AI Moderation
Catches harassment and obfuscated abuse that word lists miss, using CyndrixAI.
What it does
Word lists catch words. AI Moderation catches intent — harassment written politely, abuse spelled creatively, or a pile-on where no single message contains anything on a banned list. It scores each message across six categories: toxicity, harassment, hate, sexual content, violence and self-harm.
Each category has its own threshold, so you can be strict about hate and relaxed about mild toxicity in a gaming server. Short messages are skipped below a minimum length, because "lol ok" carries no signal and scoring it just burns Action Points.
Start in test mode
Test mode is the single most useful setting here. It scores messages and logs what it would have done without acting on anything. Run it for a few days on a busy channel, read the log, then tune your thresholds before letting it act. Turning AI moderation loose at default thresholds on day one is how you end up removing jokes.
Scope it to specific channels or categories, and add bypass roles for staff. It draws from the same shared Action Points pool as the other AI features.
Commands
This module has no commands — it's configured entirely from the dashboard.