The Hardware

CCTV / VSaaS runs on a dedicated appliance called the Edge Processor (shown in the UI as "AI Edge Processor"). It is a turnkey, pre-flashed product — like the Door & Gate Access controller and the AI Agent Processor, it arrives imaged and ready: you connect power and network, then claim it by serial number from Device Management. There is nothing to install and no software to license.

This page covers the physical device: the tiers, what's inside, how many cameras each handles, and how to install it.

Note: Hardware revisions improve over time — a newer unit may differ in appearance or exact platform. The software experience is identical across tiers and revisions. Your device's exact model is shown in its device dialog in Device Management (the header shows the accelerator model and variant).

The three tiers

The Edge Processor is built on NVIDIA's Jetson edge-AI platform and is provisioned in three capacity tiers. Which tier you receive depends on the size of your deployment.

Camera-feed guidance assumes 1080p streams. Recording without AI inference is much lighter than running person detection, which is why VSaaS-only capacity is roughly double. Higher-resolution cameras (2K/4K) consume proportionally more capacity — see Camera capacity by resolution below.

Edge Processor 32 / 64

The 32 GB and 64 GB tiers are built on the Jetson AGX Orin module in an industrial-grade carrier platform:

• Enclosure roughly 210 × 180 × 87 mm — small enough for a comms cabinet shelf.
• Five wired network ports — one 10GbE and four 1GbE. Cameras can share your existing LAN or be isolated on the appliance's spare ports.
• NVMe solid-state storage on PCIe Gen4 for local recording buffer.
• Wide-range DC power input (19–48 V) with the supplied power adapter.
• Operating temperature −10 °C to 60 °C; actively cooled.
• Full specification tables: Specifications.

Edge Processor Thor

The Thor tier is built on NVIDIA's Jetson AGX Thor (T5000) platform — at the time of writing (July 2026), the flagship of the Jetson line:

• NVIDIA Blackwell GPU with 128 GB of memory — an order of magnitude more AI compute than the Orin tiers, headroom for the largest camera counts and the heaviest AI Insights workloads.
• 1 TB NVMe storage for a deep local recording buffer.
• 5GbE RJ45 networking, plus a QSFP28 cage (4× 25GbE) for specialised deployments.
• Platform details: NVIDIA's public Jetson Thor documentation.

Camera capacity by resolution

The headline feed counts assume 1080p cameras. Capacity scales with the pixels the appliance has to process, so higher-resolution cameras reduce the count and lower-resolution cameras increase it. The table below shows the approximate number of cameras each tier handles at each resolution — the first figure is with AI inference (person detection) running, the second is record-only (VSaaS without AI).

Mixed fleets are fine — each camera simply consumes capacity in proportion to its resolution (one 4K camera ≈ four 1080p cameras). If your camera plan exceeds one appliance, a facility can run multiple Edge Processors.

These are rough guidelines, not hard limits. There is no fixed cap on the number of cameras you can add to an Edge Processor. However, going beyond the recommendations will cause the device to drop frames — which can mean less accurate AI detections, gaps in recordings, and degraded hardware-accelerated encoding for live streams. Stay within the guidance above (or add another appliance) for reliable results.

Internet upload for continuous recording

With continuous recording, the appliance uploads every camera's stream to cloud storage as-is — no transcoding or downscaling — so the internet upload you need is simply the sum of your cameras' stream bitrates. Typical H.264 camera bitrates, and the resulting upload requirement:

Reading the numbers:

• These are sustained upload rates, needed around the clock while continuous recording runs. Size your internet plan's upload speed against them, and leave ~25% headroom for live view, other site traffic, and bitrate spikes in busy scenes.
• Your cameras set the actual figure. Bitrate is configured on the camera (resolution × frame rate × codec × scene motion); check each camera's stream settings and add them up. Cameras encoding H.265/HEVC typically need about half the figures above.
• Motion-triggered recording needs far less — only clips with activity are uploaded, so most sites on motion recording use a fraction of these rates.
• Short outages don't lose footage. The appliance buffers recordings on its NVMe storage and catches the upload queue up when bandwidth frees; you can watch the queue in the device status popover (Device health & monitoring).

Which tier do I need?

Count the cameras you plan to record, then decide whether you'll run AI Insights & People on top:

Placement and installation

Treat the Edge Processor like the small server it is:

• Hard-wire it. Connect it to your network by Ethernet — it moves a lot of video. Use the 10GbE port for the uplink where your switch supports it.
• Put it near your cameras' network. The appliance pulls every camera stream continuously across your LAN; keep it on the same switch fabric as the cameras where possible.
• Give it constant power. A socket that is not on a switched circuit or timer; on a UPS if your network gear has one. Recording stops when the appliance loses power.
• Keep it ventilated. The unit is actively cooled and works hard — leave clearance around it and avoid sealed cabinets with other heat-producing equipment. You can check its running temperature any time in the device status popover (see Device health & monitoring); the UI warns that the "Device will Thermal Throttle at temperatures over 77°C".
• Place it securely. It holds a local buffer of your footage — staff-only areas, comms cabinets, or locked offices.

Network requirements

• Cameras → appliance: cameras must be reachable from the appliance on your local network (standard camera streaming ports). Cameras never need internet access.
• Appliance → cloud: ordinary outbound HTTPS and secure WebSocket connections. No inbound ports, no port-forwarding, no firewall changes are required for recording, upload, or management.
• Live view: your browser streams from the appliance peer-to-peer where possible. On the local network this is automatic. For remote viewing, the player falls back to a buffered mode automatically if a direct path can't be established.
• Bandwidth: upload consumption depends on camera count, resolution, and recording mode — see Internet upload for continuous recording for sizing guidance. The device status popover shows the current upload speed, and the Dashboard header shows your facility's measured line speed.

Identity and serial number

Each appliance is identified by its serial number, printed on the unit and its carton, and carried on a QR code. You use it once — when claiming the device (see Getting started) — and when talking to support. In Device Management, the device dialog's Details tab shows the device UUID, model, network addresses, uptime, and live processing statistics.

Related pages

• Specifications — full specification tables for the Edge Processor
• Getting started — claiming the appliance and adding cameras
• Cameras & devices — connecting and configuring cameras
• Device health & monitoring — temperature, upload queue, and status