LoyJoy Now Runs AI Models Entirely on Its Own Hardware
EU data sovereignty is becoming a strategic question
More and more companies in regulated industries, from insurers to banks to utilities, ask themselves the same question before adopting an AI solution: where is the data actually processed, and who might have access to it? This is no longer a compliance detail to sort out at the end of a vendor selection process. It increasingly decides which vendors make the shortlist in the first place.
We now have a clear answer to that.
AI models on our own hardware in a German data center
For our chat agents, we run our AI model on our own hardware in a German data center, not in a rented cloud. This removes the use of external AI subprocessors entirely for chat: your data is processed exclusively by LoyJoy itself.
There are also concrete data protection benefits: the US Cloud Act attack surface is completely eliminated. As soon as a provider with a US parent company is involved anywhere in the chain, even just through the hosting server, a theoretical access by US authorities can never be fully ruled out legally. If inference instead runs entirely on our own hardware in Germany, this risk is removed completely for chat.
If you prefer, you can still additionally choose current models such as GPT from the EU for chat, that remains your choice. But the default is now a model that we run entirely ourselves.
No compromise on performance
Moving to a self-hosted model would not have been possible at comparable quality two years ago. Today, our customers rate the model we use on accuracy, speed and reliability as on par with established large language models from well-known providers, and these are exactly the three criteria that matter in customer dialogue:
- Accuracy. The answers hold up.
- Speed. Noticeably shorter wait times.
- Reliability. Consistent quality, precise instruction-following.
This is made possible by the open-source model Gemma 4, which we host and run ourselves.
What this means for you
You no longer have to choose between AI performance and data sovereignty. The default path for chat now runs entirely at LoyJoy in Germany, without giving up current model quality. For companies in regulated industries, this is more than a technical detail, it is a building block for your own risk assessment and for demonstrating compliance to regulators and customers.
You can find the full list of our subprocessors, as always, in the Trust Center, and details on compliance and data sovereignty on our compliance page.
Where does the data actually go when an AI agent answers on your website?