About AnomaEYE
Cameras should help people respond, not just record.
AnomaEYE began as a research project in video anomaly detection. Today it is a working system: cameras stream in, a model flags incidents, a second model checks them, and the right people get the clip.
Our story
Why we built it.
Most camera networks are reviewed after the fact. Footage is searched once a loss has been reported, and the moment that mattered is buried in hours of recording.
We set out to build a model that reads behavior in video, then put it inside what a security team actually needs: one alert per incident, the clip that shows it, and a clear record of what happened next.
We would rather show you the system than describe it. This site describes what runs today, and our Technology page lists what doesn't work well yet.
Timeline
Where the project stands.
Late 2025
Research
A first video anomaly-detection model, trained and tested on public crime footage.
May 2026
Backend
A multi-tenant service for organizations, roles, cameras, incidents and alert delivery.
July 2026
Live system
Live cameras, GPU detection, the web dashboard and the Android app, with evidence clips and alerts by push, email, SMS and siren.
Today
Verified alerts
The v4 model, with a vision-language verifier that checks every candidate before an alert goes out. Available for demos and pilot conversations.
Principles
How we make decisions.
Privacy by design
Detection reads movement and behavior. Facial recognition isn't part of the workflow.
People decide
An alert prioritizes attention. A person decides what happened and what to do about it.
Honest about limits
We publish what the system does poorly and recommend a pilot before any rollout.
See the limitationsBuilt around response
One alert per incident, with the camera and clip attached, on the channels each person chose.
Team
Who's building AnomaEYE.
Talk to us
See the system working.
A 30-minute walkthrough of live detection, alerts and evidence.