An end-to-end autonomous security operations center that never sleeps, never misses a threat, and scales globally in seconds.
YOLOv8n runs directly on Jetson Nano/Orin edge hardware at up to 30 FPS. Every frame is analyzed for persons, vehicles, weapons, and anomalies. Detections with confidence scores are batched and sent over HTTPS to the Cloudflare Worker API. On-device NMS suppresses duplicate alerts within a 2-second window. The pipeline supports 6 simultaneous camera streams per device with automatic reconnection on network failure.
YOLOv8n · ONNX Runtime · TensorRTYAMNet classifies audio clips in real-time across 521 sound events. Critical triggers include gunshot, glass break, screaming, vehicle horn, and general anomaly. The audio pipeline uses a sliding 3-second window with 50% overlap for seamless detection. Anomalies below the trigger threshold are logged for pattern analysis — building a baseline of ambient noise per zone over 14 days.
YAMNet · Sliding Window · Pattern BaselinePatrol vehicles stream lat/lng/speed/heading at 1Hz. Geofence zones are defined as polygons with configurable restricted hours. Entry and exit events are computed on-device using the Ray Casting algorithm, with only the event sent to the cloud — not raw GPS.
Geofencing · Polygon Ray Casting · 1HzThe fusion engine combines four weighted signal channels (vision 35%, audio 25%, zone 20%, behavior 20%) into a 0.0–1.0 threat score. Compound rules like "weapon + stationary >5min + elevated audio" fire immediately with a critical alert tier.
Weighted Fusion · Compound Rules · 5 TiersServer-Sent Events push alert data to the React dashboard with sub-100ms latency from edge to browser. Automatic reconnection with exponential backoff. Event categories include: new_alert, alert_resolved, device_offline, device_online, zone_breach.
SSE · HTTP/2 · Auto-Reconnect275+ Cloudflare edge locations each run a full SQLite instance via D1. Reads are served from the nearest node with zero cross-region latency. Writes are replicated asynchronously to the primary. The schema is optimized for high-frequency INSERT with covering indexes on (device_id, timestamp).
SQLite @ Edge · 275+ Nodes · Zero LatencyBefore connecting real hardware, a Node.js simulator generates realistic multi-modal security events every 3–12 seconds across 6 virtual devices. The simulator models day/night patrol cycles, random threat injection, device state transitions (online/offline), and compound multi-signal alerts — giving a fully operational SOC demo without any physical hardware. Currently running with 14 active alerts across 6 simulated agents.
Node.js Simulator · Day/Night Cycles · 6 DevicesA full-featured React SPA connects to the SSE stream and polls dashboard REST endpoints. Features include: live alert feed with severity color coding, device status grid with heartbeat indicators, zone threat heatmap, time-series charts for alert frequency (last 24h, 7d), alert acknowledgment and resolution workflow, and a shift-change summary modal. The dashboard is fully mobile-responsive with a collapsible sidebar.
React 18 · SSE · Recharts · Tailwind CSS · PWA-readyPhysical security events are captured by edge agents and streamed to globally distributed Cloudflare Workers, persisted to D1, and broadcast in real-time.
Base URL: https://nxv-worker.cadev.workers.dev/api/
The MVP simulator runs 6 virtual devices generating realistic multi-modal security events every 3–12 seconds with full day/night patrol modeling.
8 tables optimized for high-frequency edge ingestion with covering indexes and TTL-based auto-purge policies.
Weighted multi-signal fusion with compound cross-signal rules and 5-tier escalation.
{{ rule.condition }}
{{ rule.desc }}
From MVP simulator to enterprise-grade autonomous threat response platform.
{{ phase.desc }}
Ready-to-use snippets for connecting real hardware to the NexVigil API.