AI & Safety

How the AI companion works

The companion is an AI model operated through Anthropic's API. Before and after every reply, Nightingain applies fixed safety rules written and reviewable by people:

• Urgent help is never gated. If a message contains language suggesting possible danger, crisis resources appear immediately — even if you've used your monthly AI messages, AI is switched off, or you're on the free plan.
• Fixed answers for things an AI shouldn't do. Questions about diagnosis, medication or dosing, or whether you're safe to drive or work get a reviewed response pointing to qualified help, without calling the AI.
• Output checks. Replies are checked for claims we don't allow (diagnosis, dosing, "you're safe to drive", claiming to be human or a clinician, promising to check on you). If one appears, a safe fallback is shown instead.
• No escalation behind your back. Nightingain never contacts your employer, emergency services, or anyone else based on what you write.

Limits

These safeguards can't recognise every crisis, and the AI can be wrong. It isn't therapy and doesn't replace people who can help. The safety rules and the companion's instructions are pending review by a licensed mental-health professional.

Costs and limits

Each plan includes a monthly number of AI messages. Nightingain also enforces an overall monthly AI budget; if it's reached, the companion pauses and everything else keeps working.