Edge Compute

Edge compute is the pattern of running inference close to the camera, on hardware we ship to your site, instead of streaming video to a cloud region. For Australian operators, it is rarely a technical preference — it is the only way to get sub-20ms response times without paying for video egress.

The numbers

< 20 ms

Typical inference latency from frame to event on a Canopy Unit. Cloud round-trips from a regional site are often 200–300ms.

0 Mbps

Video egress across the WAN. We transmit small JSON events, not frames.

100%

Of inference continues to work when the WAN drops. Events buffer locally and sync on recovery.

Why latency matters

When a queue agent needs to message a duty manager before a guest walks away from the desk, a 500ms delay is not a technicality — it means the trigger fires after the human has already moved. When a PPE agent needs to stop someone at a gate before they enter a compound, a second of latency is the difference between an alert and a near-miss report..

What happens when the network drops

Australian sites — particularly regional, industrial and construction sites — do not have the network reliability of a Sydney CBD office. Our Edge Unit is designed to keep working through outages: frames process locally, events queue on encrypted disk, and when the WAN returns, they sync in order. We have customers who have run for 21 days with no external connectivity and zero lost data.

What we run on the edge

Person, vehicle, PPE and object detection models — quantised for the hardware, signed, versioned.
A local rule engine — most triggers are thresholds over zones and windows, evaluated on the box.
A health monitor that reports temperature, FPS, latency and storage every thirty seconds.
An encrypted event queue, durable across power cycles.

Cloud services

Aggregation, dashboards, user management, and long-term analytics. These do not need sub-second response times and benefit from being centralised across a multi-site estate.

Talk to us about your network

Tell us about the sites you want to deploy at, and we will tell you whether edge is the right shape.

Start a conversation

Frequently asked questions

Why not just run everything in the cloud?

For Australian operators, cloud video processing is expensive in egress, slow in latency, and increasingly difficult to justify under APP 8. Edge inference gives sub-20ms response, zero video egress, and works when the network drops.

What is the typical latency?

Under 20ms from frame to event on a Canopy Unit. Cloud round-trips from a regional site are often 200–300ms.

Can we run a hybrid architecture?

Yes. Most Australian deployments we see are hybrid: perception and real-time action on the edge, aggregation and long-term analytics in our Sydney-hosted control plane.