet connectivity, but basic food identification and inventory management work fully offline.
Is my food data private?
Yes. By default, all food scanning happens on-device. Your inventory photos and data stay on your iPhone unless you explicitly enable iCloud sync for household sharing. Cloud AI (GPT-4o) is only used when you request recipe generation or for uncertain items you choose to re-scan — and no food images are stored on servers.
How long does a scan take?
Under 3 seconds per item in live camera mode. You can scan an entire shelf at once — the AI detects and classifies multiple items simultaneously. A full fridge scan (30-40 items) typically takes under 2 minutes.
See the AI in action.
Download ScanMyPantry — it's free and takes 30 seconds to set up.
Free — no account needed. Compatible with iPhone XS and newer, iOS 17+.