Skip to content
BionicAdri

Where your data and BionicAdri's AI live

Each client has their own server in Europe and the AI runs on GPUs in Germany and Poland. We also tell you what doesn't always stay in Europe and why.

Mapa esquemático de Europa con Alemania, Finlandia y Polonia unidas por líneas y un candado

A question that deserves an exact answer

When a company puts its invoicing, payroll and customers on a platform, the underlying question is simple: where is my data and who touches it? In this article we answer with what's there today, without any frills. It includes what isn't “all in Europe”, because saying it clearly is what builds trust.

One server per client

At BionicAdri we don't mix companies into a single database. Each client has their own server: their own installation of the platform, with their own database. What one company stores doesn't share storage with another's.

Those servers are in Hetzner data centres in Europe, in Germany and Finland.

Backups

  • Two backups a day of the entire server, not just the data.
  • Database backups every 4 hours.
  • Database backups leave a trace of who downloads or restores them.

How the servers talk to each other

The servers communicate over an encrypted private network. The AI gateway, the component through which all ADRI requests pass, is exposed only through an encrypted tunnel: it has no ports open to the internet.

On 10 October 2026 we hardened this layer:

  • firewalls on the servers;
  • access by key only;
  • services isolated from one another;
  • secrets (passwords and internal keys) rotated;
  • automatic security updates.

In addition, the application doesn't load third-party CDN code: code is served from our own servers. It's one less thing to rely on anyone for.

Where ADRI's AI runs

Here's what people ask about most. AI doesn't live in a single place, and it's worth explaining each component.

The in-house gateway

All ADRI requests go through our own AI gateway. It does four things: measures and bills each use per client, distributes work between models, applies spending caps and decides where each request goes. It is also the only place from which you can see what each company consumes.

The models

ComponentWhere it runs
Main model: Qwen (open weights, open source)Dedicated GPUs rented in Germany
Model that makes decisions, voice recognition and synthesis, and search in your documents (RAG)Another dedicated GPU in Poland

Most of our clients' AI traffic goes to the main model, the one running in Germany.

What isn't in Europe: backup

We're not going to say “zero data outside Europe” here, because that wouldn't be true. These are the exceptions:

  • Model fallback. When the main model is overloaded or fails, the gateway can pass the request to external model providers, some of them outside Europe.
  • Image and video. Image and video generation uses external services.

Why does that fallback exist?

So that work does not stop. If on a Monday at nine the main model is overloaded, we prefer your team's task to go another route rather than wait. It is a trade-off: more continuity in exchange for the request being able to leave our infrastructure in those cases.

Whether that use applies to your company depends on the configuration. If it is a condition for you, tell us before starting and we will review it with you.

If your data cannot leave your network

For companies that need it, there is another way: install the platform and a dedicated AI machine in your own office. It is the 'everything in your office' option that we explain on the pricing page. This way, your data and your queries do not leave your network.

The public website

The public website, the one you are reading now, is behind Cloudflare, which provides a distribution network and protection. It is the showcase website; your company's data is on your server, not there.

Questions we get asked

Is my data mixed with other companies' data?

Not in the database. Each client has its own server and database.

How do I know how much my company spends on AI?

Because everything goes through our own gateway: each use is measured per client and spending caps can be set. There is no 'hidden' usage behind it.

What if the data centre fails?

That is why we back up the entire server twice a day and the database every 4 hours. If you want to see how the AI part behaves during an outage, we explain it in this other article.

Who can see my backups?

Database backups leave a trace of who downloads or restores them, so that any action is logged. For how we protect clock-ins, signatures and payslips, there is a separate article.

In summary

  • Each client's data: own server, at Hetzner (Germany and Finland), with two daily server backups and database backups every 4 hours.
  • Communications: encrypted private network; the AI gateway only through an encrypted tunnel.
  • Standard AI: Qwen on GPUs in Germany; decision, voice and RAG on a GPU in Poland.
  • Exceptions: fallback with external models (some outside Europe) depending on configuration, and image and video with external services.
  • Total control: installation in your office if you need it.

You can see the rest of the guarantees in Security and compliance. And if you want to review with us what your case would look like, write to us at [email protected] or request a demo.

  • Security
  • Europe
  • Hetzner
  • RGPD
  • IA

No-obligation demo

See it with your company's data

We show you ADRI working on a real task of yours, in about 15 minutes.

[email protected]+34 604 87 17 35