How the safety rules work
Every message your kid sends, and every answer the AI writes back, passes through your family's rules first. Both directions matter: an innocent question can get an answer that isn't.
The nine defaults
These are on from the first minute: violence, weapons, self-harm, adult content, drugs, alcohol, tobacco & vaping, gambling, and profanity monitoring. The first eight block; profanity doesn't block anything, it just lets you know if swearing becomes a pattern. You can turn any of them off, but most families never touch them.
Most defaults understand intent, not just keywords: a history question about Prohibition is fine, “how do I sneak beer” is not, even though both mention alcohol. A couple (like weapons) use direct word matching for speed.
What a rule can do
- Block.The message never reaches the AI (or the AI's answer never reaches your kid). They see an honest notice with your rule's message and your name, and you get an alert.
- Ask me first.The request pauses, your kid sees “waiting for your guardian,” and you approve or decline from your email in one tap. See Approvals.
- Allow and notify. Nothing is stopped, but the moment is recorded in your activity feed, either instantly or as a pattern alert (that's how profanity monitoring works).
Sensitivity, in parent language
Rules that understand intent give each match a confidence score, and every such rule has a threshold (75% by default). Above the threshold, the rule acts. Below it, nothing happens to your kid's chat, but you may see an amber “Possible” chip in the activity feed: the classifier's honest “I wasn't sure.” If a rule misses things it should catch, lower its threshold; if it keeps flagging video-game talk, it's working as intended and you can leave it alone.
Rules in your own words
On the Policies page you can add rules the way you'd say them out loud: block a specific game by name, make dating an “ask me first” topic, get notified about a topic without blocking it. Each custom rule picks one of the three actions above and its own notification style.
Teaching modes
Two optional re-writers change how the AI teaches, not what your kid can ask. With homework mode, the AI stops writing essays your kid could hand in and tutors instead: one guiding question at a time, hints when they're stuck, and a polite refusal of “just write the essay.” With vibe coding, code answers arrive as step-by-step walkthroughs instead of one big block. Honest framing: these adjust our instructions to the AI; your kid's own words are never altered or hidden.