DeploySovereignHow we prove that a US AI never sees your documents
Using AI to analyse your documents without their content leaving Europe: the hybrid architecture, and above all how we PROVE it.
Analysing a document with AI is useful. But if the model is American, the raw content of your files crosses the Atlantic. Our answer: a European analysis layer (Mistral), and only a sanitised summary continues the pipeline.
The principle in one image
contrat, liste clients, dossier… tel quel
lit le document et en tire un résumé structuré
structure et chiffres — ni noms, ni textes bruts
la construction continue avec ce résumé, rien de plus
Raw text never leaves Europe. What circulates afterwards is a business abstraction, not your sensitive data.
"OK, but how do I PROVE it?"
That's the right question. Four levels of proof, from simplest to strongest:
- Code audit: the path is short and readable — raw data goes only to the European AI, only the summary is stored.
- Marker test: we slip a unique marker (e.g. "XJ-MARKER-4271") into a document, then inspect what actually goes out to the third-party service. The marker must be absent.
- Network traffic capture: we show that raw data goes only to the EU.
- Permanent logging: we keep an exact record of what is sent — auditable at any time.
The honest nuance
The summary itself may use a third-party service. We make it defensible with a personal data filter on that summary before any transmission. And for the strictest requirements (healthcare, public sector), we can switch to an end-to-end European AI, hosted in your own cloud.
Transparency is not a slogan: it is a protocol we can show you.
A sovereign application, delivered and hosted for you.
From idea to production, on a European cloud.
See pricing