From an existing camera feed to a record that holds up.
One path, four stages. None of it requires a control room or someone watching a screen.
The cameras already on site
All Clear reads the standard security feeds you already run. We point the system at the cameras covering the zones where PPE matters most: usually gates, laydown yards and active work fronts.
What does get added is one small computer. It sits on your property, typically in the trailer, and it is the only new equipment involved. No new poles, no new cabling, and no change to your existing surveillance setup.
Inference on a local device
A computer-vision model runs on that device and evaluates frames as they arrive. It looks for a narrow, specific thing: a person in frame without a hard hat, without a hi-vis vest, or without a mask.
Frames are evaluated and dropped. What continues down the pipeline is a description of an event, not the footage of it. It is not a productivity or behaviour-monitoring tool. It watches for a small set of safety conditions and nothing else.
A structured compliance entry
Each detection is written to the database as a row with fixed fields, so a period can be queried and exported rather than reconstructed. This is the part that gets shown to a COR auditor, an insurer, or a WCB reviewer.
Tamper-evidence is being built into this record through our Labs4 research placement this fall: each event is hash-chained to the one before it, so a record that is altered breaks the chain and the change is detectable. It is in development, and we will say so until it has been validated.
- Timestamp
- Local time, to the second
- Site and zone
- Named site and the zone it covers
- Camera
- Source camera identifier
- Event type
- Which PPE item was absent
- Confidence
- How certain the model was
- Alert
- Who was notified, and when
- Imagery
- None, unless snapshot mode is on
A text message to the person who can act
The supervisor responsible for that area gets an SMS naming the site and the zone. There is no dashboard to keep open and no app to install, because the people who need this are outside, on radio, wearing gloves.
Prevention is a real benefit and it comes along for free. But the record is what gets tested: by an auditor, by an insurer, or by an investigation months later.
Default mode keeps no imagery at all.
Frames are analyzed on site and discarded. Only the structured event is stored. No stills, no clips, no cloud storage of imagery.
Camera → local inference → event record → SMS
If your incident process needs a still attached to an event, snapshot mode can be enabled per site. It is off until you turn it on, and it is the only path where an image is retained.
Camera → local inference → stored snapshot + event record → SMS
Runs on the cameras you have
Standard site security feeds. Deployment is a connection, not a construction project. No re-cabling, and no change to how the yard is monitored.
Inference stays on site
Detection runs locally on a device on your property. In default mode video never leaves the site, and the records are held on Canadian infrastructure.
An alert to the person who can act
The supervisor responsible for the area gets a text naming the site and zone. No dashboard to keep open, no app to install.
Audit-ready by default
Every detection is a record. When a COR auditor or a WCB reviewer asks about a shift in March, there is something to open.
What deployment actually looks like.
A call about your cameras
Which sites, which zones, what the existing feeds look like, and who should receive alerts.
The risk assessment
A fixed-fee, time-boxed engagement that produces a documented compliance baseline, and tells us what the ongoing agreement should cost for your operation.
One site connected
We install the edge device against the feeds for the first zone and confirm alerts land with the right supervisor.
The log starts filling
Real records accumulate from your own site, not a demo dataset, and the compliance baseline starts building itself.