How it works
The model runs where your data already lives.
The architecture is the whole argument. Inputs, the AI engine, and outputs all sit inside one network boundary. Cloud APIs, telemetry, and external processors stay outside it, blocked. There is no outward arrow on the diagram because there is no outward flow in the system.
Everything inside the dashed perimeter runs on your hardware. The crossed-out items on the right — cloud APIs, telemetry, external processors — are what a cloud AI tool depends on and what an air-gapped deployment does without.
Deployment options
Hardware sized to the work, not the other way round.
GPU clusters
For high throughput and concurrent users — large document volumes, busy knowledge bases, and fine-tuning. Sized to the workload and housed in your data centre.
Server racks
A single rack-mounted server suits most pilots and many production deployments. It installs into existing infrastructure and runs without external connectivity.
Ruggedised edge devices
For constrained or mobile environments, a compact device runs transcription and inference on site, with the same closed boundary as a full rack.
The process
From first assessment to independent operation.
Discovery
We assess the data, the compliance requirements, and the existing infrastructure. The output is a clear scope: what runs, where, and against which obligations.
Design
We architect the solution to the specific environment — hardware, capability, and the security boundary — so the deployment fits your estate rather than forcing changes to it.
Deployment
We install and configure on site. The system goes into your network with the boundary closed; no data and no telemetry leave during installation or after it.
Training
Where the engagement includes fine-tuning, models are trained on your data, on your hardware, fully offline. The trained model is yours to keep.
Handover
Your team is trained to operate the system independently. You are not left dependent on us to run day-to-day work.
Support
Ongoing support comes from the South West team, on site where the deployment requires it rather than over a remote connection.
For the CISO reading this
What the architecture rules out, by construction.
- No data leaves the network boundary, at any stage.
- No outbound telemetry, analytics, or usage reporting.
- No third-party API calls during processing or training.
- No dependency on vendor uptime to keep working.
- No retention of your data by Electric Azimuth.
- No internet connection required to run.
Want the architecture reviewed by your team?
We will walk your IT and security people through the deployment, the data flows, and the boundary, and answer the hard questions.