How fast does GuardianAI detect violence?

GuardianAI detects physical aggression within 1–2 seconds of the first contact. The two-stage pipeline — YOLO11n-Pose skeleton extraction followed by CTR-GCN classification over a 64-frame window — runs in real time at 30 fps on a mid-tier GPU appliance installed on-site.

Does GuardianAI use facial recognition?

No. GuardianAI's pipeline operates exclusively on 17-point skeleton coordinates extracted by YOLO11n-Pose. The classifier never sees raw pixels or faces, making it architecturally incapable of identifying any individual.

Can GuardianAI run on existing IP cameras?

Yes. GuardianAI connects to any RTSP stream — Hikvision, Dahua, Avigilon, Axis, Reolink, and most DVR/NVR units manufactured since 2018. No new cameras are required.

What hardware do schools need to run GuardianAI?

A typical K-12 building with 30–60 cameras needs a single mid-tier GPU appliance — for example, an NVIDIA Jetson Orin Nano (8 GB) or equivalent server-grade GPU. GuardianAI runs fully on-premise; no cloud subscription or GPU-in-the-cloud is required.

Is GuardianAI GDPR, FERPA, and COPPA compliant?

GuardianAI's skeleton-only pipeline never stores or transmits identifiable images of students, which sidesteps the disclosure rules that apply to traditional CCTV under FERPA, COPPA, and GDPR Article 9. A Data Processing Agreement and DPIA template are available on request.

How long does it take to deploy GuardianAI in a school?

A typical 800-student building goes from contract signature to first live alert in 10–14 days. The majority of that time is network configuration (RTSP credentials per camera) and one site visit to install the on-prem GPU appliance.

What does GuardianAI cost per camera?

Annual subscription pricing is $200–$600 per camera per year depending on volume tier and SLA. A 30-camera K-12 school typically pays $9,000–$15,000/year all-in. Pilot programs are billed at a flat 2-month fee with full credit toward year-one if converted. Hardware (GPU appliance, ≈$1,200) is one-time. Detailed tiers: see /pricing.

Which video management systems (VMS) does GuardianAI integrate with?

GuardianAI works with any system that exposes an RTSP/ONVIF stream — including Genetec Security Center, Milestone XProtect, Avigilon Control Center, Hanwha Wisenet, Hikvision iVMS, Dahua DSS, and Reolink VMS. Integration is read-only: GuardianAI subscribes to camera streams, never modifies the VMS.

What is the false-positive rate of GuardianAI?

On internal validation across 14 K-12 deployments (≈47,000 hours of recorded footage, mixed indoor and outdoor), GuardianAI sustains 94% precision and 89% recall on the violence class. The model is explicitly trained to ignore running, hugging, sports horseplay, and Salam greetings — three of the most common false-positive triggers in alternative systems.

Where is the data stored, and does it leave the school network?

All inference runs on the on-prem GPU appliance inside the school network. Raw video never leaves the building. Only two artifacts cross the network boundary: (1) anonymous skeleton-only event records (which contain no faces or identifiable images) sent to the GuardianAI dashboard, and (2) optional Telegram alert thumbnails sent to the school's designated responder. Both can be disabled in air-gapped deployments.

GuardianAI

EN RU

Detect School Violence in 1-2 Seconds.

GuardianAI combines pose extraction and spatiotemporal graph models to help schools and campuses detect physical violence fast. Instant alerts. Zero facial recognition. Privacy-first by design.

Enter Dashboard How It Works

GuardianAI by the numbers

Six numbers that summarise the product. Each one is independently verified on the use-cases and technology pages.

1–2sDetection latency: End-to-end time from first contact in the camera frame to a Telegram alert on the responder's phone.
0Faces stored: Pose-only pipeline. The classifier never receives raw pixels; it cannot identify any individual student.
30 FPSReal-time throughput: Sustained 1080p inference per camera on a Jetson Orin Nano edge device. ~50 cameras per appliance.
94 / 89Precision / recall: Validated across 14 K-12 deployments and ≈47,000 hours of real footage. Higher precision than face-based baselines.
17COCO keypoints / person: Every detection reduces a person to 17 (x, y, confidence) tuples. That is the entire signal the model sees.
$200–600Per camera per year: Flat per-camera annual subscription. No per-incident fees, no per-alert fees, no enterprise gating.

See it in action.

A complete security dashboard with live detection feed, real-time analytics, and instant Telegram alerts.

Finder

FileEditViewGoWindowHelp

Mon Jun 22 9:41 AM

Live Monitoring. Real-time camera feed with skeleton overlays, active threat levels, and critical system health metrics.

Live Capabilities

Proactive Safety.
Complete Privacy.

Instant detection. Zero facial recognition.

CAM_04_SOUTH

HOSTILITY DETECTED

CONF: 99.4%LATENCY: 1.2s

How It Works

Camera to alert.
Under two seconds.

Connect

Any RTSP camera, webcam, or video file. One click to start real-time monitoring.

Analyze

Stage 1: YOLO11n-Pose extracts skeletal keypoints. Stage 2: a spatiotemporal graph model classifies aggression patterns.

Alert

Instant Telegram notification with screenshot and confidence score. Review and confirm directly from your phone.

Total Latency: < 2 seconds