Conversational safety

Version 1.0 — Effective date: 2026-07-25

This is a courtesy translation. The French version in French HTML version is the official version and prevails in case of any discrepancy.

This protocol applies to conversations held in the Savantis application. It complements each agent's specific notices.

1. Nature of the service

Savantis agents are artificial intelligence software, not people. They do not replace emergency services, health professionals, or crisis intervention services.

2. Crisis messages

A deterministic filter examines the message locally when it is sent to recognize explicit wording related to suicide, self-harm, or immediate danger.

When it is triggered:

• the message is not sent to the AI model;
• the usual conversational response is interrupted;
• the application displays Canadian crisis resources and reminds the user to call emergency services in case of immediate danger;
• only a technical trigger event is logged.

The message text, conversation, and user identity are not copied to this safety log.

The filter can be mistaken or miss some wording. It is not a clinical assessment.

3. Long sessions

During a continuous conversation, the application periodically displays a non-blocking reminder to take a break and that the agent is artificial intelligence. The interval can be changed in settings.

The log records only that the reminder was displayed, without conversation content.

4. Adult users

Savantis is reserved for people aged 18 or older. A separate, timestamped confirmation is required before account activation. No date of birth is requested.

5. Contact

To report a safety concern related to this protocol: support@savantisai.com.