Wrapper or standalone platform?

What is a wrapper?

A wrapper is a shell. The term means an interface that sends requests to another provider’s AI model. The interface can offer a lot: chat, assistants, search in documents, permissions. But the answer is computed by the model provider’s model, in that provider’s data centre. Many offerings on the market work this way, even when the interface is hosted in Germany.

Examples, according to the providers (as of 11 October 2026):

• DeutschlandGPT bundles models from several providers, including OpenAI, Anthropic and Google, and stores data in the Open Telekom Cloud in Germany.
• Langdock offers models from more than 30 providers; the platform runs on Microsoft Azure in the EU.
• amber uses models from OpenAI, Anthropic, Google, Mistral AI and Z.ai, operated in the EU. At its core is the Data Layer, which opens up knowledge from the connected systems with the existing permissions. Platform and Data Layer run in the T Cloud in Germany and the EU and, according to the provider, also on-premises in the Enterprise package; there, customers can also use their own model keys.
• Kommunen-Chat is hosted at Hetzner in Germany and uses models from Mistral AI in France.
• LLMoin, according to the City of Hamburg, uses an OpenAI model on Microsoft infrastructure in Europe.

This is not a judgement: for many tasks, a wrapper is a good choice. But it is a different design from a platform that runs the model itself.

What is a standalone platform?

A standalone platform runs the AI model itself: in its own data centre or on a server at the customer’s site. The request does not leave the platform provider’s environment. These are usually open-source models that the provider selects, tests and updates. Model vendors such as Mistral AI or Aleph Alpha also offer to run their models at the customer’s site.

How do the two differ?

Feature	Wrapper	Standalone platform
Where the model runs	At the model provider	At the platform provider or at your site
Who processes the data	Platform and model provider	Platform provider; with a server on your premises, you yourself
Sub-processors	Model provider is on the list	Model provider only if additionally used
US CLOUD Act	Applies if a model provider is subject to US law	Does not apply if platform and model have no US connection
Choice of models	Often large, newest models available quickly	Selected models, new ones via updates
Operation without internet	Not possible	Possible with a server on your premises
Getting started	Often low-cost per account and month	Depending on the provider, per person or per server

What does this mean for data protection and contracts?

• Chain of processing: The model provider is a sub-processor of the platform (Art. 28(2) and (4) GDPR). You need its terms, its location and its commitments, not only those of the platform.
• Third country: If a model provider is located outside the EU or is subject to US law, Art. 44 et seq. GDPR and the US CLOUD Act come into play. More in the guide US CLOUD Act.
• Storage at the model provider: Some providers log requests for abuse detection, others commit to zero data retention. What that means is explained in the guide Zero data retention.
• Professional secrecy: Under § 203 StGB (German Criminal Code, professional secrecy), every service provider involved must be bound to confidentiality, including the model provider.

How is basebox built?

basebox is a standalone platform. The AI model in basebox runs in the cloud on our servers in the Munich-East data centre, or on a server just for you, at your site and, if you wish, without an internet connection. basebox does not develop its own model but runs open-source models itself; new models come via updates, tested by us beforehand. Your data does not train any model.

Other models are possible, but never without your knowledge:

• App with an external model: An external model is only available as a separate app, if you explicitly ask for it to be set up. This app runs on Google Vertex in the EU; Google is then a sub-processor, and because Google is a US company, the US CLOUD Act applies (app with an external model).
• Additional computing power: In the cloud, we use it where you want it, for example at STACKIT or Mistral AI or through your own framework agreement; that provider is then on the list of sub-processors. We only use offerings that do not train on your data.
• Hybrid: basebox runs on hardware in your own network; the AI model comes from elsewhere.

What do wrappers offer that basebox does not?

• Free choice of model for every person: Many wrappers offer models from numerous vendors in one interface. With basebox, everyone works with the AI model in basebox.
• Newest models from major providers, often on the day they are released.
• Low-cost start for small teams, sometimes cancellable monthly; the basebox cloud starts at 5 people and runs for at least one year.

What suits whom?

• A wrapper suits you if your data may go to the model providers and you value a choice of many models.
• A standalone platform such as basebox suits you if patient, citizen or social data is to go in and the model is to run in Germany or on your own premises. The calculator works out the prices.