In pictures
Before you begin
The permission to manage automoderation in Spider (rules.manage).
Step by step
Open Moderation › Smart automod and turn the module on.
Choose an alerts channel for the staff only and the sensitivity.
Keep the types to watch, add the channels where members tease each other to the ignored channels, then save.
Try a few sentences in Test a message to see what Spider thinks.
How it works
A first filter, on Spider’s server, picks the suspicious messages, even disguised: digits for letters, spaced or repeated letters. Only those messages are sent to the AI, with the six messages before them. The AI decides whether it is a real attack or a joke between friends, and gives a severity from 1 to 3. Other messages never leave Spider.
Alerts
Each alert arrives in the chosen channel with the message, the member, the type, the severity and the AI’s explanation. The staff picks: Delete, Warn, Timeout 10 min or Reject. Warn and Timeout create an infraction in the member’s case and remove the message. Nothing happens without a click, unless Delete very serious messages is on: threats and hate of severity 3 are then removed at once.
Limits and privacy
Staff who can manage messages and bots are never analysed. Spider analyses at most 120 suspicious messages per hour and per server, and 6 per member in 10 minutes. The AI can be wrong: rejected alerts lower the rate of confirmed alerts shown at the bottom of the page. Spider keeps the last 200 alerts. The module needs access to message content.
The team answers on the Spider server. Ask on Discord