Sourced articles on on-premise AI, data sovereignty, and the regulation that really affects the Mittelstand — and honest answers to the questions that always come up in the intro call.
Endless generations nobody stops, an inbox full of ‘Other’, invented search results – five real failure patterns from our agent engineering and the architecture that follows.
Read article →Free models, exposed credentials, anonymization that does not hold — and why the Trusted Computing Base is the real question. With sources.
Read article →A sovereign cloud is still a cloud. Your engineering data leaves your premises, the model weights belong to someone else, and the contract term is shorter than your equipment lifecycle. And then there is physics: robotics, manufacturing IoT, and production control rooms need response times a cloud connection simply cannot deliver. A sovereign cloud contract is a rented flat in a foreign city. We deliver a house.
The physical location is not the same as legal certainty. A US-owned provider is subject to the US CLOUD Act — regardless of whether the servers sit in Frankfurt. A German address is a legal mitigation, not a solution. If your data sits in your own server room, there is no third party that could be compelled to hand them over — not because it is contractually excluded, but because there is physically no one for an order to be addressed to.
An integrator sells you the metal — cleanly racked. But connecting your data via RAG and agent orchestration are rarely in-house. We deliver all three layers as one vendor: the hardware, the local model, and the agent layer that orchestrates your business processes. One responsibility, one invoice, one phone number.
The hardware stays, the model moves up. When a better open model appears, it can be rolled out as an update — we agree any model change with you beforehand, since it can change how your workflows behave. You are not buying a frozen model, but a platform that can stay current, without your data ever leaving the building.
Not necessarily. We deliver the appliance ready to install and hand it over in full with documentation — your team uses the system through an interface they know from the cloud. If you'd rather not run day-to-day operations yourself, you can add support optionally; scope and terms are agreed in the intro call.
From site visit to production we work in clearly defined phases, billed per phase — how long it takes depends on scope and data maturity. If the first phase — the readiness check — shows your data is not ready yet, we say so before any hardware is ordered. Honesty is cheaper than a pilot that never reaches production.
Ask it in the intro call — thirty minutes, engineer to engineer, no sales funnel.