Skip to main content

Notes on data protection

The models provided by mittwald’s AI hosting are tailored for data-sensitive applications used by agencies or their end customers. All models run on mittwald's infrastructure and are integrated into our existing hosting environment. If mittwald’s AI hosting is accessed through a mittwald hosting product, all data processing remains entirely within our infrastructure.

All communications with the offered models are encrypted during transmission according to the state of the art (https).

The models provided are not self-learning; each model remains in the state in which it was published. Queries or submitted data do not influence the model, nor are they aggregated, stored, or used to improve the models.

Our models operate in a stateless manner, meaning no content-related information from submitted inputs, outputs, or prompts is stored or processed for other purposes. However, for billing, performance analysis, and system stability improvements, we retain log files and metadata of submitted requests. These include IP address, API key used, timestamp of the request, and number of tokens transmitted—but not their content.

Our product is designed for GDPR-compliant usage, as no data is transferred to third parties and no content data is stored. However, we point out that GDPR and legal compliance ultimately depend on the specific use case. For example, the models in mittwald’s AI hosting can be used to build your own AI applications. In these cases, it may be necessary to obtain user consent if personal data is processed, or the algorithm generated with our product may not comply with legal requirements such as the EU AI Act. We have no influence on this and refer to the Terms of Use.

If user applications store model-generated content or embeddings, this storage or processing occurs within the services and hosting products used by the application—but not within the scope of mittwald’s AI hosting. For example, Open WebUI can run in mittwald’s container hosting. The chat history or uploaded documents are stored within these containers in the corresponding hosting infrastructure, not in the models themselves.