Safety & trust

Coach Neil runs under a fixed system prompt that never breaks: no medical, clinical, legal or mental-health advice; no collecting personal details; nothing frightening or adult. It stays focused on tutoring, encouragement and feelings support.

Every message your child sends is checked by a deterministic keyword safety net AND a separate AI classifier. The final read is always the MORE cautious of the two — the AI can only raise the alarm level the keyword net already set, never lower it.

If a message signals real distress, Coach Neil replies with a fixed, gentle, pre-written message (no live model call — safety never depends on the AI provider being up) and a flag is raised straight to your parent panel. For higher-severity flags, we also try to alert you directly by email when one is configured for your account — this is best-effort, not guaranteed, so the panel is the reliable place to check.

Nothing your child says to Coach Neil is private from you. Full chat history is visible in your parent panel, any time.

The parent panel — progress, ratings, writing history, wellbeing flags — sits behind a PIN only you set. Your child signs in to their own account and cannot open the panel, even if they know the PIN is there.

Only a parent account carries an email address. Child accounts are created by a parent from the panel with just a display name, a username and a password — your child never has, needs, or is shown an email address of their own.

Every child has a daily cap on AI usage that a parent controls. If a cap is ever crossed, further AI-backed replies pause for a rolling 24 hours — this limit never touches the safety path above, which always runs regardless of budget.

Coach Neil's own replies are moderated by an automated content filter before they're stored or shown to your child — this is an automated check, not a person reading chats in real time.