Built for whoever comes in through the console.

There are organisations with several locations, technical people at head office, none at the branches and no control room with shifts: medium-sized companies, schools, logistics, clinics. For them video is a tool you look at when something happens, and nobody watches the sixteen-camera mosaic.

Rhombus is built for that. The cameras record inside themselves, analyse what they see in the camera itself and send up the light stuff: what it saw, when, and one image. The video is fetched when somebody asks for it. You search by what happened instead of rewinding, and you administer it from a browser like any other tool in the company.

And it adds what an organisation like that needs together: doors, temperature or humidity sensors, an alert for a door left open or for noise, and the alerts arriving where those people already work —the company chat, the email, the mobile—. That is not one more function: it is what makes somebody actually read the alert.

Where the recordings live: the video in the camera; the index, the users, the permissions and whatever is archived, in the supplier's cloud — which by design has an administration route in. What you should demand is the same as with the others: minimum permissions, compulsory two-factor, a visible access log and, in writing, who can see what from the supplier's side.

If the line goes down the camera carries on recording and what goes down is the live view from outside, the alerts and the search. When it comes back, the cloud catches up. It works for nearly everything except live monitoring.

The cost of uploading video: low and steady while nobody watches. The peak turns up when somebody watches and when archiving to the cloud — if a lot is archived and from a lot of cameras, the sustained flow stops being small. It is worked out with the list of what is going to be archived in front of you.

RGPD and international transfers: a supplier from outside the EU, with everything that drags along —a processor contract, guarantees for the transfer, the region in writing, a prior assessment where it applies, information given to the staff—. And a question that with young platforms gets asked plainly: which region you get, with your contract, and what happens if the supplier changes it.

The dependency, and the sixth question. Its own cameras are not reused with another system, so leaving means changing hardware; third-party ones stay. And the one that is not technical: it is a younger and smaller manufacturer than the big names on this list, which brings good things —they develop fast, you can get through to support— and makes you ask about support and replacements in your country, about the references in Europe and about what would happen if the company is sold.

Camera and licence by term, with the console, the updates and the support inside. The cloud archive and any functions contracted above the base go separately.

As you grow two things get forgotten and both show up in year three. The dates: if the cameras went in in batches, the renewals get scattered across the calendar. And the archive: it is contracted for a few critical cameras, it grows without anyone deciding it and it ends up being half the bill. That is why we ask for the estimated five-year cost in writing, which is the number you have to compare a perpetual licence plus its maintenance against.

It comes with a fair amount and working from day one: people and vehicles, search by what was seen instead of by the time, number plates and faces on the models and licences that include it, zones, timetables, loitering, a door left open, occupancy. It runs in the camera, so there is no analytics server to buy.

About the part that recognises faces, two things: it is subject to strict data protection rules, in many workplaces it cannot be switched on without a properly built legal basis, and switching it on «because it comes included» is the kind of decision you pay for in an inspection. It is decided with the data protection officer in the room and it is documented.

The ceiling: it is analytics of categories and of search. It knows there is a person, a car, an open door. It does not understand a scene with several things at once, nor situations belonging to an industrial process, and there is no setting that takes it there. What you can do is pull the events out over the API and let another system decide.

And the part we disclose: in the same house there is IRIS Neural, the NVMS from Infinity Neural —the other company in the group, so we recommend it with an interest of our own and we say so right here—. In many installations they live together: the VMS governs cameras and recording, IRIS brings the understanding of the scene. With the manufacturer's own cameras you have to be specific: if there are third-party cameras, IRIS takes those same streams; if everything is the manufacturer's hardware, the normal thing is to put cameras of our own at the points that have to be understood, or to check first what access to the video your licence allows.