// Manifesto

Why we build on open models.

Not for ideology. For a concrete reason: on sensitive data, who owns and controls the model matters more than a few points of performance on a benchmark.

Definition

"Open" models, in brief.

An open (or open-weight) model is an AI model whose weights — the trained "brain" — are publicly available and can be run on infrastructure of your choosing. The alternative is closed models, accessible only through a vendor's API: powerful, but "black boxes" you have to send your data to in order to get an answer.

// the practical difference

With a closed model, the data goes to the model; with an open model, the model comes to the data.

Four reasons this matters.

1 · Data sovereignty

With an open model running internally, sensitive data never leaves your perimeter. No sending it to foreign APIs, no transit through cloud infrastructure outside Switzerland, no risk of your content feeding the training of a third party's model.

2 · No lock-in

A model you own is portable and stable. You don't depend on one vendor's decisions: price increases, changed terms, deprecated features or sudden service interruptions don't expose you. The system you build today remains yours tomorrow.

3 · Transparency and control

Open models are inspectable. You can verify their behavior, adapt them to your domain, subject them to security review and document them for an audit. In a regulated context, being able to explain how a system works matters just as much as the fact that it works.

4 · Economics at scale

Above a certain volume of usage, owning your own infrastructure becomes more predictable and more economical than a per-call cost that grows without limit. The costs are yours, controllable and plannable.

// Intellectual honesty

What about closed models? Let's not pretend they have no advantages.

It would be dishonest to say open models always win. On some particularly complex tasks, the best closed models remain ahead today, and adopting them is immediate because it requires no infrastructure.

That's why our approach is pragmatic, not dogmatic:

default

Open by default on everything that touches sensitive, confidential or regulated data.

hybrid

Hybrid where a non-critical task genuinely benefits from a frontier model's capability — and always with anonymized or non-sensitive data.

boundary

The decision on where the line sits is yours, made together with us based on facts, not accepted by default.

The point isn't "open versus closed." It's putting every piece of data where it belongs.

Want to work out the right mix for you?

Book a call →← Back to home