Where your data goes, step by step
Follow one message from your browser to the AI model and back: where it is processed, who may keep a copy, and the one option that can send anything outside Europe.
Four steps, the same for every message
The dashed line marks the EU and EEA. Nothing crosses it unless you switch on Google web search.
- 01
Your browser
An encrypted connection takes your message to our servers in a Hetzner data centre in Germany.
- 02
EurekaGPT
Our server adds two rules to every request: EU providers only, and zero data retention. Both are fixed in our code.
- 03
CORTECS
CORTECS (Cortecs GmbH, Vienna) picks a provider that meets both rules. If none can, the request fails and you are told.
- 04
The model
The answer comes back with a record of who served it and where. The Trust panel under the answer shows it.
“Established in the EU” and “processed in the EEA” are different facts
Both are true for EurekaGPT. One does not prove the other, so we state them separately.
Jurisdiction
Which law applies to the company
The company that serves the model is established in the EU. For a US-made model, that is a European company such as Google Cloud EMEA Ltd or Microsoft Ireland Operations Ltd.
Data residency
Where your data is processed
Your message, and the technical data around it, are processed in the EEA. This is a separate rule.
What we say: served by providers established in the EU, processed in the EEA.
Why we say EEA (the EU plus Iceland, Liechtenstein and Norway) and not a country: the short answer.
The model and the web search are two different things
The model writes the answer, and it always runs in the EEA. Web search is a separate setting with three options. It is the only part that can reach outside Europe, and only if you choose it.
| Option | What happens | Label on the answer |
|---|---|---|
| Off | The model answers from what it knows and any documents you attached. No web search. | In Europe |
| EU search default | Our server searches Staan, a European search engine run by European Search Perspective SAS. It gives the results to the model, which never goes online itself. | In Europe · web search |
| Google’s own search, built into Gemini models. Your search words go to Google in the United States. Off by default, needs your consent, not available in Ephemeral chats. | Left Europe · Google |
Choosing a Gemini model does not switch Google on: it stays off until you turn it on yourself. Try the Google switch on the home page.
Zero data retention and Ephemeral cover different places
They apply at two different points of the route. See both on a timeline.
Zero data retention Always on
The company that runs the model may not store, log or train on your message. We add the rule to every request, and no setting removes it.
Ephemeral Off by default
By default we save your chats in Germany, for 7, 30 or 90 days (the default) or 1 year after the last message. An Ephemeral chat is never stored: no messages, no title, no files.
Your conversations are stored in the EU so you have a history. Or choose Ephemeral, which keeps no message, no title and no file. The models that answer are EU-resident and keep nothing of the call.
What is fixed, and what you choose
Always on, whatever you choose
- The rule “EU providers only” is fixed in our code. No setting turns it off.
- Zero data retention is requested on every message. If it cannot be met, the request stops.
- Documents are indexed by Mistral Embed; scanned pages are read by Mistral Small, a French model running in the EU. The index stays on our servers in Germany.
- Code the assistant writes runs in a sealed container on our servers, with no internet access.
- Every answer is labelled with what happened, including when it left the EU.
Only if you choose it
- Search words leave the EU only with Google web search, which you switch on and confirm.
- An Ephemeral chat is never stored: no messages, no title, no files. You cannot come back to it.