Why sovereign Dutch medical speech recognition has to run on-premise.

In Dutch healthcare and government, the best cloud speech model in the world is unusable if the audio has to leave the building. That constraint — not raw accuracy — is what shapes the market.

The cloud is a non-starter for regulated Dutch audio

Patient recordings, mental-health sessions and government dictation carry some of the most sensitive data there is. GDPR, NEN 7510, the EU AI Act and data-residency expectations make sending that audio to a US-hosted API a non-starter for many buyers — regardless of how good the model is.

So the real question isn't "most accurate" — it's "most accurate that runs here"

The engines that can run on-premise have historically been well behind on Dutch. The engines that are accurate are hosted. That gap is the whole opportunity: accuracy that a hospital, a GGZ provider or a ministry can run inside its own walls, air-gapped if needed.

Our position

We build for exactly that constraint: a Dutch speech engine that leads the public benchmark and runs fully on-premise, with hybrid post-quantum encryption for data at rest. Data never leaves the customer's infrastructure, and the accuracy doesn't drop to buy that.

Why now

Regulation is the tailwind. As the EU AI Act and sector rules tighten, "where does the data go?" becomes the first question in every procurement — and sovereign, on-premise AI stops being a nice-to-have.