Deployment and requirements

Can IREX use my existing cameras, or do I need new ones?

In most cases, yes — reusing your existing fleet is the standard deployment model. IREX deploys across the cameras a city or organization already operates, preserving your current infrastructure, and supports multi-brand cameras: any camera that streams RTSP with H.264 Baseline Profile at 720 × 480 or better can be connected, and all ONVIF Profile S cameras are compatible. Two things to plan for. First, there is no plug-and-play connection — for compatibility and security reasons, a qualified network engineer configures each camera and router (network settings, NTP time sync, current firmware, strong passwords, and stream parameters). Second, cameras that do not meet the ONVIF/RTSP baseline, and use cases with stricter optics requirements such as traffic-violation enforcement, may need to be replaced or supplemented. Traffic enforcement is the main example: it has its own published installation specifications per violation type, covering image quality in pixels per meter, illumination, mounting height, and camera angles — see Camera requirements.

What are the minimum requirements to connect a camera?

The platform-wide minimum for any camera is:

• Video streaming over RTP/RTSP, using TCP or UDP (TCP recommended)
• H.264 Baseline Profile video compression
• Resolution of at least 720 × 480 (NTSC) — individual analytics modules may require more
• A valid RTSP URL with a static IP address

All ONVIF Profile S cameras can be connected. The recommended configuration is 1920 × 1080 or 1920 × 1440 on the primary stream at variable bitrate capped at 5 Mbps (or constant bitrate at 5 Mbps where variable bitrate is unavailable or the scene is very busy), plus an optional 740 × 480 secondary stream used for video playback in layouts on video walls. Each camera also needs correct NTP time sync — accurate time and time zone are essential for video archiving. Full detail is in System requirements.

Which video and audio codecs are supported?

For video, H.264 is supported — H.264 Baseline Profile is the platform's connection baseline and the setting to use on your cameras today. H.265 (HEVC) support is in development. MJPEG is not supported. Audio is optional; when you use it, the supported audio codecs are AAC (recommended), G.711 µ-law, and PCM, and a camera needs to support at least one of them. Audio is what the AudioTrack analytics module uses to detect events such as gunshots, screaming, glass breaking, and loud noises. Size camera bitrate and network bandwidth on H.264.

What deployment models are available?

Two commercial models are offered for the IREX Ethical AI Platform:

• User-hosted private cloud — the platform runs on your own infrastructure as an IREX Private Cloud instance. This is the model for sovereign, government, and large-scale customers that require data residency and full data ownership. It is subject to a minimum order quantity of 1,000 cameras, and the IREX Private Cloud subscription is included.
• IREX-hosted SaaS — IREX hosts the platform for you. No 1,000-camera minimum applies, so this suits smaller deployments.

Separately, there are three ways to connect cameras, and they can be mixed in one deployment: direct connection (cameras stream over fiber or high-availability broadband to the private cloud, with no servers at camera sites); high-availability edge servers (redundant local AI processing for bandwidth-constrained or unreliable links); and cost-effective edge servers (a single edge server for smaller sites). Edge processing runs analytics next to the cameras and sends only events to the data center, with full video on demand — it sharply reduces bandwidth, keeps analytics and recording running through network outages and syncs events automatically afterward, and encrypts traffic over an automatic VPN. It is recommended wherever bandwidth is limited, network disruptions are likely, or site conditions are tough; the trade-off is on-premises hardware, which is where high availability and fail-over add cost. See Edge servers.

What server hardware is required, and how is a deployment sized?

The server infrastructure has three linearly scalable groups: a Kubernetes processing subcluster (master, worker, and load-balancer nodes, sized by the compute the deployment needs), a Ceph storage subcluster (hot nodes on SSD and warm nodes on HDD, sized by retention), and auxiliary servers for metrics, logs, and configuration management. The inference layer is hardware-agnostic — standard CPU nodes and NVIDIA CUDA GPU nodes are both supported, and the CPU/GPU mix is chosen per deployment. IREX does not publish a fixed cameras-per-server or GPU-count figure, because the requirement depends on your specific mix. Sizing is done by IREX engineers from four inputs you provide: total camera count with approximate bitrates, the type and approximate number of video analytics modules, the retention period for video and events, and the maximum size of the person and vehicle databases. The final bill of materials is produced during the system design phase. Software prerequisites and the data-center network design are covered in System requirements.

How long does a typical deployment take?

A citywide IREX Ethical AI deployment typically takes 6 to 12 months end to end. Delivery follows a stage-gated seven-phase lifecycle, with a formal Go/No-Go checkpoint and a written deliverable at the end of each phase: discovery and demo; analysis and requirements; pilot or proof of value; system design and commercials; procurement and mobilization; build, deploy and integrate; and go-live, hypercare and run. A pilot runs 4 to 8 weeks against real camera streams with pre-agreed, measurable success criteria, and closes with a pilot acceptance report — at citywide scale it is optional but strongly recommended. The largest single variable in the overall schedule is the procurement route: a sole-source award moves roughly two to four times faster than an open tender. Site readiness and the number of systems to integrate with are the other main drivers.