Edge AI. Selective cloud. Customer-owned data.
Gobi runs detection locally on the node, sends only meaningful events to the cloud for enrichment, and renders everything in a clean searchable dashboard. No proprietary cameras. No port forwarding. No vendor lock-in.
Architecture, layer by layer.
📷 Camera compatibility
Any IP camera with an RTSP stream. That covers most modern business and municipal cameras. ONVIF discovery is supported. Mixed brands per site are fine.
🟧 The Gobi node
A small Linux device sized to your site (Pi 5, Beelink N100, OptiPlex, PowerEdge or rugged industrial). Runs detection, hosts forensic recording, and tunnels out securely.
⚡ Frigate (local detection)
Open-source NVR with hardware-accelerated detection. Filters out wind, rain, shadows. Tags people, vehicles, pets, packages and more — locally, without sending raw video to the cloud.
🧠 AI enrichment pipeline
Only meaningful events go upstream. AWS Rekognition handles fast labels, plates, and optional face matching. Claude Vision generates rich scene descriptions and structured tags for search.
🔒 Cloudflare Tunnel
Outbound-only secure tunnel. No port forwarding, no static IPs, no firewall holes — and no public exposure of your cameras.
🛰 Tailscale
Private mesh network for fleet management. Manage every Gobi node — including ones on cellular or Starlink — from one place.
📊 Dashboard
Live grid, daily timelapse, AI Motion Log search, sensor widgets, role-based access. Clean, fast, mobile-friendly.
⚙ Node-RED & MQTT
No-code automation layer. Connect cameras, sensors, gates, relays, PLCs, Modbus devices, alerts, dashboards. Wire anything to anything.
🗄 Data ownership
Forensic video and snapshots stay on the node. Only events and metadata are pushed for search. Snapshot ZIPs are downloadable. Customers own their cameras, their feeds, and their data — including face libraries and plate lists.