Every guide on this keyword sells the central station. This one is about what happens when the alarm is already verified before it leaves the building.

It means a central station contract. A panel sensor (door contact, motion PIR, glass break) trips, the panel dials the central station over cellular or IP, a human operator reads the zone label, calls the property to verify it is not a false alarm, and only then dispatches police. ADT, Emergency24, Pye-Barker, Smith Thompson, and GeoArm all describe variations of this same pipeline. The monthly fee ($15 to $50 per month) and the per-camera-pricing on the video add-on exist to pay for that human operator and the dispatch relationship.

When the AI runs on the DVR's HDMI output, the classified event IS the alarm. There is no unverified signal arriving at a central station with a zone label attached. The alarm arrives with an event_class (person_in_zone, loiter, vehicle_dwell, tamper), a tile.label pulled from the recorder's own camera-name strip, a 480x270 JPEG thumbnail of the triggering tile, and a deep link to the exact timestamp on the recorder. The verification already happened on-device. The operator's job collapses from 'scrub video to decide if this is real' to 'decide what to do about this verified event.'

On a 25-tile DVR multiview, the detector produces about 412,803 candidate detections per 24 hours. The four-stage filter pipeline (overlay_mask, tile-grid zones, multi-frame persistence, tile.label dedup) drops 99.942% of those before anything crosses the network. Only 241 events become alarms. That is a 0.058% emission rate. It is what makes the 'the alarm is the verification' claim actually hold: if you emit a fraction of a percent of candidate detections, each surviving event has already passed four independent suppression checks.

Yes. Most municipalities (Los Angeles, Dallas, Phoenix, and roughly 200 others) charge $50 to $400 per unverified police dispatch on repeat offenders, and many require Enhanced Call Verification (ECV) or Verified Response before dispatch at all. Traditional panel monitoring fails ECV by definition because the central station only has the zone label and a phone tree. A pre-verified video alarm ships with the thumbnail that shows the actual person in the actual zone at the actual timestamp, which is precisely what Verified Response ordinances (Salt Lake City, Seattle, Las Vegas precedent) require. The dispatch is either preempted by the thumbnail itself or satisfied by it.

Cyrano does not replace the DVR or NVR. It plugs into the HDMI output of whichever recorder is already there, passes the signal through to the wall monitor so the guard sees the same multiview, and captures frames from the passthrough for inference. The overlay_mask template library already covers Hikvision DS-7xxx, Dahua XVR and NVR, Lorex, Amcrest, Reolink NVR, Uniview, Swann, Night Owl, Q-See, ANNKE, EZVIZ, Bosch DIVAR, Honeywell Performance, and Panasonic WJ-NX. Recorder brand does not matter as long as it puts a multiview on HDMI-out, which all of them do.

Yes, but the 24/7 role is monitoring the event stream, not the raw video. The Cyrano unit runs inference around the clock on-device. Any event that passes all four filter stages lands in the operator inbox with its thumbnail and recorder deep link. You can route that inbox to your own staff, to a white-label alarm response vendor, or to a guard service. The crucial difference is that the human is not triaging unverified signals; they are triaging already-classified events. One person can reliably cover 20 to 50 properties this way because the per-property alert volume collapses.

Nine fields plus a thumbnail. tile.label (camera name the DVR stamps on the strip, e.g. 'Loading Dock NE'), tile.index (row-major position in the grid), tile.coords (x,y,w,h in composite-frame pixels), property (site ID), layout_id (recorder layout, e.g. '4x4-std'), overlay_mask (array of overlay classes blanked, e.g. ['clock','cam_name_strip','channel_bug']), event_class (person_in_zone / vehicle_dwell / loiter / tamper), iso8601_ts (recorder clock timestamp), latency_ms (capture-to-delivery time). The thumbnail is a 480x270 JPEG crop of just the triggering tile, stored at an S3 key the event references.

Detection keeps running. Events keep getting generated, filtered, and written to a local queue on the Cyrano device. Thumbnails are generated locally. When the link comes back up, the queue drains to the cloud in timestamp order and the operator dashboard backfills. A traditional panel calling a central station over cellular fails when the cell link fails; a cloud camera with a subscription fails when the upstream fails. The HDMI edge-tap path is offline-durable because the AI decision and the thumbnail both happen before the network ever enters the picture.

A central station plan for a small commercial property runs $15 to $50 per month, and video add-ons (Alarm.com, Verisure, cloud cameras) typically add $10 to $25 per camera per month. A 16-camera property easily hits $250+ per month with dispatch fees layered on top. The HDMI edge-tap is $450 one-time for the device and $200 per month for the property, independent of camera count up to 25. At 50 properties the monthly cost is $10,000 flat versus anywhere from $12,000 to $30,000 for the central-station-plus-video-addon stack.

Neither, in the strict sense. The panel (door contacts, motion PIRs, glass break) still does what it does and still reaches whichever central station you already pay. The HDMI edge-tap operates in parallel on the camera side, emitting pre-verified video alarms into your operator inbox. Many customers eventually drop the panel contract once they realize the video alarms are more actionable and cheaper, but nothing forces that. They run side-by-side until you are ready.

Five common routes: (1) notify the on-site guard with the thumbnail and recorder deep link; (2) route the event to a white-label monitoring vendor with video response; (3) call the tenant or property manager through an auto-dialer with the thumbnail link via SMS; (4) submit the event thumbnail as ECV evidence to local PD where Verified Response is required; (5) escalate to an armed-response dispatch service where the ordinance allows. All five are configurable per event_class and per property so LOW-threat events (delivery person at noon) and HIGH-threat events (masked person at 2am) follow different response paths.

Verkada and Eagle Eye are cloud VMS platforms that push continuous video from the camera or recorder to a vendor bucket and run analytics in their cloud. The property still pays the upstream bandwidth cost, the raw footage still lives on a vendor server, and the 'alarm' is still a post-hoc annotation on stored video. Cyrano's HDMI edge-tap runs inference on-device before anything leaves the building. The alarm is emitted at classification time, with the thumbnail generated locally, and nothing about the raw video is uploaded. The posture is fundamentally different on privacy, bandwidth, and latency.