AI that only sees what it is allowed to see.
Assistants, search over your own documents and automation, connected to your systems. With a permission check before every answer and a model that meets your data protection requirements.
1Typical projects
Where language models actually take work off your hands.
- Assistants
- For staff or students, with answers drawn from your own sources.
- Search over your own documents
- Policies, manuals, wikis: ask questions instead of searching through folders.
- Analysis & summaries
- Turn long texts, feedback or minutes into structured overviews.
- Automation
- Pre-sort requests, prepare drafts, extract data from documents.
- AI in Moodle and ILIAS
- Support for teachers and learners directly in the learning platform.
- Review of existing AI projects
- Permissions, data flows and prompt injection, checked as in a security audit.
2Example
What a person may not see, the assistant does not see either.
SituationAn assistant is to answer questions about internal documents. Not every person may see every document.
- ①Permissions before the answer. The assistant only sees what the person may see.
- ②Model to your requirements. Run locally or with a provider in the EU.
- ③Traceable. Which sources went into which answer.
3Frequently asked questions
Where does our data go?
Where you decide: to your own servers or to a provider in the EU. We agree on this in writing before the project starts.
Which model do you use?
That depends on the task, data protection and budget. We compare options and recommend one, instead of selling one model for everything.
Do we need our own hardware?
Not necessarily. For locally run models yes, for EU-hosted models no.
What about prompt injection?
We treat inputs and documents as untrusted. The assistant only gets the permissions it needs, and we test specifically against injection.
4 / Contact
What do you want to use AI for?
Describe your project in three short steps. We usually reply within one working day.