.webp){kind=logo}
Watch Agent is included in the Platinum tier and works alongside icetana Safety and Security. While Safety and Security learns what is unusual, Watch Agent detects the specific incidents you define.
You tell us the risk. Our team builds and tunes the detection to suit your site.
Describe the risk you actually have, for example the loading dock that floods, the west car park after 11pm.
Antara Core runs detection on your own servers with no internet connection. If your site allows outbound traffic, the same events can point at an external LLM instead.
All detections appear in one icetana Livewall, giving operators a single queue to monitor and respond to.
Same events, same Livewall, same tuning. The choice is about your network, your data rules and your budget shape. Not about detection quality.
Note: a provider can change or retire a model underneath you. When that happens the event has to be re-tuned.
On Antara Core the model moves only when we move it, and we test your events against it before the release ships.
Watch Agent runs on a schedule tailored to each event and camera, balancing coverage with compute requirements.
Which cameras carry which events. Your highest-risk views get the closest attention. A quiet back corridor does not need the same, and paying for it takes capacity away from the views that matter.
How quickly each one has to surface. Some events need to reach an operator in seconds. Hazard work is usually written against a five-minute food-court response and twenty minutes elsewhere, and we configure comfortably inside it.
Compute is finite. Every event you switch on, on every camera you add it to, competes for the same GPU as the events already running.
Turning everything up everywhere is the quickest way to slow the whole system down, including the detections you cared about most. We size the configuration against your hardware and tell you where the ceiling sits.
We separate proven detections from those that still need validation on your cameras. If your event is not listed, tell us what you need and we’ll tell you where it stands.
A specific incident you want found on camera, described in plain language and tuned until it detects reliably. "Liquid spilled on the ground" and "vehicle in a pedestrian zone" are watch events. When one is detected it appears on your Livewall with the camera, the time and the clip.
You define it; we engineer it. You describe the risk on your site in plain language, and our AI team turns that into an event, tests it against real footage and tunes it until the coverage is worth the compute it consumes.
The split is deliberate. An untested description fires on the wrong things, and every event you run costs compute you are paying for whether it finds anything or not.
How fast is a configuration. Cameras are checked on a schedule set for your site, and the events that need to reach an operator quickest are configured to do so.
For hazard and safety work that lands comfortably inside the response times sites actually contract to.
If you have something where seconds genuinely matter, tell us during sizing. We will configure for it and tell you what it costs in compute.
They are complementary, and they come together in the Platinum tier.
Safety and Security learns the rhythm of your site and flags what is unusual, the person moving against the flow, the vehicle where vehicles are not.
Watch Agent looks for a specific thing you have named, whether or not it is out of the ordinary.
You need both, because the two answer different questions: one tells you something here is off, the other tells you the thing you were worried about just happened. They run on the same cameras and land in the same queue on the same Livewall.
Yes, on Antara Core. Detection, reporting and the Livewall all run on servers inside your network, which is why the product suits dark sites, air gapped networks and anywhere with strict data sovereignty rules. Running on an external LLM does require internet access.
Yes, on Antara Core. Detection, reporting and the Livewall all run on servers inside your network, which is why it suits dark sites, air-gapped networks and sites with strict data sovereignty rules. Running on an external LLM does require outbound internet access.
It depends on your camera count, which events, and the server they run on. Rather than quote a number that will be wrong for your site, we size it with you before you buy and tell you where the ceiling sits on your hardware, including what happens to the events you already have when you add another.
On Antara Core, nothing changes until icetana ships a release, and we test your events against a new model before that release goes out.
On an external LLM, the provider controls the timing, and a model change can move detection behaviour underneath you. We will flag it and re-tune, but the exposure is real and you should price it in when you choose the engine.
Most security incidents are visible in the footage before anyone knew to look. icetana AI makes sure your team sees them in time.
.webp){kind=logo}