Key features

This page describes what the IREX platform provides, capability by capability. Each entry links to the chapter that documents how to configure and use it.

Carrier-grade video management and analytics

IREX is a user-hosted platform for a centralized video network powered by AI computer vision. It deploys across the cameras already in place, from one building to an entire country, and preserves the existing infrastructure investment. The architecture is carrier-grade and multi-tenant, with hierarchical role-based access control, so several agencies or departments can share one deployment without sharing each other's data. Distributed deployment with edge analytics and remote storage is supported for sites with limited or unreliable connectivity.

See Platform architecture, Manage users and user groups, and Edge servers.

Searchveillance™: instant search at petabyte scale

Searchveillance™ retrieves people, vehicles, and events from petabytes of recorded video in under a second. Proprietary descriptor clustering and indexing makes this possible across billions of "analog" objects such as faces, person figures, and vehicles. For those there is no identifier to look up: the match has to be computed. A search can start from the map, from filter panels, from a photo, or from a natural-language request.

Scalability is independent across AI modules, camera count, event volume, concurrent users, and retention period, so a growing archive does not slow retrieval.

See Investigate a case, Find a person, and Find a vehicle.

Ask IREX: natural-language investigations

Ask IREX accepts an investigation described in ordinary words, interprets the intent, cross-references the relevant watchlists, and runs the multi-step search through the platform API. It is available from the main IREX interface and from the Sover secure messenger, so an analyst at a desk and an officer in the field use the same capability. Every request runs inside the requesting operator's permissions, requires a Case ID where the underlying action requires one, and is written to the audit log.

Ask IREX is shipping in beta on selected instances. See Ask IREX for what it does and Find events with Ask IREX for how to use it.

Real-time alerts and evidence

The platform detects threats in live video with low latency and raises notifications to real-time crime centers, 911 dispatch centers, and messenger applications. Detectable events include weapons, watchlist matches, unattended items, fire and smoke, dangerous crowding, perimeter intrusion, and conditions defined by a text prompt. Alarm monitors decide which events reach which operators, with the camera snapshot, location, and timestamp attached.

See Get alerted in real-time and Alarm monitors.

StreamVLM™: detectors defined by a prompt

Alongside the high-frame-rate detectors optimized for tracking people and vehicles, IREX integrates Vision-Language Models (VLMs) that create a working detector from a plain-text description. A prompt such as "a person lying on the ground" or "standing water in the underpass" becomes an active detector without dataset collection or model training. Municipal conditions that never justified a dedicated module, including illegal dumping, graffiti, damaged fencing, and uncleared snow, become detectable.

StreamVLM™ is shipping in beta on selected instances, and requires a local GPU node for VLM inference. See StreamVLM.

Face recognition at scale

The IREX face recognition engine is trained on a proprietary dataset of 40 million images. It is tuned for CCTV conditions rather than passport photography: off-angle faces, low light, motion, and distance. It monitors watchlists across hundreds of thousands of cameras in real time.

The engine also reads facial attributes, including gender, age, ethnic group, facial hair, and glasses. That supports an investigation when no photograph of the suspect exists and only a description does. Face-based access control with liveness and anti-spoofing checks is available for sensitive entry points.

Recognition is restricted to persons pre-registered on a watchlist, and every search requires a Case ID. See FaceTrack Pro, Engineered for ethics, and Case ID and accountability.

Person re-identification

When a face is not visible, masked, or turned away, appearance-based re-identification continues the track. A specialized engine builds a profile for each person from distinctive attributes such as clothing texture, hat and hair color, and accessories including bags and backpacks. This is what allows a route to be reconstructed across cameras that never captured a usable facial image.

See Find a person.

Multi-camera traffic analytics

IREX coordinates several cameras at once to follow vehicles across all lanes as they approach an intersection, correlating vehicle movement with traffic light state and pedestrian activity. This is the difference between reading one camera at a time and analyzing an intersection as a single scene.

All vehicle types are supported, including two-wheelers. Beyond the license plate, the system reads make, model, color, and body type. Integration with external law enforcement databases allows detection of stolen, uninsured, and unregistered vehicles. Supported enforcement scenarios include red-light and stop-line violations, forbidden maneuvers, wrong-way driving, right-of-way violations, railway crossing violations, lane restrictions, illegal parking, and average speed control.

See CarTrack Pro, Manage traffic at crossroads, and Plate number recognition.

Railway and transportation safety

Dedicated safety analytics cover railways, subways, and air transportation. Detected conditions include a person falling onto the tracks, unattended luggage, dangerous crowding at a platform edge, fire and smoke, and loitering consistent with suicide risk. Incidents of this kind are rare, which is exactly why training data for them is scarce; IREX assembled its datasets with transportation agencies and by generating synthetic 3D environments.

See ObjectTrack Pro and MotionTrack Pro.

Crowd management

People counting runs in environments from a stadium queue to a large open area, estimating up to 3,000 people in a region of interest. The module raises real-time alerts when density crosses a configured threshold and supports precise post-event counting inside any user-defined part of the camera view.

See CrowdCount.

Scalable multi-layered map

The map and floor-plan environment works offline and groups hundreds of thousands of cameras with seamless zoom. User-defined layers place other devices on the same canvas, including ATMs, legacy CCTV systems, and intercoms. From the map an operator opens snapshot previews, plays video instantly, searches for objects, and follows a person or vehicle visually between cameras.

See Navigate around Smart City and Map marks and mark types.

Comprehensive activity logging

Following the principle of transparency by design, the platform logs every user action, with particular rigor for high-risk AI features such as suspect search and tracking. Each entry carries the Case ID or other lawful reason for the investigation. Audit tools let supervisors and ethics committees examine suspicious or unauthorized activity, which is what reduces the risk of the technology being misused.

See Case ID and accountability and View the Logbook.

Collaboration and secure communications

IREX provides a shared working environment for agencies, city departments, campus teams, and transport authorities, with sensitive data processed and stored according to national regulations. Access rights are granular: per camera, per recording, per uploaded media item, per analytics module. Real-time alerts reach teams through Sover, the end-to-end-encrypted enterprise messenger, where responders coordinate, upload photo and video evidence, and manage an incident from any device. The same Sover interface also accepts natural-language queries to the platform.

See Collaborate and Ask IREX.

Continuous event export and dashboards

Events export continuously to Apache Superset, where agencies and partners build their own dashboards and reports. Common uses are traffic and crowd heatmaps with average speed statistics, suspect analytics that surface high-activity places and times, and compliance auditing that flags irregular analytics usage.

See Integration with Apache Superset and Analyze event statistics.

Hardware-agnostic analytics pipelines

The platform runs on the IREX Private Cloud, which installs on bare metal, virtual machines, private cloud, or hybrid infrastructure, with no third-party proprietary code in the stack. The analytics layer supports CPU inference through the proprietary Synet framework and GPU inference through NVIDIA CUDA. The CPU and GPU mix is sized per deployment by IREX engineering against camera count, retention period, AI workload, and the servers already owned.

See System requirements and Data and technology sovereignty.