Skip to content

Sovereign AI chat

The chat is where most people start: ask a question, get an answer, get work done. Two things make it different from the chat you already know.

Your data stays in the EU and never trains a model. Not as a setting you have to find — as the default the workspace runs on.

The assistant can act, not only answer. It researches, drafts, reads your documents, prepares work in connected systems — and every step that reaches the outside world waits for your approval.

A US provider is subject to US law wherever its servers physically stand. That is not a hosting question, it is a legal one, and it applies to everything you type into a chat window — which for most companies means pricing, personnel matters, contracts and half-finished ideas.

Your chat history is trade secrets. It is worth treating it that way.

On this platform: EU-only processing, no training on customer data, no resale, and an audit log that records which model saw what.

Open-weight models on EU infrastructure handle the everyday work. For the tasks where a frontier model is genuinely better, your workspace policy decides whether that is allowed — and each such call is logged.

Two controls do the work:

  • The model tier picks the class: EU-sovereign, frontier, or cheapest-that-fits.
  • The PII flag arms the gate: a request marked as containing personal data is refused a non-EU model centrally, not by convention.

The effect: the default path never leaves the EU, and using a frontier model is a deliberate, logged decision — the kind of evidence an EU AI Act file wants to see.

Beyond answering, the assistant can run real tasks: research a topic across sources, draft a document from your own files, prepare a reply, schedule something, pull numbers out of a connected system.

Anything with a consequence outside the workspace — a message sent, a document published, money moved — becomes an approval you decide on. The work happens without you; the commitment does not.

Upload contracts, manuals, policies or reports and ask questions against them. Answers cite the passage they came from, so you can check the source instead of trusting a summary.

Files stay in your workspace and are processed in the EU, like everything else.

Chat is flat on every plan — the conversation does not run on a meter, and heavy use never leaves you staring at an empty wallet mid-thought.

Credits pay for autonomous work: tasks the AI runs on its own, image and video generation, coding sessions. That way the thing you do all day is predictable, and the thing that consumes real compute is what you pay for.