Five AI eyes on your store.
Zero video leaves the building.
Nepal's first AI-powered CCTV automation platform. Pose, behavior and shoplifting detection running on your own hardware. Faces blurred before anything is uploaded — only JSON metadata reaches the cloud.
One Windows .exe. Five concurrent models.
Cammitra runs everything on the customer's own PC. The cloud only ever sees JSON. That's the entire trick — and it's why our pricing is a fraction of cloud video classifiers.
On-prem AI pipeline
YOLO + pose + behavior + ViViT shoplifting + tripwire — all running locally on consumer hardware. ONNX Runtime, not PyTorch.
Self-learning brain
Doubt clips auto-label via Gemini. Every 500 new labels triggers a YOLO fine-tune. Agent hot-reloads — no restart needed.
Privacy-first
Faces blurred at the edge before any upload. Raw video never touches the cloud. GDPR-friendly. Built for India's DPDP Act.
Smart alert rules
Natural language: "Alert me if someone loiters near the cash register for over 2 minutes after 9 PM." Sector-aware, AI-fired.
WhatsApp + email alerts
2-second face-blurred clips attached to every alert. Sent via WhatsApp Cloud API. Confirmed-theft faces auto-flag your visitor watchlist.
Multi-sector ready
Retail, traffic, hospitals, schools, banks (CBDC-ready). One codebase, sector-aware sidebars and alert rules.
Install once. The brain grows by itself.
A 4-step setup that takes about 15 minutes per store. After that, the system improves on its own as it sees more footage.
Install the agent
Download the SaneyAgent .exe on the store's PC. Paste your API key. Point it at your RTSP cameras.
Edge inference fires
5 ML models run locally at sub-50ms. JSON metadata is the only thing sent over the wire — never video.
You get alerts
Real-time dashboard updates via SSE. WhatsApp + email alerts include short, face-blurred clips.
Models retrain themselves
Doubt clips auto-label via Gemini. Every 500 new samples triggers a YOLO fine-tune. Agent hot-reloads.
Cam-Mitra vs. cloud video classifiers
Competitors like Veesion and Everseen stream raw video to their cloud and charge accordingly. We don't.
| Capability | Cam-Mitra | Cloud classifiers |
|---|---|---|
| Where inference runs | Customer's own PC | Vendor cloud |
| Video leaves premises | ✕ Never | Yes — full RTSP streams |
| ML signals concurrent | 5 (pose, behavior, shoplift, tripwire, face) | Usually 1 video classifier |
| Face-blur before upload | ✓ Always | Optional / post-hoc |
| Self-learning loop | ✓ Auto-retrain every 500 labels | Manual model updates |
| WhatsApp clip alerts | ✓ Built-in | Often email only |
| Hardware requirement | Any 4-year-old i5 + optional RTX | None on-prem (cloud cost) |
| Multi-sector platform | ✓ Retail, traffic, hospital, school | Retail only |
Built for retail. Designed for everything else.
The same edge pipeline ships across verticals. Different sector, different alert rules — same agent binary.
Retail / Mall
Smart alert rules · employee + customer analytics · pilot underway
Traffic & Roads
Red light violations · hit-and-run · wrong-way · speed monitoring
Govt. Offices
Restricted access · visitor verification · anti-corruption monitoring
Hospitals
ICU access · patient wandering · emergency crowd · restricted wards
Schools
Stranger danger · bullying · after-hours intrusion · student safety
Airports
Unattended baggage · terminal breach · crowd surge · perimeter
Disaster Control
Smoke · fire · flood · evacuation · rescue zone alerts
Banks / CBDC
CBDC kiosk security · terminal fraud · NRB pilot 2026 alignment
Put five AI models in your store this week.
14-day free trial. No credit card. One camera included. Cancel any time — your data stays on your hardware regardless.