Product.
What is Semaphor?
Semaphor is AI video monitoring assistance for security operators. It watches the same monitoring screen your guards watch, separates every camera pane, detects people and vehicles, and sends focused alerts with the original frames and an AI scene report to one guard inbox. You keep your cameras and camera software. Read the full overview in what is Semaphor.
How does Semaphor work, step by step?
It runs one loop continuously: observe the screen, react to people, vehicles and substantial scene changes, analyze the event with a vision-language model, and deliver a card to the inbox. Duplicate suppression and cooldowns keep repeat activity from flooding the queue. The walkthrough is on how Semaphor works.
What does an alert actually contain?
A preliminary alert arrives immediately with the original frames, so the guard never waits on the AI to see that something moved. A vision-language model then updates the same card with a scene report: what is happening, what it observed, a recommendation and what remains uncertain. See alert-first review and AI scene reports.
Compatibility.
Does Semaphor work with our existing camera software?
Yes. Semaphor works from the rendered monitoring screen, whether that is a VMS client, a video wall or a camera dashboard. There is no vendor API to connect and no camera credentials to share, so there is no integration project. We confirm capture quality on your layouts during a pilot. Details are on works with existing camera software.
Do we need new cameras or a new VMS?
No. If a camera appears on your monitoring screen, Semaphor can observe it, whatever the brand or age of the camera behind it. That makes it one of the simplest ways to add AI to existing security cameras without a rip-and-replace project or a new license tier on your video platform.
What happens when our layout cycles through cameras?
Semaphor gives each view a visual identity and follows it as layouts cycle, resize or rearrange. A camera cycled off the screen is marked unobserved and shown as a coverage gap. It is never assumed quiet. Read how in camera view tracking.
Guards & AI.
Does the AI decide whether to dispatch?
No. The guard stays responsible for every action. The AI supplies a scene report and a recommendation, shown beside the original evidence and its uncertainty. Semaphor never closes an alert on its own and never calls for a response. That is human-in-the-loop security AI by design, and it is how accountability stays with a named person.
Will Semaphor replace our guards?
No. Semaphor does the continuous watching that wears people down, and brings the right camera forward. Your guards do the judging, the post orders and the response. The goal is to cover more cameras without more guards, not to take guards off the post.
How does Semaphor cut down on nuisance alerts?
It alerts on detected people and vehicles, not every pixel change from rain, shadows or headlights. Event routing adds duplicate suppression and cooldowns, so the same parked car does not alert all night. The guard can override priority on any card. See duplicate alert suppression and our guide to reducing nuisance alerts.
Does Semaphor detect fire and smoke?
Fire and smoke detection is in active development, with a trained model evaluated offline on held-out video. Early access is open to pilot partners. It is designed to complement certified fire alarm systems, not replace them. Follow progress on what we are building next.
Data & privacy.
Where does our video get processed?
Person and vehicle detection runs locally on the monitoring station. The vision-language model can run locally or on hosted inference. When hosted, only selected event stills are sent, over verified HTTPS with authentication and optional mutual TLS. No raw video stream leaves the site. See security and privacy.
Does Semaphor use facial recognition?
No. Track IDs are numbered per camera clip to follow movement within a view. They do not identify who anyone is, and Semaphor does not perform facial recognition. Read more in person and vehicle detection.
What evidence does Semaphor keep?
Every processed frame and crop keeps its original image with an exact SHA-256 hash and perceptual hashes. The original evidence sits beside every AI assessment, manifests export as JSON, and retention is automatic and bounded in guard mode. Details are on evidence and audit trail.
Pilots & pricing.
How does a pilot work?
Start with one monitoring station. Together we confirm capture quality, dashboard layouts and the situations your operators need surfaced. Then we measure missed events, nuisance alerts, response delay and evidence review time, and you expand station by station where results earn it. See start a pilot and how to run an AI monitoring pilot.
How much does Semaphor cost?
Semaphor is built to be accessible to guard companies and monitoring centers that cannot justify premium enterprise platforms. Pricing depends on stations, cameras and how you want the AI review hosted, so we scope it with you during a live demo. Book a live demo and we will walk through it on your own camera screens.
Can we see it before committing to anything?
Yes. The home page has a recorded demo on public research footage with detection overlays, a track inspector and saved AI scene reports. For the real test, book a live demo on your own camera screens; it takes one conversation to set up.