Drop objects on the belt. Show SorterAI a few sample photos. It learns your sort rules — colour, shape, custom labels — and routes them to the right bin in milliseconds. Linux on the edge, Next.js in the cloud, secure mTLS in between.
Manage one belt or a hundred from the same dashboard. Push new rules in seconds. Roll back broken updates automatically.
Run YOLO + CLIP locally on the Jetson, or fall back to Anthropic / OpenAI vision models routed through our secure proxy.
Upload a handful of reference photos. SorterAI builds CLIP embeddings and matches live frames in under 50 ms.
Compose conditions like Lego blocks: ‘red AND round → reject’. Edit rules from the touch panel or your laptop. Hot-reloaded in 2s.
Signed system bundles, A/B partitions, instant rollback if a health check fails. Canary, staged or fleet-wide.
Each belt provisions an ED25519 keypair on first boot. 90-day cert rotation, full audit log, JWT-signed commands.
Raspberry Pi 5 + optional Jetson Orin Nano + ESP32 bridge for hard real-time. Bill of materials in the docs.
Power on, scan the QR, enter the 6-char code from the touch panel.
Drop a few photos of each object class. We embed them with CLIP.
Visual editor: ‘color is red’, ‘shape is round’, route → lane.
Live cam, throughput stats, alerts. Rollback any change in one click.
Start lean with cloud-AI. Upgrade to local inference when you need offline reliability or sub-50 ms decisions.