Your AI, in your building.
One box, sized to your business, running your models, your memory and your tools. It answers when the internet doesn't, it keeps your documents where they already are, and nobody else gets to change its terms.
Sovereignty is continuity.
In June 2026 an export directive knocked frontier models offline overnight for foreign users. Businesses that had quietly built their operations on a remote API woke up to find the API was a policy decision made in another country — one they had no vote in and no notice of.
That is the argument for local, and it isn't a privacy argument. It's an uptime argument. Your bookkeeping, your support desk, your document search and your internal knowledge should not have a kill switch held by someone whose incentives you don't control. A model running on hardware in your office cannot be switched off by press release.
We still use frontier models where they earn their keep — for the irreducible reasoning, on demand, with your consent. The difference is that they become an accelerator you can unplug rather than a dependency you can't survive.
Discovery, box, wiring, care.
- Discovery. We map what your business actually runs on, what leaves the building today, and where a model is quietly deciding something code should decide. You get a written sizing and roadmap — yours to keep, and credited against the build if you go ahead.
- The sized box. We specify the smallest machine that comfortably does your job, buy it in your name at cost, and provision it. Not the biggest box we can sell you — the one your workload actually needs.
- Local models & Mnemos memory. Open-weight models run on the appliance. Mnemos gives them durable memory of your business — decisions, documents, history — so the assistant stops starting from zero every morning.
- MCP wiring into your tools. The appliance is connected to the software you already use through Model Context Protocol connectors, so the AI can read and act inside your real workflows. Everything irreversible waits for a human.
- Sovereignty Care. A monthly plan keeps it current: model and security updates, backup verification, health checks, and a human to call when something looks wrong.
Three sizes, bought at cost.
Indicative street prices as of August 2026, for orientation only — hardware is quoted at the time of purchase, bought in your name, and invoiced at cost with the supplier receipt attached. The right tier falls out of discovery.
Value
A Ryzen AI Max class mini-workstation. The right answer for a small team that wants local drafting, document search and bookkeeping support without a server room.
Compact
Mac mini or Mac Studio class. Quiet, small, unified memory — the office-friendly tier when the appliance lives in a room where people work.
Performance
NVIDIA DGX Spark class. For larger models, heavier concurrency, or workloads where the appliance is doing real inference all day rather than in bursts.
Start with discovery.
Tell us what your business runs on and what you'd hate to lose access to. Discovery is €1,500, you keep the roadmap either way, and it's credited against the build if you go ahead. Sovereign, humanity-first AI is not a slogan here — it's the reason the box is in your building and not ours.