Book a call ↗

Public sector & regulated teams

Sovereign AI

AI on your own hardware, on-premises or fully air-gapped, for teams that cannot send data to a hosted model.

Book a 45‑minute discovery call
← All industries

From challenge to value

Nothing leaves
the building.

Client work first, then applications we build for teams like yours.

01
From a client engagement

Air-gapped AI inference

The challenge

Models must run where the data is generated, with no route to the public internet and no hosted API to call.

What we build

A Kubernetes platform that installs and updates from an offline registry, shares GPUs across workloads and keeps telemetry inside the perimeter.

Engineering considerations
  • Offline software supply chain
  • Patching without connectivity
  • Showing auditors that nothing leaves
Project outcome
Video AI that runs where the cameras are.
Faster incident response.Video AI runs locally, inside the customer site.
Routine releases.Helm rollouts from an offline registry replaced releases that took days.
Better use of every GPU.Slicing and time-slicing share each card across streams.
Security technology · individual engagementView case study
02
Example application

A private cloud in a week

The challenge

Teams that cannot use public cloud still need what it provides: Kubernetes, storage, identity, GitOps and observability, plus evidence for auditors.

What we build

The Sovereign Platform: a security-hardened, 100% open-source private cloud on your hardware, built as code from firewall to observability, with controls mapped to SOC 2, ISO 27001 and CIS benchmarks.

  • Bare metal to production in one week
  • Zero licence fees
  • A quarterly compliance evidence pack
Engineering considerations
  • An air-gap mirror for images and packages
  • Identity, just-in-time access and runtime security
  • A restore rehearsed before hand-over
Delivery by design

Five days to production.

The Sovereign Platform build

  1. 01Day 1Bare metal, network, DNS, PKI
  2. 02Day 2Kubernetes, etcd, storage
  3. 03Days 3–4Identity, GitOps, security, observability
  4. 04Day 5Validation, restore rehearsal, hand-over
Bare metal to production in a week.
03
Example application

Open models on your own GPUs

The challenge

If your work runs on a model you rent, you do not get a say when it changes, and sensitive prompts and documents cannot go to a hosted API.

What we build

The AI Control Plane: KServe and vLLM on your GPU node pools, a versioned and signed model registry, llm-d routing and per-team quotas, with every request authenticated and audited.

  • Prompts, documents and embeddings stay in the building
  • The same answer tomorrow as today
  • Assistants, summarisation and document analysis first
Engineering considerations
  • GPU partitioning and per-team quotas
  • Guardrails at the gateway
  • Cost and usage per team
Architecture in practice

Your data. Your boundary.

The AI Control Plane, conceptually

Your environmentLocal
Prompts & documents
Open models on your GPUs
Audit trail

Access control · Lineage · Residency checks

Nothing leaves the perimeter

Built with accountability

Trust you can
build on.

Independent assurance behind how we protect information and operate our business.

Information securityDocumented controlsClear accountability
Discuss your security requirements
INTERCERT ISO/IEC 27001 certification mark
Certified · INTERCERTISO/IEC 27001:2022

Information security management

AICPA SOC 2 for Service Organizations logo
CertifiedSOC 2 Type II

Controls assessed over time

Your next step

Tell us what has
to run, and where.

Your users, workloads and constraints are where the design starts.

Book a 45‑minute discovery call