How Vengeea compares to Icetana — unsupervised anomaly detection integrated with existing VMS vs predefined on-prem detection modules on any existing IP camera.
Icetana uses unsupervised machine learning to learn a per-camera baseline of "normal" and flag anomalies. Integrates with existing VMS platforms (Milestone, Genetec, Digifort). Strong at reducing operator noise on cameras where an unsupervised model works well — the trade-off is a learning window before the baseline is trustworthy, and an alert that says "unusual" rather than naming what happened.
Vengeea is an on-premises AI video analytics platform with predefined detection modules — intrusion, weapon, face, fire/smoke and LPR are live today, fight/aggression, theft and liquid-leak are in active rollout — that ship pre-trained and are enabled per-camera with no baseline-learning period. Runs standalone or alongside any VMS.
Where the two platforms diverge in practice — during procurement, at rollout, and after go-live. The core split is architectural: Icetana's unsupervised model has to learn what a camera's normal looks like before it's useful, while Vengeea's modules ship pre-trained and start naming events on day one — the trade-off shows up in time-to-value and in how much operator judgement each alert still requires.
| Dimension | Vengeea | Icetana |
|---|---|---|
| Detection approach | Predefined, pre-trained modules per event type | Unsupervised anomaly detection per camera |
| Time to first useful alert | Sub-second after enable | Baseline-learning window (days) per camera |
| What triggers an alert | Named event: intrusion, weapon, fire, LPR match, etc. | Anomaly relative to the learned scene baseline |
| Deployment model | On-prem appliance; optional cloud sync | On-prem or private-cloud, VMS-integrated |
| Camera compatibility | Any RTSP / ONVIF | Via VMS integration (Milestone, Genetec, Digifort, etc.) |
| Standalone use without a VMS | Yes — Vengeea's own dashboard | Depends on the deployment — typically paired with a VMS |
| Named-event modules | Live: intrusion, weapon, face, fire/smoke, LPR. Rolling out: fight, theft, leak | Anomaly-driven; specific events depend on configuration |
| How new detectors are added | Toggle a module or add a custom YOLO/ONNX | Anomaly model tuning per camera |
| Alert interpretability | Named event class with confidence + snapshot | "Unusual" — operator judgement on the flagged clip |
| Best-fit customer | Facilities that need specific named-event detection an operator can act on without reviewing footage first | Facilities with large camera counts and low-signal baselines where reducing noise is the priority |
Because Vengeea runs alongside during the pilot and doesn't touch existing cameras or the VMS, the whole process is reversible until you decide to commit.
Icetana's approach is unsupervised anomaly detection — it learns what "normal" looks like on each camera during a baseline window and flags deviations from it. Vengeea's approach is predefined detection modules — intrusion, weapon, face, fire/smoke and LPR live today, fight, theft and leak rolling out — that ship pre-trained and produce named-event alerts from the moment they're enabled. Both are on-prem; they solve different halves of the operator-experience problem: Icetana catches "something's off here" on cameras with no obvious event class, Vengeea tells the operator exactly what happened and where.
It depends on the camera and the event. Icetana's scene-learning cuts noise well on cameras with a clear, stable baseline — a busy loading dock that's always busy learns as "normal" fast. Vengeea's per-module confidence tuning and zone polygons cut noise on cameras where the operator cares about a specific event class regardless of how busy the scene is — a weapon in a crowded lobby still fires even though "crowded lobby" is the baseline. In parallel pilots, customers often end up using both — different tools for different cameras, which is exactly the kind of finding a two-to-four-week pilot on your own footage is meant to surface before you commit to either.
No. Vengeea has its own dashboard and can run standalone. If you have a VMS (Milestone, Genetec, Digifort, Exacq, Avigilon Unity), Vengeea reads the same RTSP streams alongside it and can forward events back as bookmarks via ONVIF or webhook.
Yes. Both read RTSP; running them in parallel is the typical pilot arrangement and often the long-term steady state.
Vengeea's per-module confidence tuning + zone polygons + rule scheduling reduces false alerts significantly, but it does so by pinning down what event you care about on which camera. If the customer wants a "tell me when the camera looks unusual" experience without pre-defining event classes, Icetana's approach fits that framing more directly.
1–3 day integration. No hardware replacement. Reversible pilot.
