Platform and products

What is the IREX Ethical AI Platform?

The IREX Ethical AI Platform is a video intelligence platform for public safety, built and licensed by IrexAI Inc. It runs on the cameras an organization already has — from a single site to an entire country — and turns that network into searchable video intelligence: AI analytics detect threats on live streams, Searchveillance™ retrieves people, vehicles, and events from the archive, and investigation agents answer questions asked in plain language. Accountability is enforced by the architecture rather than by policy alone: recognition is restricted to pre-registered persons and vehicles, permissions gate every screen, and every privacy-sensitive action is bound to a Case ID and written to an audit log that cannot be erased. A single instance is proven in production at 100,000+ cameras, all running real-time AI analytics with 30-day instant search, and the architecture scales linearly to 1,000,000 cameras, 100,000 users, and a one-year retention period without redesign. IREX is deployed in production in 8 countries — the United States, the United Kingdom, Ireland, Peru, Israel, Bosnia and Herzegovina, Mongolia, and Kazakhstan — managing 250,000+ cameras globally. See Key features for a capability-by-capability description.

What video analytics does the platform provide?

The platform runs 10+ AI video analytics suites, each covering one operator-facing capability and grouping several shipping modules:

• Face and facial features — watchlist suspect recognition, plus attribute search (gender, age, ethnic group, facial hair, glasses) when no suspect photograph exists. Module: FaceTrack Pro.
• License plate and vehicle features — plate recognition plus make, model, color, type, and body style; stolen, uninsured, and unregistered vehicle detection through external-database integration. Modules: CarTrack Pro, plate number recognition, CarCount.
• Perimeter security — intrusion detection, restricted-zone monitoring, abandoned-object and loitering detection. Modules: MotionTrack Pro, ObjectTrack Pro.
• Railway and transportation safety — people on tracks, train surfing, platform-edge crowding, unattended luggage.
• Crowd analytics — real-time dangerous-crowding alerts and precise post-event counting, up to 3,000 people in a region of interest. Module: CrowdCount.
• Person re-identification — appearance-based tracking by clothing texture, accessories, and hat or hair color when the face is occluded, masked, or off-angle.
• Multi-camera traffic analytics — red-light, stop-line, wrong-way, forbidden-maneuver, illegal-parking, right-of-way, railway-crossing, and average-speed enforcement; two-wheeler detection; congestion analytics; bike-lane monitoring. Module: BikeLaneTrack.
• Universal VLM analytics — custom detectors written as plain-text prompts. This is StreamVLM™, shipping in beta on selected instances.
• Access-control biometrics — face-based access control with liveness and anti-spoofing checks at sensitive entry points.
• Fire and smoke alarm — early fire and smoke detection indoors and outdoors. Module: FireTrack.
• Camera tamper and quality control — obstruction, defocus, repositioning, and quality degradation, with maintenance tickets raised automatically. Module: TamperTrack.
• Audio analytics — AudioTrack detects gunshots, screaming, glass breaking, and loud noises on camera channels that carry a microphone.

Firearms are covered by the GunTrack module, which detects weapons carried or pulled out in real time in public spaces such as schools, transport hubs, banks, and critical infrastructure. The catalog grows with each release, and which modules can run together on one camera channel is documented per module. Full detail is in Video analytics.

What is StreamVLM, and how is it different from conventional video analytics?

StreamVLM™ creates a working video detector from a sentence: an operator describes a condition in plain text — for example, "a person lying on the ground," "standing water or flooding in the underpass," or "fresh graffiti on a wall" — and that description becomes an active detector on the selected cameras, with no dataset collection, no labeling, and no model training. A Vision-Language Model reads image and text together and scores sampled frames, and a per-prompt confidence threshold and alert cooldown decide what qualifies as an event. The difference from conventional analytics is where new capability comes from: a trained module needs a dataset, a training cycle, and a release, which is worth doing for faces, plates, weapons, and fire, but not for the long tail of conditions specific to one municipality — so the two complement each other, with trained modules working every frame to follow people and vehicles continuously and StreamVLM recognizing a described scene from sampled frames. A single camera channel supports up to 100 prompt-defined detectors at once, prompts can be adjusted without taking the system offline, and every prompt, edit, and resulting event is written to the audit log. Inference runs on a GPU node inside the customer's network, so frames and results stay there; StreamVLM is shipping in beta on selected instances, is included in the standard IREX price, and adds no cost beyond that local GPU node where a deployment requires one. See StreamVLM.

What is Ask IREX, and how does natural-language search work?

Ask IREX brings natural-language queries and investigation agents to the platform: instead of translating an intention into filters — pick the cameras, set the time window, choose the object class, set the attributes, then repeat when the first attempt returns too much or too little — an operator states the investigation the way an investigator would state it. The agent interprets the intent, cross-references the relevant watchlists, runs the multi-step search through the platform API, and returns results in context with the filters already applied for further refinement, so an occasional user becomes productive without learning filter semantics and an investigation that would take a sequence of manual searches runs as one request. It is available from the main IREX interface and from the Sover secure messenger, where a field officer can send a message or a photo from any device and receive a short summary plus a link that opens the same results in the workspace; both surfaces reach the same agents and the same API. Searches run inside the same permission boundaries, Case ID requirement, and non-erasable audit logging that govern every IREX search, and underneath them the Searchveillance™ search engine retrieves people, vehicles, and events from petabyte-scale archives; IREX partner MetroCCTV reports results in under a second, a 100x+ reduction in average investigation time, and 1,000+ cameras covered per control-room operator. Ask IREX is shipping in beta on selected instances and is included in the standard IREX price. See Ask IREX.

What are Sover and the IREX Private Cloud, and how do they relate to the IREX platform?

Sover is a secure, user-hosted unified communications platform for sovereign governments and regulated organizations that need full control over how their data is processed and stored: end-to-end encrypted messaging, voice and video calls, group chats, video sharing, workspaces, bots, and LLM integrations, deployable on-premises or in a private cloud and available on Windows, macOS, Linux, Android, iOS, and web browsers. Sover is also the secure front end to the IREX Ethical AI Platform: real-time IREX alerts arrive in dedicated security spaces, and officers can query the IREX archive in plain language from a Sover chat through Ask IREX, under the same Case ID and audit-logging controls. IrexAI Inc. offers Sover alongside the IREX platform and to existing IREX customers, so that integration and platform support come from one place.

The IREX Private Cloud is the platform underneath both: IrexAI's bare-metal-to-Kubernetes platform, built entirely on open-source components with no third-party proprietary code, which installs on bare metal, virtual machines, private or public cloud, or hybrid infrastructure. It hosts the IREX video pipeline — ingest and archive, analytics on CPU and GPU, the event bus, and the search index — and it also hosts Sover. See Platform architecture.

What is the current platform version, and where can I find what's new?

The current version of the IREX Ethical AI Platform is 4.39, released in August 2026. It deepens Logbook search and export, adds frame-by-frame navigation and reverse archive playback, introduces Minivans recognition in the CarTrack Pro module, broadens the Apache Superset statistics, and streamlines camera settings and Alarm Monitor schedules. Every release has its own What's New page describing the new features and improvements, with screenshots and links into the user guide; the version list in the left navigation goes back to 4.13. See What's new in IREX 4.39.