Open models, operated entirely in Germany
Intra AI uses open-weight models exclusively. We keep widening and updating our model selection.
- Qwen
- Gemma
- OpenAI
- Mistral
- DeepSeek
The models from these providers are publicly available and are run by German operators in Germany. Your inputs never reach the cloud outside the EU.
Open weights, run in Germany
At its core, a model is a file containing weights. Whoever published it does not decide where your data goes. That is decided by whoever operates the model. That is precisely why we use open models on infrastructure in Germany, rather than sending every request to a third-party provider’s API. Your administrators decide which models are available in your instance. However, you retain the option to connect any endpoints to Intra AI under your own responsibility.
Why we run open models only
An open model is published as a file of weights. Anyone who wants to use it downloads it and runs it on hardware they own or rent. That is the distinction that matters in data protection terms: whoever published the model does not decide where your data goes, whoever runs it does.
What if you still want to use other models?
Technically, an external AI provider can be connected as well. Whether that makes sense and is legally sound in your case is something we look at together: with U.S. providers we can no longer guarantee data processing exempt from the U.S. Cloud Act. We’ll discuss this further during our initial consultation.
What your documents pass through before a model answers
For document search your text is split into passages and turned into vectors that sit in your instance’s own database. The retrieved passages are then ordered by relevance. Open models do both, and we run them in Germany.
What this gives your data protection officer
For a record of processing activities it means one named processor per model endpoint rather than an open chain of AI vendors, and no transfer to the United States for AI inference. That your content is never used for training is a commitment we make contractually. The Article 28 GDPR data processing agreement, with the list of subprocessors, is available before you sign.
Frequently asked questions
Which AI models does Intra AI run?
Open-weight models only. The providers whose models are in the selection today are named above; which of them are available inside your instance is your administrators’ decision.
What does “open weights” mean?
The model’s trained parameters are published, so anyone can download them and run the model themselves. We run open-weight models only. That is a decision about how we operate, not a limit of the platform: an administrator can add a further provider at runtime.
Are my inputs sent to the providers of those models?
No. These companies have published the model weights; the models are operated by a German provider in Germany. There is no connection to an interface provided by the mentioned providers.
Where are the models run?
In Germany. Development, data storage, and AI inference take place exclusively on German servers.
Is my data used for training?
No. Neither we nor the subprocessors we use apply your content to train or retrain a model. We commit to that contractually.
Can we connect our own or other models?
Yes. An administrator can add a further provider with a compatible endpoint to your instance, including a model you run yourself. Our default is open models that we run in Germany; if you want to depart from that, talk to us first so we can work through what it means for your data processing agreement.
Which model suits you is a conversation, not a checkbox
Which one answers your questions best is something we work out together – including when the smaller one is enough.
Qwen, Gemma, OpenAI, Mistral and DeepSeek are trademarks of their respective owners. They are named here solely to identify the models whose open weights we operate. No affiliation, sponsorship or endorsement by those companies is implied.