How to get alerts from CCTV footage

Most camera systems are very good at recording and very bad at telling you anything. The closest they come is motion detection, which on a busy site fires on every customer, delivery and pigeon until somebody mutes it. An alert you ignore is the same as no alert, so the theft, the fall or the person in the stockroom gets discovered the next morning, on the recording, when there's nothing left to do about it.

Why motion alerts don't work

Motion detection answers exactly one question: did some pixels change. It can't tell a concealment from a customer reaching for the top shelf, because the recorder knows that something moved and nothing else. On a quiet site that's tolerable. On a busy one everything is motion, so the alerts become noise, the noise gets muted, and the cameras go back to being expensive hindsight.

The fix is to alert on what happened rather than on movement. Svid has a vision model read the footage as it arrives, describing what's going on in every frame: people, objects, actions, hazards. Once the system knows what's in the picture, an alert can mean something specific. Not “motion on camera 4” but “someone concealed an item near the till”.

Pick what you want to be told about

Alerts come from checks, paid watchers that run over the same pipeline. Each one looks for a particular kind of event: Theft Watch flags concealment as it happens, Weapon Alert does what the name suggests, Fall / Slip Detection catches someone going down. There are 31 in the Store, each with plain-English copy on how it works and a live monthly price computed from your own usage, and you only pay for the ones you turn on. The aim is to be told about the handful of events that matter on your site, and nothing else.

1. 01

   Create an account

   Sign up at app.svid.ai. Four onboarding questions suggest a check bundle and a watch level, both editable before you pay, and the price is shown with its derivation: based on your own usage, with a $9 a month minimum.

2. 02

   Add your footage

   The Add tab takes uploads, a pasted video URL (the title fills itself in), a live phone monitor at app.svid.ai/live.html once you've signed in on the device, or an RTSP IP camera captured on our side.

3. 03

   Pick your checks

   Choose the events that actually matter where you are. A shop probably wants Theft Watch; a warehouse cares more about Fall / Slip Detection. A check runs across a whole folder of footage in one action, so the till cameras get watched for theft without the car park paying for it.

4. 04

   Choose how alerts reach you

   On screen in the Alerts tab, by email, or by JSON webhook into your POS, Slack, or any system your team already watches.

How alerts behave

One event fires one alert. If the same check fires again within two minutes, that's recorded as the event continuing, not a new siren, so a five-minute incident is one entry in the Alerts tab rather than a stack of duplicates. And because alerting checks judge a window of recent frames rather than a single image, what gets flagged is an event with a shape, someone going down and staying down, rather than one ambiguous frame. The moment behind every alert is already searchable footage too; if you want the five minutes either side of it, you search for it instead of scrubbing for it.

What it costs

The monthly price is based on your own usage for your cameras, with a $9 a month minimum, worked out from four questions at signup and shown with the full derivation before you pay: frames per month, AI analyses, and your price. Each check you add is priced in the Store from your own usage, as the extra it would add per month, so a two-camera shop and a distribution site pay very different amounts for the same check. The pricing section shows the arithmetic.

How do alerts reach me?

On screen in the Alerts tab, by email, or as a JSON webhook into a system you already run. Everything flagged stays in the Alerts tab, filterable, whichever route you choose.

Can alerts go into Slack or my POS?

Yes. Alerts can be delivered as a JSON webhook into any existing system, and POS and Slack are common destinations. It's a generic webhook rather than a native integration, which means it works with whatever you've already got.

Will one incident spam me with alerts?

No. One event fires one alert, and the same check re-firing within two minutes is recorded as a continuation of that event. A long incident reads as one alert carrying on, not a fresh siren every few seconds.

Will I get flooded with false alarms like motion detection?

Alerts are raised on described events, not on movement, so a passing pigeon doesn't qualify. Alerting checks also judge a window of recent frames rather than one image, so what gets flagged is a sequence of events, not a single odd-looking frame. And you choose which events you want to hear about in the first place.