Theft Watch: shoplifting detection on your existing CCTV

What it does

Svid already reads every frame of your footage and describes what's happening in it. Theft Watch is a paid check over those descriptions that looks specifically for theft behaviour: items going into pockets and bags, tags being worked at, someone moving goods towards an exit without passing a till. Because it judges a window of recent frames, it sees the sequence rather than one ambiguous still, and when it sees something it flags the moment rather than quietly logging it.

What an alert looks like

The alert lands on screen and by email with a plain-English description of what was seen and a link to the exact moment in the footage. If you'd rather wire it into your own systems, every alert is also available as a JSON webhook, which shops typically point at Slack or the POS back office.

The alert is a flag for a human, not a verdict. Staff open the clip, see what actually happened, and decide what to do while the person is still in the shop. Anything that wasn't caught live is still searchable afterwards, so the end-of-day review is a query, not a scrub.

Where it runs

Like every check, Theft Watch is switched on per camera, so you can run it on the shop floor and the stockroom door and leave the office camera alone. It works on RTSP feeds from your existing recorder, on uploaded files, and on a live phone camera propped where coverage is thin. For sites where footage can't leave the building, there's a fully on-prem appliance mode.

What it costs

There's no rate card. The Store prices Theft Watch from your own usage before you buy: your cameras, hours a day and watch level become a live figure, shown as ‘+$X/mo at your usage’ next to the check. It's added to a monthly subscription priced from your own usage with a $9 a month floor, and the whole derivation is on screen at signup. The pricing section walks through the arithmetic.

How quickly do alerts arrive?

Frames are judged as they arrive, so the flag goes out while the behaviour is still fresh, on screen, by email or by webhook. If the check fires again within two minutes, that's recorded as the same event continuing, not a new siren.

What about false alarms?

Theft Watch is tuned to flag for human review rather than to accuse. Each alert includes the clip and a description, so a glance settles it. Staff stop watching feeds and only look when there's something to look at.

Does it work on recorded footage too?

Yes. Live and recorded video go through the same pipeline, so you can run it over last week's export to see what it would have caught before you point a live camera at it.