How we build it, and how you keep it
The biggest worry when buying from a small supplier is dependence. So the architecture is set up so that you can carry on without us. That is a decision in the design, not a reassurance in a meeting.
Built so you can leave
- 1Your own environment
Every client runs their own automation environment, on their own infrastructure or in their own cloud account. We manage it; you own it. If the relationship ends, everything stays where it is and keeps running.
- 2Error handling before features
What happens when an integration goes down, a format changes or a field is empty? We build that first. A process that stops and says so is better than a process that quietly passes on the wrong data.
- 3Everything is logged
For every order or invoice processed you can see what happened and why. You need that for your own checks, for the GDPR, and it is the basis of the monthly report.
- 4People decide where people should
Below the confidence threshold the work goes to a person instead of through. Where that threshold sits is decided together with whoever does the work today.
- 5Documentation as we build
Not a handover afterwards but documentation that keeps pace: what each integration does, which assumptions are baked in, and where to look when something breaks.
- 6Licences and hosting in your name
Software licences and hosting run directly through you: no reselling, no margin, no keys left behind with us. If you would rather not manage that yourself we can host it — and even then the environment stays transferable. That is the condition we hold ourselves to.
Where your data stays
The moment we classify email, process invoices or read CVs, we process personal data on your behalf. That makes us the processor and you the controller, with the obligations that come with it. We arrange that before we build rather than after your accountant asks.
Where a process needs a language model, the default is that we run our own — on your infrastructure or on ours, inside Europe. That is not a principle for its own sake. It means there is no processor to sign an agreement with, no transfer outside the EU to justify, and „your data is not used to train models” is simply true rather than a promise borrowed from a supplier.
Worth knowing, because it is the distinction almost nobody makes: an EU region with an American provider is not sovereignty. Azure OpenAI in West Europe and AWS in Frankfurt fall under the US CLOUD Act — the provider can be legally compelled to hand over data regardless of where the servers sit. Sovereign means a model on your own hardware, or a provider that is itself subject to European law.
- Data processing agreement as standard. Required under article 28 GDPR. You get it with the proposal, not on delivery.
- Sub-processors named. If an external service is involved, the proposal says which one, what for, and where the processing happens.
- Nothing outside Europe unwritten. No data goes to a non-European service unless it is in the proposal before we build. That is the one promise we make without conditions.
- No training on your data. Trivially true when the model is ours.
The honest trade-off
Models you run yourself are weaker at difficult extraction than the largest hosted ones. The design absorbs that: every extracted field carries a confidence score, and below the threshold a person decides. The effect of a weaker model is therefore more exceptions, not more errors — and that is measurable, adjustable and visible to you.
It also means no per-token cost, and no model drift. A hosted model changes underneath you; one you deploy does not. For a process that approves invoices, being able to reproduce last month's result is not a luxury.
Start with one process
Tell us which work gets retyped most often at your company. If automating it turns out to be a bad idea, you will hear that in the first conversation — before there is an invoice.
The first conversation is free and without obligation. Want it substantiated? The Automation Check is €295 — deducted if you have us build within 30 days.