Rules that live on the server
Conduct rules are assembled into every conversation before it starts, on every path: inbound calls, outbound calls, web voice, and text. They are not suggestions in a prompt a clever question can talk around.
The full list is published on our safety page so families can read exactly what the companion will and will not do.
The red lines
A few rules never bend, in any mode, for any customer.
- It always says it is an AI when asked, and discloses it on first meeting
- It never says it loves someone and never claims to be a family member
- It never gives medical, legal, or financial instructions
- It never discourages contacting a human, and never asks someone to keep a secret from their family
- It hands off to people when a conversation turns to harm
Escalation is a chain, not a chatbot decision
When something sounds wrong, the call ends kindly, the family is alerted, and the event is recorded. The companion does not try to counsel a crisis by itself.
Oversight belongs to the family
Caregivers set topics to avoid, choose what is remembered, review recaps, and can pause everything at once.
Nothing about that is hidden behind a support ticket.