What it is.
The guard inbox is where Semaphor meets the guard. It is the Event stage of the loop: a single, persistent queue of alert cards drawn from every camera on the monitoring screen. Each card is a camera incident, carrying the original frames, the trigger reason, a priority, the AI scene report and a full history of what has happened to it. The inbox count shows in the browser tab, so a guard working in another window still sees when something needs attention.
Anatomy of an alert card.
- Camera and time. Which view, when the activity was captured and how the camera was identified.
- Trigger. Why it surfaced: a person, a vehicle or a substantial scene change.
- Original evidence. The frames behind the alert, hashed at capture. See evidence and audit trail.
- Priority. Critical, High, Unassessed or Low, marked as an AI suggestion or a guard decision. See alert prioritization.
- AI scene report. Summary, observations, recommendation and uncertainty, labelled with the frames it examined.
- Status. New, acknowledged or closed, with any review pending or unavailable shown plainly.
Why it matters.
One place to look.
Guards should not have to watch a video wall, a VMS event list and a separate analytics console at once. Every camera, from every system on the screen, reports into one queue with one format.
The guard stays in control.
The AI can suggest and raise priority. It cannot close a card, lower a guard’s decision or dispatch anyone. A quiet camera, a missing camera or a model failure never closes an incident on its own. Only the guard closes the card. That is the human-in-the-loop principle, built into the workflow rather than written in a policy.
Nothing falls between shifts.
Unresolved cards persist — across a browser refresh, a restart and a shift change. The incoming guard sees exactly what is open, what was acknowledged but not closed, and what the AI said about each one.
How guards work the queue.
Acknowledge.
Mark the alert as seen. The card stays open, and if genuinely newer evidence raises its priority, it comes back to the guard’s attention.
Review.
Open the original frames, read the scene report, pull the camera up in the VMS, follow post orders. Adjust priority with a reason if needed.
Close after review.
Record an explicit outcome. Closing gives repeated review-level activity on that camera a short cooldown; new urgent activity still breaks through.
A sorted queue, not a feed.
Unresolved cards are shown Critical first, then High, Unassessed and Low, oldest first within each level. Closed cards drop to the bottom. The most urgent unresolved situation is always at the top.
One card per situation.
Repeated activity on a camera updates its open card instead of stacking new ones, thanks to duplicate alert suppression. A person pacing the lobby for ten minutes is one card with a growing story, not twenty alerts.
Protected against stale updates.
Cards are versioned. If an urgent update arrives while the guard is acknowledging, the acknowledgement does not silently swallow it. An AI result about frames captured before the guard’s decision cannot overwrite that decision.
Built for shift handoff.
Handoff is where monitoring operations lose information: a verbal “keep an eye on camera 12” that never makes it to the next guard. With Semaphor, the open cards are the handoff. Every acknowledged-but-open incident, every priority override and its reason, the latest AI note and the retained frames are on the card. Read our guide to security shift handoff for a checklist that uses it.
Where it matters.
- Building lobbies — a single desk guard fielding visitors while the inbox watches the rest of the building.
- Warehouses and yards — overnight gate and dock activity tracked across a long shift.
- Houses of worship — volunteer and part-time security teams who need a clear, simple queue.
It is the daily tool for security guard companies and remote monitoring centers alike.
Frequently asked questions.
What is the difference between acknowledging and closing an alert?
Acknowledge means “I’ve seen it” — the card stays open. Close after review records an explicit outcome and resolves the card. They are separate on purpose.
Can an alert close itself?
No. Quiet cameras, missing cameras and AI failures never close an incident. Only a guard closes a card.
What happens to open alerts if the system restarts?
Inbox state is saved persistently. Open cards, acknowledgements, priorities and AI notes are all there after a restart.
Work a real queue from your own cameras: book a live demo or start a pilot.