Guide/Protect your server/AI judgments & privacy
AI judgments & privacy — Discord Bot Guide
Where the model is consulted, what it can and cannot decide, how to switch each use off, and what leaves the server.
Quick answer
Turn a use off
Dashboard → Configuration → AI judgments → switch off "automod" (or lower its confidence bar)Ask the guide
Step by step
In short
The model never acts on its own. It answers narrow yes/no or which-one questions that code then uses; every action still goes through the same thresholds, permissions and hierarchy checks. Configuration → AI judgments switches each use on or off.
| Use | Question the model answers | What happens with the answer |
|---|---|---|
automod | Is this borderline message a scam, threat, disguised slur, harassment, advertising or NSFW? | Adds heat (aiScam, aiThreat, aiHarassment, aiNsfw, aiAdvertising factors). Heat thresholds decide, not the model. |
secondOpinion | Do these messages really deserve the automatic timeout / kick / ban / quarantine about to land? | A positive disagreement HOLDS the action and alerts staff. Silence or uncertainty changes nothing. |
compromise | Did this long-standing member stop sounding like themselves while posting a pitch? | Critical alert + heat. Never a ban on its own. |
privileged | Does this destructive admin action read as a takeover rather than housekeeping? | Anti-nuke treats one such action as its threshold; whitelisted staff only get an alert. |
antiraid | Does this joining profile (name, age, avatar) look like a raid account? | Points added to the join risk score. |
cohort | Is this group of joins one operator, or real people arriving together? | Contain the cohort, or stand the rate response down for a genuine influx. |
names | Is this nickname or voice-room name a slur, sexual, impersonation or an invite? | Name reset. |
autoresponders | Which of the server's written triggers does this message mean? | Match by meaning instead of literal text. |
routing | Which team is this ticket for; which command does this plain-English request want? | Triage note; assistant plan shown for confirmation. |
answers | Which of the server's written answers covers this question? | Selects one of yours verbatim, or nothing. |
appeals | How does this appeal read? | A hint for staff. Never decides. |
Confidence
Each use has its own bar under Configuration → AI judgments. Adding heat can end in a ban, so its bar is high; picking an answer is harmless, so its bar is low. Below the bar the answer is ignored and the rule-based path stands alone.
What leaves the server
- Only the text being judged plus the minimum context (a username, an account age in days, the names of a few roles). No message history is uploaded, no member lists, no server id.
- Nothing is stored by the model; verdicts are cached briefly by the bot so the same message is judged once.
- If the model is unreachable, every protection keeps working on rules alone — an outage can never make the bot softer.
- Switching the master toggle off stops every call for this server immediately. The operator can also revoke the key globally.
Status (calls, cache hits, failures) is shown live at the top of the AI page; the key itself is never displayed anywhere.
Copy and try
Turn a use off
Dashboard → Configuration → AI judgments → switch off "automod" (or lower its confidence bar)Good to know
- Without a key every protection runs on rules alone; nothing depends on the model.
- Only the judged text and minimal context leave the server; no history or member lists.
Still stuck? Ask in the support server.