Control

An artificial-intelligence system can make mistakes.

Avoiding them means identifying the sensitive steps at design time, and placing the controls there.

A person in the loop

We define together the moments when a person validates, and they are written into the scope before the work. The file comes back to them with what they need to decide: the invoice and the purchase order side by side, the quote ready to reread, the prepared message.

At each run, you receive an email with the result, and the history is in your dashboard. The system validates no entry, initiates no payment, and sends no quote.

Answers you can check

A model queried on its own invents. The system answers from your documents, names the file and the passage the answer is drawn from, and flags the questions left unanswered instead of reconstructing one.

Its scope is the collection you entrust to it; it does not search the Internet.

Capped costs

Every installation includes a gateway, and no call to a model bypasses it. You read the spending use case by use case in your dashboard, and you set the monthly ceiling there.

Once the ceiling is reached, calls are refused until you raise it.

Access defined one by one

The system's access to your tools is defined one by one, read-only or read and write, and an access the use case does not need is not requested.

The connection to your mailbox goes through a consent you give yourself. You grant it and withdraw it from the Connections page of your dashboard. Access tokens are encrypted, and they appear neither in your browser nor in the conversation.

Where the model runs, and the sharing between clients

The model runs where you decide. On our hardware in Lausanne, we can tell you on which machine. Each client has its own installation, its own network and its own volumes, and its documents are not indexed with another's.

Telephony, when it is installed, goes through a third-party provider. We check the software in the catalogue on our own machines before installing it.

Swiss data protection act, GDPR, AI Act: the law that applies to you

In Switzerland, the data protection act applies, as to any data processing. A company that processes data of people based in the Union also falls under the GDPR.

The European AI Act has applied since 2 August 2026 to any company that places a system on the Union market, or whose results are used there: it follows your activity in France, not your registered office in Switzerland. Sorting invoices or answering from one's own documents does not fall within its high-risk categories.

What going live requires to be compliant is included in its price: the description of the system, the written limits of what it may read and write, and the log of its runs.

Three questions to ask any provider

  • Where does the model run, and on which machine?
  • What is shared between clients?
  • Which software has been checked, and by whom?

Get in touch

Thirty minutes, by video call. We look at your documents and say what we can do, including when the answer is no.

Book a call

In writing. A person reads your message, and replies within 24 hours.

bonjour@advazia.ch +41 78 305 23 12

Avenue Édouard-Dapples 21
1006 Lausanne, Switzerland