Stop finding out afterwards.

Last night your operation was told things. A resident phoned. Somebody posted in a group at two in the morning. A responder mentioned something at shift change. A camera fired an alert nobody had time to look at. By this morning, almost none of it can be found.

Almost nothing in this industry does the second job. Software watches, records, analyses and displays, and then waits for somebody to be looking at the right screen at the right moment. Usually nobody is.

In South Africa, safety is delivered by whoever is closest to it. A precinct. An estate. A watch. A security company. Each runs its own cameras, its own control room, its own book, and each sees its own patch and nothing past the boundary.

It is the resident who phones. The message in a neighbourhood group at two in the morning. The hotline. The responder who walks in at shift change and mentions something. The officer's report, the plate read, the alert from a camera that actually fired.

In most operations that human half evaporates. It lives in a controller's memory, or it scrolls past in a group nobody can search. SENTINEL ingests it as it arrives and puts it on the same record as everything else. A system can only reason about what reaches it, and what was never written down was never going to reach anyone.

Cameras are the hard part

Cameras themselves are rarely the problem. What they sit on usually is, and it degrades quietly: a wireless bridge put up on a budget some years ago, a recorder that can no longer be reached from outside, a feed that drops at exactly the wrong moment. Platforms sold on clever analytics assume the picture arrives. More often than most operations realise, it does not arrive reliably.

Port forwarding set up by whoever was cheapest is still open, still fragile and still unencrypted. An exposed recorder or camera carrying the installer's default password is not a piece of technical untidiness. It is unauthorised access to personal information. Data leaks out of it, the integrity of the system is compromised, and so is the evidentiary value of everything that comes off it, which matters most at exactly the moment that evidence is needed in a legal process.

So we start with network continuity before anything else, and we go and fix it, including in community areas where nobody else would take the job on. Devices are reached over our own VPN system, on multiple secure protocols and multiple independent paths, so losing one route does not lose the camera. Reachability is shown per device on the map as a fact rather than an assumption. An AI watchdog tests the path continuously and repairs it before anyone has to be woken. Every camera, recorder and link is tracked as an asset down to serial, model and firmware, and checked against what the device itself says it is.

It is the same principle our own control centre runs on. A controller cannot deliver a service on a camera they cannot see.

Without a licence, or hardware you were told to buy

SENTINEL is essentially “middleware”. The manufacturer's per-camera licence, charged forever for permission to view hardware you already paid for, stops being your problem. And the conventional route to capabilities like search across a whole estate arrives as a specification for cameras, recorders and servers, written by somebody who has never seen your site. It is a capital project before it is a capability. We do that layer in software, over the cameras you already have, so what your operation can do is not decided by what somebody talked you into buying.

SENTINEL is handed the operation's own data and works with the person in front of it. It follows a thread. It chases the lead you did not think to follow. It tests a hunch against the record, and it tells you plainly when the evidence is too thin to carry the conclusion.

Describe a person or a scene in ordinary words and it finds the moments, across every camera, site and hour. What used to be a morning in the footage becomes a sentence and an answer.

Ask it a question about the operation and it writes its own queries to answer it, chains as many as the question needs, reasons across the results, and builds the report while you watch. It states the sample size behind every conclusion, and it says where the data is thin.

Somebody has to be looking at it, at the moment it matters, and usually nobody is. SENTINEL is not finished when it knows something. It is finished when the person who can do something about it has it in their hand.

The precinct next door saw the same vehicle two hours earlier, and neither of you will ever know. We intend to change that, and we will say plainly that it is an intention rather than a finished thing.

Where the business heads on both sides agree, SENTINEL is built so neighbouring operators can pass intelligence at the federated level while each keeps its business entirely separate. Your clients, your cameras, your contracts and your book stay yours. What crosses is a vehicle, a pattern, a person moving between areas, under an agreement both sides sign, on databases that stay apart.

It only works between management who actually want the same outcome: less crime, safer areas, and clients who get more than they paid for. That is not a sentiment. It is the entry requirement, and it is why this can never be a feature somebody simply buys.

Every login is named and scoped to what that person may see. Terms are accepted on the record, and the version they accepted is kept. Every view, snapshot and export is logged. A unique identity watermark sits on live tiles, snapshots and exports, so a leak is attributable to whoever took it.

Shared logins are this industry's standard practice and the textbook POPIA failure. When footage leaks, nobody can say who watched it or who exported it. This is the answer to that.

The same record shows what was sent, to whom, when, and what was done about it. That is what makes passing intelligence between organisations possible at all, which is why governance is not a page at the back of this site.