Frontier models
Best results, fastest, no training on your data. The trade-off is plain: the data leaves the EU.
01Where does your data go once AI touches it?
Sensitive data doesn’t have to leave your environment for AI to be useful. We build with cloud, private and locally hosted models, choosing the setup that fits the data, the risk and the job.
The question that stalls AI projects
Most companies know where their data is stored. With AI a second question appears: where does it go the moment a model processes it.
There’s no single right answer. There are three paths with different costs, guarantees and limits, and the right one depends on your data.
Why no client names2x10
Our clients hire us to build what gives them an edge. Confidentiality is part of the job. Publishing the details could give that edge away. We’ve delivered in regulated environments and on projects worth hundreds of millions.
Where it runs
Three paths, and how we choose
EU region
The same top models on European hardware. Not a proxy to the US: the model runs in the region.
02Austrian hosting
Open-source models in an Austrian data center we partner with. The data never leaves the country.
03Most of the time the right answer is a mix: the sensitive part local, the rest in an EU region. And one honest limitation belongs here. Open-source models are, as a rule of thumb, about half a year behind the frontier. For most tasks that’s enough.For the most demanding ones it isn’t always
What we've built
An internal system for specialist processes at a client, with the model running on our own infrastructure in Vienna. No US hyperscaler anywhere in the processing path.
For health-grade data, processing runs in an EU region with an additional security layer in front of it. Identifiers are pseudonymized before anything is logged, and special categories of personal data are filtered out of what the system keeps.
Documents run through a model in an EU region, behind an abstraction so the model underneath can be swapped without touching the pipeline. The intermediate images live in EU storage only, reachable through short-lived access and deleted immediately after processing.
Production questions2x10
What makes them hold up
Four questions decide this, and only one of them is about where the data is hosted.
- Who operates it in year two?Starting a local model is easy. Running it for two years, with the outages, fallbacks and upgrades, is the part nobody demonstrates.
- Cloud or not cloud?The real question is which data flow may go where, decided per use case. A blanket policy is comfortable and usually wrong in both directions at once.
- What happens when it’s misconfigured?It has to refuse service rather than expose data, which is an architecture decision, not an operational detail.
- What does "GDPR-compliant" actually mean?Where the data sits, who can see it, and what each path costs. We show all three, including the one that argues against the option we would otherwise sell you.
What we could build for you
- We rebuild existing AI applications so sensitive data never leaves your infrastructure.
- We determine the right path per use case, instead of running everything locally on principle.
- We keep models interchangeable, so no provider ever becomes impossible to leave.
- We make it traceable what a model gets to see, and what it doesn’t.
- And whatever you have in mind that AI and solid software engineering could make possible.
Under the hood
Deployment · self-hosting · EU regions · open-source models · model routing · Security · zero trust · encryption at rest · fail-closed design · data classification · Governance · prompt logging · audit trails · rules for what may leave the system
Most engagements start small: talking it through, a few weeks of engineering, something real you can judge. If it works, we take it further.If it doesn’t, you find out early
More work2x10
Platforms
Built to last
View themePrototype to Production
Make it real
View themeAI Governance
You stay in charge
View themeN° 06 Contact
Tell me what you’re planning
No form, no ticket system. Email me directly and you get me, not an inbox someone else manages.
Moritz Miedler · Founder, replies personally2x10 Technologies · Viennahello@2x10.com