GDPR-Compliant AI Infrastructure in Europe

No. All inference runs as stateless computations. We do not store the content of your prompts or AI responses in databases or object storage, only processing them in RAM during computation.

No. Customer data is never used to train or fine-tune our models. Training only occurs on separate, controlled datasets – never on individual customer prompts or responses.

By default, we only log technical metadata (e.g., customer ID, model, token count, response time) for operations, billing, and capacity planning. No prompt or response texts are stored in the default mode.

Yes. For certain customers and use cases, we can activate content logging for debugging or quality work. In that case, prompts and responses are logged for a limited time period and automatically deleted after X days according to the agreed retention policy.

We have a documented data flow map, clear logging policies, and separate environments per customer. Additionally, you can access our security procedures, audit documentation, and – for larger customers – conduct technical and legal reviews together with us.

For customers in defense and critical infrastructure, we can run dedicated instances with enhanced transparency. They gain access to their own log data, can review which fields are logged, and set their own strict retention (e.g., automatic deletion after X days).