Security
Three simple rules: we only use the necessary information, only authorised personnel have access to it, and sensitive actions require approval from someone on your team.
002
Multiple safeguards
We apply multiple layers of protection: the information used, who has access to it, what the AI is permitted to do, where human validation is required, and what is logged. Security never relies on a simple prompt given to the AI.
003
Planned from the start
The essentials
We only use the information necessary for the work entrusted to us.
Limited access
Each person, assistant, or connected tool only has the permissions necessary for their role.
Your approval when needed
Sensitive actions await human approval before being sent. We determine what counts as sensitive together with you.
Everything is searchable
For every major action, you know where it came from, its context, and who authorized it.
004
Who can do what?
Nothing is sent automatically by default. Together, we assign each action to one of these three levels.
Automated
Repetitive tasks that can be eliminated.
Categorize, summarize, draft
On your terms
The action only proceeds if your rules are met.
Rule met → the action runs; otherwise, your team is notified
Your agreement
Sensitive decisions and edge cases.
Everything is prepared, a person makes the decision
005
Data sovereignty
We leave you with maximum control over it.
Hosted in Europe
Your data is hosted on servers located in the European Union.
GDPR Compliance
A data processing agreement is signed with each client.
Never used to train AI
Your data is never used to train artificial intelligence models.
You retain full ownership
Your data and automations belong to you: you can retrieve or delete everything at any time.
On-site, if you prefer
The tools can be installed on your own servers.
Your data, at every step
Everything is set up before any data is used.
Receive
For specific use cases
Use
Only what is useful
Keep
For a duration agreed upon with you
Delete
Planned from the start
Received messages
Incoming messages never issue instructions to the AI.
Incoming emails, messages, and documents are treated as information, never as instructions. They cannot alter the permissions or rules established with you.
Message received
"Could you please confirm Thursday's appointment?"
Reading
Read as information, not as an instruction
Your rules
Unchanged permissions and rules
Authorized response
Response prepared according to your guidelines
In technical terms: protection against prompt injection.