Do we need a data centre or a Kubernetes cluster?
No. The default deployment is a single workstation with a
consumer GPU (e.g. RTX 4070-class) sitting in your server room
or under a desk. We help you size it on the demo call.
What if our internet connection goes down?
DEM keeps working. The model, the index, and the UI all live
on your LAN. The internet is only used for software updates
(and the Sovereign tier doesn't even need that).
Which model does DEM use under the hood?
An open-weights model selected for European-language quality
and on-prem feasibility. We pin a specific version per release
so behaviour doesn't drift mid-quarter. We can run customer-
specific models on the Sovereign tier.
How does this comply with the EU AI Act?
DEM is a general-purpose AI system used as a knowledge tool —
not a high-risk system in the Act's sense. We provide
transparency documentation, model cards, and an audit log to
support your internal AI governance. We're not your lawyers,
but we make compliance easier rather than harder.
Can a junior employee ask anything they want?
Only what their folder permissions allow. DEM respects the
same access controls your file shares already enforce, and
role-based filters can further restrict by document tag.
What if a document needs to be forgotten?
One click in the admin UI removes the document and every
embedding derived from it. We log the deletion for your DPO's
records.
How long does deployment take?
A small SME with a few hundred thousand documents is usually
live in two to five working days, including the first round of
ingestion. Bigger archives take longer to ingest — but DEM is
answering questions while it ingests.