A detection that lands on a screen nobody is watching is not a safety system. IREX routes events to the people authorized to act on them, on the device they actually carry, with the frame that triggered the alert attached.
What Triggers an Alert
IREX supports low-latency, video-based threat detection with notification through real-time crime centers, 911 dispatch centers, and secure messenger applications. Detectable events include the appearance of weapons, known suspects on a watchlist, unattended items, fire and smoke, and custom events defined by a natural-language prompt.
Alerts carry their context: the snapshot, the camera, the location on the map, the time, and the analytics that fired. An operator verifies rather than guesses, which is what keeps a false positive from becoming a dispatched unit.
Nothing responds autonomously. Detections are signals for a person to verify, and the verification, the decision, and the responder assignment are all written to the audit trail against the same Case ID.
What Feeds the Monitors
The same control-room wall as on the home page: twelve capability classes, from weapons and watchlists to fire, crowds, track intrusion and prompt-defined StreamVLM™ detectors (beta, selected instances). Every tile is a detector whose events an alarm monitor can carry. Hover to pause the wall, or click a tile for what it detects and where it runs.
Lawful search, bound to a Case ID
ALPR/LPR plate, make, model and color search
Intrusion, loitering, line-crossing
People on the track, train surfing
Density, flow and crowding alerts
Across non-overlapping cameras
Flow, violations and corridor tracking
A new detector from a single sentence
“Asian, glasses, moustache”
Early visual detection of fire and smoke
Obstruction, defocus and outage
Route reconstruction across cameras
The Alarm Monitor
An alarm monitor is a virtual space holding one sub-stream of the event flow: the alarms that one person or one team can genuinely process. In a large deployment the whole stream far exceeds what anyone can watch, so the monitor, not the camera list, is the unit of work.
The Alarms screen opens on the monitors the signed-in user’s groups are entitled to, and on nothing else. Choose one and its event cards fill the screen with the alarm details and the map location in a sidebar beside them. Unviewed cards are shaded dark red, monitors holding unread alarms sort to the top of the list with a red dot, and a bell in the header carries the count of monitors still waiting. An operator clears cards one at a time with Mark as Read, frees the screen with Hide Viewed, and hands back a monitor outside their own responsibility with Mark All as Viewed. Up to five monitors can be pinned as favorites. IREX’s own guidance on sizing them is blunt: do not put too many cameras on one monitor, split a large area into districts one monitor at a time, and break a long person list into shorter lists rather than let it flood a face-recognition monitor.
See Real-Time Crime Centers →
Creating a monitor is a short form: a self-explanatory name, the user groups that will have it, an event limit for how many alarms the timeline holds, and a response protocol, the text that reminds an operator what to do when this monitor fires and that they reopen from the card menu as View Protocol. The protocol is a reminder, not an automation: the decision stays with the person. A creator can only select groups they themselves belong to, so wider access is an administrator’s act, granted monitor by monitor on the user group’s own screen. Two switches decide where events travel beyond the screen, Enable messenger notifications and Send events to external systems. Then Set Filter decides what the monitor is for at all: analytics modules, sensors, cameras, locations, person lists, vehicle lists, priorities, objects, appearance, brand, color, and speed. Since version 4.40 a monitor carries several separate filters rather than one, each created with its own name and its own set of criteria, and Preview checks a filter before it is applied: it shows which alarms will appear on the monitor and at what rate, so the volume a filter would produce is something a supervisor looks at rather than something a shift discovers. Traffic violations are a filter category of their own, which sends captured violations to a monitor separately from the other video analytics events. Every managing action on a monitor, creating it, editing it, refiltering it, scheduling it, or deleting it, requires a Case ID.
Docs: Assign Monitors to User Groups ↗
Where Alerts Go
Alarm monitors group cameras, locations and event types onto operator workstations, so a shift sees the events it owns rather than everything at once.
Events reach the dispatch workflow already carrying the evidence, which is where the reported 10x or better response-time reduction comes from.
Switch on messenger notifications and the monitor gets a group chat of its own in the secure space, with the authorized users already in it. Each message carries the event snapshot, the event and camera names, the extras that apply such as a plate number, the location, and a link that plays the event back in the IREX player. Members comment, forward the alert, and upload evidence of their own, so the chat becomes the record of what the team did about it.
See Sover Communications →A monitor can call a URL you nominate the moment an event lands on it, the mechanics of which are below. Documented uses: a face match opening a door at an office entrance, a “Safety line crossed” event playing a step-back warning at a metro station, and motion in a protected area starting lights and sirens. Access-control and device integrations such as Sigur and Netping are managed per user group under external connections.
Continuous export to Apache Superset for heatmaps, trend analysis, and compliance auditing over the alerting itself.
The Alert Itself
Attaching the evidence to the alert is what lets verification happen on the card, before anyone is sent anywhere.
A card leads with the event class the analytics fired, over the frame that triggered it, with the camera name and the location tags underneath. Click the snapshot and the details sidebar opens with the event name, the location on the map, and the event tags that apply, such as a vehicle type or an estimated age. The three-dot menu is what an operator reaches for next: Download Snapshot for the file, Copy Link for a short link that plays the event back in the IREX player and can be pasted into a message, Camera Settings to check what the spotting camera is running, view-only unless the operator’s role allows editing, and Route to draw the person or vehicle’s waypoints on the map in chronological order.
Docs: Review an Event Card ↗
Getting Alerts Out
Sover carries an alert to a person. The other two carry it to software, and which one to use is a question of volume: a webhook for the alerts that should each trigger an action, Apache Superset for the event stream as a whole.
Switch on messenger notifications for a monitor and every alarm on it reaches a group chat in the Sover secure messenger: the event snapshot, the event and camera names, the extras that apply such as a plate, the location, and a link that plays the event back in the IREX player, on whatever device the responder is holding.
See Sover Communications →For the alerts that should each do something. Switch on Send events to external systems and the monitor calls a URL you nominate the moment an event lands on it, with the HTTP method you choose and, for POST, the frames attached; events arrive as JSON. The REST API creates the plate list, the monitor, its filter and its webhook without touching the interface, which is the API guide’s seven-step Track a Vehicle tutorial. One call per alert, so it suits a monitor that fires a few times a day, not one that fires a thousand.
Docs: Track a Vehicle through the API ↗For the stream as a whole. Every event is exported continuously into the platform’s ClickHouse event database, and Apache Superset queries it: more than 35 columns per event, from camera, module, event type and priority to plate, vehicle color and speed, recognized person and similarity, and crowd size, through filters or standard SQL. Dashboards show alarm rates per monitor, heatmaps, watchlist-hit frequency and the irregular usage patterns compliance auditing looks for, and export as PDF, image or CSV.
Docs: Apache Superset Integration ↗Keeping the Stream Workable
From the Record
Three from the archive: the alarm monitor as a product, the case for putting an alert in front of a team rather than a screen, and how permissions keep one agency’s alarms out of another agency’s view.
How to Use IREX Alarm Monitors to Get Notified about Public Safety Threats IREX Alarm Monitors provides real-time alerts about public safety threats to authorized users such as monitoring station operators, police officers, security guards, and venue managers. Multiple… Read More → 
Real-Time Alerts and Collaborative Security: Saving time saves lives When bad things happen, there is no time to waste. Time is often the deciding factor. When fire breaks out, or there is a seriously injured person lying in the street who hasn’t got long, or a child… Read More → 
Demo: IREX Permission Management for Multi-Tenant Smart City Platforms IREX is a multi-tenant platform that enables multiple organizations to use the shared city infrastructure. For example, cameras and databases can be securely shared among multiple… Read More → That is the design problem the platform takes seriously. Each detector carries its own confidence threshold and alert cooldown, alarm monitors scope events by camera and type to the roles that own them, and a weapon detector was specifically reworked to reduce false alerts. Thresholds are tuned on your own cameras during the pilot.
A monitor is shared with named user groups, and the Alarms screen shows a user only the monitors their groups hold. Whoever creates a monitor can select only the groups they themselves belong to, so widening access is an administrator’s deliberate act, performed monitor by monitor on the user group’s own screen. Cameras and the other shared resources, database lists and locations among them, are assigned to groups the same way, and roles decide what a member of a group can do rather than see. Grouping the resources themselves is a page of its own: Resource Groups, in the Collaboration section of the user menu, is where a set of video sources is created as a named group and one or more user groups are granted access to it, there or on the Resource Groups tab of the User Groups page. Once a group exists, the Cameras, Events and Plans pages and the alarm-monitor filter settings can all filter by it, so the same partition that governs access also shapes what a monitor watches. That is what makes one platform workable for several agencies sharing a city’s infrastructure.
Yes. Any alarm monitor can call a URL you nominate as soon as an event lands on it, using the HTTP method you select and, for POST, attaching the event frames; events arrive at that URL as JSON. Documented deployments use it to open a door on a face match, to play a step-back announcement when a safety line is crossed at a metro station, and to start lights and sirens on motion in a protected area. Named CAD, RMS or dispatch product integrations are not part of the documented set, so treat a specific dispatch system as an integration to scope in the pilot rather than a box already ticked.
No, and deliberately not. Detections are signals for a human to verify; no response runs autonomously.
An optional service in which alarms from your cameras are transmitted to a call center that verifies the event and responds per an agreed response protocol, calling the contacts on your call list, including first responders, in order.
Yes, through Sover, the end-to-end encrypted messenger. Each alarm monitor with messenger notifications switched on gets its own group chat, and the same thread accepts evidence uploads and natural-language follow-up queries back to the platform.
Keep Reading
A pilot proves the whole chain: detection on your cameras, routing to your operators, and the audit record that follows.