Straight answers

Frequently asked questions

These are the questions people actually send us, answered the way we would answer them on a call — including the couple of places where the honest answer is “not yet.”

Your existing cameras

The question we get most often, and the one most quotes quietly dodge.

Do we have to replace our cameras to get AI?
No. The AI runs on an appliance in your rack, on a GPU — not in the camera. Any camera that can hand over a standard video stream becomes an AI camera the day it is connected, and it stays one, because the intelligence sits on hardware you upgrade in one slot instead of replacing every camera on the wall.
Which cameras work with Omnis?
Any IP camera that meets four conditions: it exposes ONVIF Profile S or a plain RTSP URL, the stream is H.264 or H.265, it is 2 MP (1080p) or better at 15 fps or better on the main stream, and it is reachable on the network. That covers essentially every business-grade IP camera made in the last decade. Cloud-locked subscription cameras that will not give up a local stream do not work, and analog cameras work only through an existing encoder or recorder that already publishes RTSP.
Is the AI better in the camera, or in a server?
In a server, for two reasons. An AI chip inside a camera is fixed for life at 1–4 TOPS, while a central GPU delivers 800+ TOPS shared across every camera — so each camera effectively gets more compute than a dedicated chip would give it, plus the larger models a camera chip cannot run at all. And anything that needs two cameras at once, like following one person across a site or searching every camera for a red pickup, cannot be done in-camera, because each camera only knows what it can see.
What happens when better AI models come out?
New capability arrives as a software update on hardware you already own. When a new GPU generation ships you change one card, not two hundred cameras — which is the whole reason the intelligence sits in the rack rather than in the camera.

What it actually does

Less a list of detections, more a set of jobs you give a camera.

What does Omnis actually detect?
Every event arrives with a plain-English summary, so a hundred events read in two minutes. You can draw a box on any spot — a register, a fire exit, a parking space — and see everyone who was there; describe what you are looking for in words across every camera and every recording; or click once for a person's entire visit, entry to exit. It also reads plates, counts queues and staffed registers, measures dwell time by display and aisle, spots spills and near-falls, runs an opening-readiness check against your own checklist, and sends a morning briefing — including the overnight report that says nothing happened and proves it.
Can it tell us something that has nothing to do with security?
Yes, and that is the point. The same cameras that watch a back door after hours can count how long the line at the counter was and which hour was worst, and put that in front of a manager every morning. One of those is loss prevention. The other is sales.
How fast can we find footage of a specific incident?
Minutes, not an afternoon. Because every detection is indexed as it happens, finding the person or vehicle that was in one spot during one window is a query rather than an investigation — and a case file with a timeline, snapshots, video links and a written narrative is built in minutes and shared by link, with no USB drives.

Multiple locations

What is true today, stated precisely, because you will test it.

Can we run several locations from one place?
Yes. One appliance per location, and one identical operator interface at every site — same screen, same search, same alerts. Your team works any site from a phone or a browser, with no VPN to stand up and nothing port-forwarded anywhere. Alerts from every location arrive in one place, routed by site and by role, and an incident raised at one location can be pushed to the others in a single tap.
How many cameras does each location support?
One appliance handles 20 to 100 cameras at full AI workload. There is no minimum — a twelve-camera site is fine. Past 100 cameras at a single site you add a second appliance, and nothing already installed gets replaced.

Security and your data

Usually the section that matters most to whoever has to answer for it.

Where does our video go? Is this cloud-based?
Your recordings stay on an appliance on your own property. There is no vendor cloud holding your footage and no per-camera cloud subscription. Remote viewing works because the appliance reaches out through an encrypted tunnel it starts itself, not because your video was uploaded somewhere.
Does our footage get used to train your AI?
No. Every frame is processed on the GPU in your building, no video or snapshots or derived data reach any third-party AI service, and your footage never trains anything of ours. That is a property of the architecture — the models are on the appliance — rather than a policy promise.
What are the cybersecurity risks of putting cameras on our network?
With most systems the risks are real: an internet-facing login page to attack, and camera firmware that phones home. Omnis is built so neither is available. The appliance accepts no inbound connections at all — no port forwarding, no DMZ, no login page published to the internet — so there is nothing at the address to scan or brute-force. The cameras sit on an isolated network with no route out, so they cannot phone home or be conscripted into a botnet. Access is per-person and revocable in a click, with an audit trail of who viewed and exported what. We do not claim it is unhackable; no honest vendor can say that. What we claim is narrower and more useful: the doors attackers actually use are not locked here, they are absent.
Do you need ports opened, or a VPN?
Neither. The appliance dials out; nothing dials in. That also means no firewall changes to request and no VPN to maintain for remote access or support.
Who can see our footage, and can we prove it?
Only the people you name, and yes. Access is per-person rather than a shared recorder password, you revoke someone in a single click, and an audit trail records who viewed and who exported what. Our own remote support is consent-based over the same outbound tunnel, with no standing access and every session logged.

Living with it

Storage, failures, and how you know it is still working.

How much storage do we need?
Budget roughly 1 TB per camera per 30 days at 4 MP, and treat retention as a dial rather than a fixed decision. Drives go in as a protected set, so a single drive failing costs you the drive and nothing else — no lost recordings.
What happens if the internet goes down?
Recording continues, and alerts queue and deliver when the connection returns. The AI does not depend on a connection because it is not in the cloud — it is running on the appliance in the building.
How would we know if it stopped recording?
The appliance tells you. It runs more than thirty-five continuous self-checks on connections, recording, storage and performance. A camera that drops at 3 AM is restarted automatically, and when something cannot be fixed you get a clear alert with next steps instead of silence — which is the real answer to discovering months later that half the cameras were not recording.
How long does installation take, and do we need IT involved?
The appliance arrives pre-built and pre-configured, so the install is power and network. There are no firewall changes to make and no VPN to stand up, and cameras are added through a wizard in the operator interface in minutes rather than from a command line. Labour and timeline for a specific site are worth a walkthrough first.

Something we haven’t answered?

Pricing depends almost entirely on how many cameras you have, so rather than publish a number that would be wrong for you, we would rather spend thirty minutes understanding the site. Same for anything specific to your building, your existing recorder, or the system you already have quoted.

Ask us directly