AI Moderation: Extra Checks in Released Channels

The AI moderation checks messages in channels you explicitly release, in addition to the word filter, for toxic, insulting or hateful content. On a hit, Kage reports it in the mod log channel and only deletes the message if you have switched that on.

What applies to all AI features (switch, daily budget, data privacy) is in AI features. Note there in particular which data goes to the AI provider: the text of the checked messages is transmitted.

Requirements

• Access to the module: You sign in to the dashboard with Discord and need the Manage Server permission (or access granted for this module).
• AI features active under AI Controls, and an AI provider set up by the operator.
• A mod log channel in the Moderation module, and the module must be switched on. Otherwise the AI detects hits but reports nothing. Kage needs Send Messages and Embed Links there.
• For deleting: Kage needs Manage Messages in the checked channel.
• Plan: The AI features are an add-on product, the plan notice at the top shows the status.

Setting up

1. In the dashboard, open AI moderation.
2. Under Allowed channels, choose the channels where checking should take place. By default no channel is released. At least one is needed to activate.
3. Under Automatically delete flagged messages, decide whether Kage deletes hits. Default: only log.
4. Turn on AI moderation active and save.

What is checked

Kage only checks a message if everything applies:

• The sender is not a bot.
• The sender does not have the Manage Messages permission. Moderators and admins are never checked.
• The message has at least 5 characters. Shorter ones like "ok" never go to the AI.
• The channel is among the allowed channels.
• The daily budget still suffices.

Then the AI gets the text (shortened to 500 characters) and answers only with OK or TOXIC and a short reason. Anything Kage does not clearly recognize as TOXIC counts as no hit.

The check costs one call per message. Choose the channels carefully, then. If the budget is used up or the AI is unreachable, the message is left untouched.

What happens on a hit

1. If Automatically delete is on, Kage deletes the message.
2. Kage posts a card "AI moderation triggered" in the mod log channel with member, channel, action ("Deleted" or "Logged only"), the AI's reason and the message text.
3. The event appears in the module's Logs tab.

There is no automatic penalty against the member and no entry in the warnings. You decide yourself whether to act.

Deleting is risky: The AI can be wrong. A wrongly flagged message is an unnecessary entry when only logging, but a lost member message when deleting. That is why deleting is off by default.

Settings at a glance

Checking the result

1. Release a channel: Choose a test channel and save.
2. Test message: With an account without moderation permissions, write a clearly insulting message.
3. Mod log: The card appears in the mod log channel. With deleting switched off, the message stays.

Troubleshooting

Nothing happens

Check: Are the main switch and AI moderation active on? Is the channel released? Does the account lack Manage Messages? Is the message long enough? Is the daily budget not used up?

Hits do not appear in the mod log

No mod log channel is chosen in the Moderation module, the module is switched off, or Kage may not write there.

The message was not deleted despite a hit

Automatically delete is off, or Kage does not have Manage Messages in the channel.

Too many false alarms

Leave deleting off and limit the release to a few channels. Add the word filter for clear-cut terms.

The budget is used up quickly

Each checked message costs one call. Release fewer or less active channels.

Further reading

Classic moderation: Moderation, Word filter, Anti-spam.