Skip to content
Sovereignty Hub
← Back

Architecture

How the system is built, and why nothing leaves it.

The full picture for the people doing due diligence: how your data moves through the system, how the AI works on your own documents, and how the whole thing is sealed inside infrastructure you control in the UK or EEA. No black boxes — every claim here is something your technical team can verify.

Inside the system

How the system actually works

Most of this site explains why the system is private and how it's secured. This part is for the people who want to know how it works underneath: where data goes when someone uses it, how the AI answers questions about your own documents, and why none of it ever leaves your environment. No marketing here, just the mechanics.

Getting in — once

  1. Encrypted Tunnel
  2. Single Sign-On
  3. AI Workspace

Every question — once you're in

  1. Query
  2. RAG Retrieval
  3. LLM
  4. Response

Where your data goes

Getting in

Reaching the system runs over an encrypted tunnel with single sign-on — the same kind of secure access your organisation already uses for its other systems. You sign in once. From that point you're working inside your own environment, and nothing about getting in repeats with every question.

What happens with each question

Once you're in, it feels like any chat interface — but each question takes a specific path. When the question is about your own documents, it doesn't go to the model blind. The system first runs retrieval: it searches your indexed documents and pulls out the passages most relevant to the question. This is the "RAG" you'll have heard about — retrieval-augmented generation. Only those relevant passages are handed to the model, which reads them and answers, and the answer returns to the person who asked. Retrieve, then answer — every time, all inside your environment.

Why you can trust the answers

Because the model answers from the passages retrieved out of your own documents — not from whatever it picked up during training — the answers stay grounded in your material. It points back to the source, so anyone can check where an answer came from rather than taking it on faith. And because it's reading your documents at the moment you ask, not relying on having been trained on them, it works on material it has never seen before the instant you add it.

The model itself stays exactly as delivered: it does not learn from your data, it does not change between questions, and it does not drift as some outside vendor updates it.

Why none of it leaves

You'll have noticed the diagram has no way out, and that's deliberate. The system is built so there's no outbound path for your data to take in the first place — not a policy that says "don't send data out", but an environment with no door for it to leave through. The model runs inside your environment, the documents stay inside it, and the only thing that travels is the answer, back to the person who asked.

How that's enforced, and how we keep our own hands off it, is what the rest of this page covers.

The Sovereign Glass Box

Every byte of client data resides on bare-metal servers located in UK/EEA data centres (London/Helsinki). The infrastructure uses a self-hosted sovereign access layer with no outbound network path. Data that never traverses an egress boundary cannot be intercepted, subpoenaed, or exfiltrated.

This is not a contractual promise. It is a physical constraint. Our integration and maintenance services are operated by a Scottish limited company. The underlying bare-metal infrastructure is provided by Tier-1 UK/EEA partners, ensuring data residency remains strictly within approved jurisdictions. Zero-Egress Routing ensures that data egress is not merely prohibited by policy, but rendered architecturally impossible.

Full transparency. Full auditability. Your IT team inspects every component. You see everything; nothing leaves.

Retained Services

Zero Standing Privileges & Retained Support

Your IT team retains absolute administrative control at all times. We hold zero persistent credentials to your infrastructure.

Zero Standing Privileges — No persistent access to client systems
Just-In-Time Gateway Access — Temporary port access, granted and revoked by your IT Lead
Audited Terminal Sessions — Full keystroke logging and session capture

Just-In-Time Access Model

We operate a Zero Standing Privileges support model: no engineer holds any persistent credentials to your systems. When scheduled work is needed — updates, security patching — access requires two actions by your Nominated IT Lead, taken together: opening the designated gateway port, and installing our temporary access key on the AI server. Neither is sufficient on its own. We then carry out the work through audited terminal sessions with full keystroke capture. When it's done, the key and any temporary account are destroyed and your team closes the port. Every access window is logged, time-bounded, and independently verifiable.

01 Access Granted Your IT Lead performs two separate actions, both required: opens the designated gateway port, and installs our temporary access key on the AI server. Neither alone grants access — it takes both, and both are in your team's hands.
02 Maintenance We carry out the scheduled work through an audited terminal session — full keystroke capture, scoped to the agreed operations.
03 Access Revoked The temporary key and any ephemeral account are destroyed, and your IT Lead closes the gateway port. The access window is logged, time-bounded, and independently verifiable.

Review the Full Deployment Architecture

We'll walk your technical and compliance people through the zero-egress deployment model, jurisdictional isolation controls, and the Zero Standing Privileges support framework.

Talk to our team Trust Centre →