See what's in the box.
AR-assisted inventory management for field technicians. Scan toolboxes with your camera, identify tools with on-device ML, manage containers and inventory with voice commands, and track everything offline. 130+ predefined tool types across 11 categories, all stored locally with Drift/SQLite.
Know what you have. Find it fast.
Field technicians carry dozens of tools across multiple containers, trucks, and job sites. Knowing exactly what's in each toolbox—and where that toolbox is—shouldn't require spreadsheets, manual counts, or digging through every drawer.
AR Toolbox uses your phone's camera and on-device machine learning to identify tools instantly. Point at an open toolbox, and the app shows you what's there, what's missing, and where everything should be—all overlaid in augmented reality. Hands full? Use voice commands to log items, check inventory, or find where a tool is without touching your phone.
What's built.
Five core modules covering the full inventory management workflow.
Container Management
Create and organize containers—toolboxes, bags, drawers, cabinets, vehicles. Color coding, location tagging, and nested containers with parent-child relationships.
Inventory Tracking
130+ predefined tool types across 11 categories. Manual entry with a tool type picker, or scan to add. Track quantity, condition, brand, and notes per item. Search and filter across all containers.
Camera Scanner
Point your phone at a toolbox and capture the scene. The detection pipeline identifies tools in the frame, presents candidates with confidence scores, and lets you confirm, reject, or correct matches before saving to inventory.
Voice Commands
"What's in the truck box?" "Where's my multimeter?" "Add socket set to drawer 3." Natural language intent parsing for hands-free inventory operations. Query, add, find, and scan—all by voice.
Reports & History
Full scan history with per-session stats. Review past scans, see what was detected, and track inventory changes over time.
Offline-First
Drift ORM with SQLite as the primary data store. All data lives on-device. ML inference runs locally. Works in basements, crawlspaces, job sites, and dead zones—no internet required.
Capture. Detect. Review. Save.
The scanner workflow is designed for speed and accuracy. The camera captures the scene, the ML model identifies tools, and you review the results before anything hits your inventory.
- Camera capture with permission handling
- On-device object detection with bounding boxes
- Candidate matching against your tool type database
- Review flow: confirm, reject, or correct each detection
- Save confirmed items to inventory with container assignment
- Scan session recorded in history for audit
Scanner interface
Hands-free inventory.
When your hands are full of tools, you shouldn't need to type. Speak naturally and the app parses your intent into inventory actions.
- "What's in [container]" — Query container contents
- "Where is [tool]" — Find tool location
- "Add [tool] to [container]" — Add to inventory
- "Scan [container]" — Start scan mode
- "Help" — Show available commands
- Platform-native speech-to-text, works on mobile
Voice command interface
See the data where it matters.
AR overlays show tool information anchored to the real-world position of each tool. No switching between camera and list views—the data is right where the tools are.
- Real-time AR anchoring via ARKit/ARCore
- Tool labels positioned at detection points
- Status indicators (present, missing, low quantity)
- Tap overlays for detailed information
- Currently in early development (stub phase)
AR overlay view
11 tool categories. 130+ types.
Predefined tool types seeded into the local database, covering the tools field technicians actually carry.
Sockets
8mm–24mm in deep and shallow variants. Standard and impact sockets with drive size tracking.
Wrenches
Combination, adjustable, allen, and torque wrenches. Size and type tracked per item.
Screwdrivers
Phillips, flathead, Torx, and hex drivers. Tip size and length variants.
Pliers
Needle nose, channel locks, lineman's pliers, and specialty gripping tools.
Power Tools
Drills, impact drivers, reciprocating saws, and other battery or corded tools.
Measuring
Tape measures, levels, squares, calipers, and other precision measuring instruments.
Cutting
Utility knives, tin snips, hacksaws, and other cutting tools for various materials.
HVAC
Manifold gauges, vacuum pumps, leak detectors, refrigerant scales, and flaring tools.
Electrical
Multimeters, wire strippers, fish tapes, voltage testers, and circuit tracers.
Safety
Gloves, safety glasses, ear protection, and other PPE with expiration tracking.
Specialty
Refrigerant scales, flaring tools, tube cutters, and other trade-specific equipment.
Data model.
Six tables in a local Drift/SQLite database. Everything offline-first.
Containers
Toolboxes, bags, drawers, cabinets, vehicles. Color coded with location tags and parent-child nesting.
Tool Types
130+ predefined types seeded across 11 categories. The reference library for what the app can identify.
Inventory Items
Individual tools linked to containers and tool types. Quantity, condition, brand, and notes per item.
Scans
Scan session records with timestamps and container association. The audit trail for every scan.
Scan Items
Individual detections per scan. Confidence scores, user confirmation status, and matched tool type.
Expected Items
What should be in each container. Used to flag missing tools during scans and generate discrepancy reports.
Technology.
Privacy-first architecture. All processing on-device. No cloud dependency.
Flutter + Dart
Cross-platform mobile app targeting Android and iOS from a single codebase. Riverpod 2.x for reactive state management with stream providers for database reactivity.
Drift / SQLite
Offline-first local database with 6 tables, 5 DAOs, and schema versioning. Freezed models for type-safe, immutable data classes with code generation.
On-Device ML
TFLite on Android, CoreML on iOS for tool detection. No images leave the device. Currently using a mock detector while the trained model is in development.
AR Layer
ARKit (iOS) and ARCore (Android) for real-world anchoring of tool overlays. Currently in stub phase—the architecture is ready, awaiting ML model integration.
Your data stays on your device.
Privacy-first. No cloud dependency. You control what leaves your phone.
On-Device ML
All inference runs locally via TFLite or CoreML. No images ever leave your phone.
Local Storage
All inventory data stored in a local SQLite database via Drift ORM. Camera frames are processed and discarded.
Optional Sync
Cloud sync planned for team sharing and backup. You'll control what syncs and when. Not yet implemented.
Built for the field.
AR Toolbox is designed for technicians who need to track inventory across:
What's next.
The core app is ~45% complete with containers, inventory, scanner workflow, voice parsing, and reports all functional. Still in development: