The World’s Most Powerful Video Analytics Platform

No other vendor ships this much video intelligence on one platform: the world’s first Vision-Language Model analytics for public safety, the broadest set of specialized high-performance detection modules, and a search engine that answers questions across petabytes of video. All of it runs on the cameras you already own, under one contract and one SLA.

Book a Demo StreamVLM™Ask IREX & Video Search
100,000+
Cameras Running Analytics on One Instance
<1 s
Video Search across Petabytes of Footage
1,000+
Cameras per Operator in the Control Room
250,000+
Cameras Deployed Globally

Why IREX

Three Things Nobody Else Ships Together

IREX calls itself the world’s most powerful video analytics platform, and the claim rests on three legs rather than an adjective. The first is StreamVLM™: the world’s first Vision-Language Model analytics for public safety. It turns a sentence into a working detector with no training and no dataset collection, so the number of things the platform can detect is no longer bounded by a catalog. It is shipping in beta on selected instances.

The second is the broadest set of specialized, high-performance analytics modules on a single platform. Each module is a product in its own right with several detectors inside it. Ethical facial recognition alone covers watchlist matching, attribute search across gender, age band, ethnic appearance, facial hair, glasses, masks and headwear, and liveness and anti-spoofing checks. Plates and vehicle attributes, multi-camera traffic enforcement, two-wheelers, weapons, gunshots and other sounds, perimeter, rail and transit, crowds, fire, and camera integrity each get the same depth.

The third is a petabyte-scale video search engine. Every detection is indexed as it happens, so an investigator asks a question of the whole archive instead of scrubbing footage. Partners report answers in under a second across petabytes and a 100x or better reduction in investigation time. All three ship together under one contract, one SLA, and one upgrade path.

The Three Legs

Prompt Detectors, Petabyte Search, Cameras That Work Together

StreamVLM™: Write a Sentence, Get a Detector

Every city has a list of conditions no vendor sells a module for: a person lying on the ground, livestock wandering onto a highway, flooding in an underpass, a damaged fence, illegal dumping, graffiti, uncleared snow. StreamVLM™ evaluates live frames against a plain-English prompt and raises an alert like any other event, with multiple prompt-defined detectors per camera, each with its own confidence threshold and cooldown. It is included in the standard IREX price and shipping in beta on selected instances; the only added cost is a GPU node where a deployment needs one.

Explore StreamVLM™
Two horses step off the gravel shoulder onto a two-lane highway outside Ulaanbaatar in late-afternoon light, one of them already standing inside the white edge line.

Search Petabytes of Video in under a Second

Analytics that only raise alarms leave the archive dark. IREX indexes what every module detects: proprietary descriptor clustering scales to billions of face, figure, and vehicle objects, and the architecture scales linearly to 1,000,000 cameras and a one-year retention period without re-architecture. Search from the map, from filters on faces, vehicles, and plates, or in plain language through Ask IREX, whose agents interpret the intent, check the relevant watchlists, and run the multi-step search across the archive (beta, selected instances). Partners report results in under a second across petabytes, a 100x or better reduction in investigation time, and 1,000+ cameras held by a single operator.

See Ask IREX & Video Search
A dark-blue Kia sedan rolls toward a US border-crossing inspection canopy at night, its rear Mexican plate legible above the bumper.

Multi-Camera Object Tracking and Traffic Analytics

Cameras that work together see what single cameras cannot. Multi-camera object tracking follows a confirmed person or vehicle camera to camera on the map and on indoor floor plans, forwards and backwards in time, with every step bound to the same Case ID. Multi-camera traffic analytics coordinates the cameras on an intersection approach, correlates vehicle movement with the signal state and with pedestrians, and assembles court-ready evidence for red-light, stop-line, wrong-way, illegal-parking, right-of-way, and average-speed violations. In January 2025 it shipped as the world’s first cloud-based multi-camera traffic management system, and one city deployment detected 2.3 million violations in a single year.

See Multi-Camera Object Tracking
A dark rapid-transit diagram of New York with three numbered detections of one subject at Times Square, Union Square and Fulton Street, joined by a cyan path running south, and a card at the side carrying a portrait and a case reference.

The Modules

What the Platform Detects

Every module ships with the platform, and modules combine freely on the same camera. This grid is the complete list, so nothing in the section is reachable only from a dropdown; multi-camera tracking and StreamVLM™ also carry their own sections above, and ethical facial recognition runs the missing-persons watchlists IREX maintains with child-protection partners.

How Modules Are Used

Two Ways to Use Every Module

Every module on this page works in both modes at once. The same detection that raises an alert in the control room is indexed for the investigator who comes looking a week later.

  1. Real-Time Alerts

    Every module raises events as they happen: a watchlist match, a weapon, a person on the tracks, a plate of interest, a StreamVLM™ condition. Alerts reach the people you nominate through real-time crime centers, 911 and dispatch workflows, and the Sover secure messenger, with the triggering frame, the camera, and the map location attached, for a person to verify.

    See Real-Time Alerts & Evidence

  2. Ask IREX & Video Search

    The same detections are indexed the moment they happen, so every module also becomes a way to search the archive: find every appearance of a face, a vehicle, or a plate, filter on attributes, or describe what you are looking for in plain language and let Ask IREX run the multi-step investigation (beta, selected instances). Events also export continuously to dashboards for heatmaps, trends, and compliance auditing.

    See Ask IREX & Video Search

Hardware

Standard Servers, Your Existing Cameras

Commercial Off-the-Shelf Servers

Every module runs on standard, commercially available servers on the IREX Private Cloud: bare metal, VMs, private or hybrid cloud. No proprietary appliance.

CPU for Most Modules

The proprietary Synet inference framework runs most modules in real time on standard CPUs, so a deployment can start on the servers you already own.

GPU Where It Counts

StreamVLM™ requires an NVIDIA GPU node for VLM inference. Weapon detection runs on CPU with limited performance, so a GPU is strongly recommended for it. IREX engineering sizes the mix per deployment.

No Specialized Cameras

Works with any camera that supports ONVIF or streams RTSP with H.264 or H.265. Resolution and placement are set per module, and traffic enforcement has its own installation specification per violation type.

See Camera Requirements

Ethical AI

Accountable by Design

Every module on this page runs inside the IREX Ethical AI framework: six pillars of Transparency by Design and a mandatory Case ID on every high-risk action.

  • Recognition modules identify only pre-registered people and vehicles: wanted suspects, missing people, vehicles flagged in a case. There is no recognition or tracking of random individuals in public spaces.
  • Every privacy-sensitive search runs under a Case ID that records the lawful grounds, and every action is written to an append-only, tamper-evident audit log readable by supervisors and oversight bodies.
  • Detections are signals for a person to verify. Nothing responds autonomously.
  • Accuracy is measured on your own cameras during the pilot, against success criteria agreed in writing, and the measured numbers go into the contract. IREX publishes no per-module accuracy figure.
See How Ethical AI Works at IREX

FAQ

Why do you call it the world’s most powerful video analytics platform?

Because of three things that ship together nowhere else: the world’s first Vision-Language Model analytics for public safety (StreamVLM™, shipping in beta on selected instances), the broadest set of specialized high-performance analytics modules on one platform, and a petabyte-scale search engine over everything those modules detect. The proof points behind each leg are on this page, and every one of them is measured on your own cameras during a pilot.

Are the modules sold separately?

No. Every analytics module ships with the platform under one contract, one SLA, and one upgrade path, with no per-feature fee. StreamVLM™ and Ask IREX are included in the standard price as well, in beta on selected instances; the only additional cost is a GPU node where a deployment requires one.

Can several modules run on the same camera?

Yes, and since version 4.38 it is the documented way the platform runs rather than a special case: several modules and detectors work on one camera at the same time, with no extra hardware and no duplicated video streams. Face recognition, object tracking, and traffic monitoring can run on the same channel, which widens coverage without adding cameras and keeps GPU and bandwidth use efficient. The number of active AI modules is also one of the dimensions the platform scales linearly across. When an operator switches an extra module on, the platform lists the modules already active on that camera and asks for confirmation before it starts: the procedure is on the module compatibility page.

Do we need GPUs?

Most modules run on CPU through the proprietary Synet inference framework. StreamVLM™ requires a GPU node, and a GPU is strongly recommended for weapon detection, which runs on CPU with limited performance. IREX engineering sizes the CPU and GPU mix per deployment against camera count, module mix, retention, and the servers you already own.

How accurate is the AI?

The face recognition engine is trained on a proprietary 40-million-image dataset optimized for real-world CCTV, and was ranked #1 in accuracy among US companies and top 10 globally in the 2021 NIST Face Recognition Vendor Test on the datasets closest to street-camera conditions. IREX publishes no other per-module figure: in every deployment, accuracy is benchmarked on your own cameras during a pilot and the measured numbers go into the contract.

What camera quality do the modules need?

It varies by module and by violation type. Any camera that supports ONVIF or streams RTSP with H.264 or H.265 can be connected, but facial recognition, weapon detection, and StreamVLM™ need higher-resolution imagery of the scene than a bare connection, and traffic enforcement has its own installation specification. Per-module mount, angle, and resolution requirements are published in the camera requirements, and a site survey confirms them before deployment.

See Every Module on Your Own Cameras

A pilot runs the modules you care about on your footage, with success criteria agreed in writing before it starts. Bring a case you already solved and time it.

Book a Demo How to Buy