GRIMASA
DownloadLog in
DOCSSafety & moderation

On-air guardrails.

Stop phrases, topic blocklists, rate limits, and audit logs keep your stream safe even when chat tries to push the AI off-script.

Stop phrases

Per-character list of phrases the host will refuse to say. If chat tries to coerce the AI into saying one, the host pivots to a generic on-topic response.

  • Set up to 50 stop phrases per character
  • Phrases are case-insensitive substring matches
  • Update them while streaming — changes take effect on the next response

Topic blocklists

Block entire topics, not just phrases. Useful for keeping your gaming stream from drifting into politics or any space your audience isn't there for.

Rate limiting

Per-user and per-event throttles. A spam bot in chat can't drain your AI quota — once a user hits the per-minute cap, the host stops responding to them until the window resets.

Audit logs

Every AI response, event trigger, and scene switch is logged. After the stream you can review the timeline, find the moment something went wrong, and tune your rules accordingly.

Emergency stop

Hardware push-to-stop button (default F9) cuts the AI mid-sentence. Use it if something inappropriate makes it through and you want full silence while you assess.

Live AI moderation isn't perfect. Treat the host as an amateur improv partner who occasionally says unexpected things — be ready to take over.