Automod and Safety

Automod is the fast rules engine. It handles objective patterns like spam, caps, links, blocked words, and repeated messages before the server gets noisy.

Core Rules

• Invite links: blocks Discord invites when Block invites is enabled.
• Scam links: catches common scam wording such as fake rewards, Nitro bait, or wallet prompts when Block scam links is on.
• Malicious links: catches file and malware-style links when Block malicious links is on.
• Mention spam: triggers when a message meets or exceeds the Mention spam limit.
• Emoji spam: triggers when a message meets or exceeds the Emoji spam limit.
• Excessive caps: checks the uppercase ratio after the message reaches Caps minimum letters.
• Repeated messages: counts matching messages inside the Duplicate window.
• Blocked words: checks your manual word list.
• Regex filters: checks your advanced patterns line by line.

What Each Setting Does

• Default action: the action AutoMod uses when a rule fires.
• Delete messages: removes the triggering message immediately when possible.
• Notify staff: logs the incident to the suspicious/logs flow.
• Trusted roles: members with these roles are skipped by AutoMod. This is the main bypass setting.
• Mention spam limit: how many mentions a single message can contain before it trips the rule.
• Emoji spam limit: how many emojis a single message can contain before it trips the rule.
• Caps threshold: the percent of letters that must be uppercase before the caps rule fires.
• Caps minimum letters: the minimum number of letters required before the caps rule even checks. This is what prevents short messages like THIS from getting flagged unless you intentionally lower it.
• Repeated message threshold: how many duplicate messages from the same user are needed before the repeat rule fires.
• Duplicate window: how long the engine keeps comparing recent messages for repeats.
• Rule cooldown: how long the same user is temporarily ignored for the same trigger after AutoMod already acted.

AI Moderation

AI Moderation is the softer scoring layer. It does not rely on one exact phrase. Instead, it assigns local heuristic scores to:

• Toxicity
• Harassment
• Threats
• Scams

Each threshold runs from 1 to 100.

• Lower values are more sensitive.
• Higher values are less sensitive.

AI Mode

• Assist moderators: produces signals that can also take automatic action if human approval is disabled.
• Shadow score only: records signals without taking action.

Require Human Approval

When this is enabled, AI Moderation only raises a review signal. It does not auto-punish.

Recommended Starting Point

• AutoMod action: delete
• Notify staff: on
• Caps threshold: 80
• Caps minimum letters: 20
• Repeated message threshold: 4
• Duplicate window: 20 seconds
• AI thresholds: toxicity 75, harassment 80, threats 85, scams 80

Tips

• If caps feels too aggressive, raise Caps minimum letters before raising the threshold itself.
• Keep blocked words precise. Very broad words create false positives.
• Start AI with Require human approval enabled until you trust the signal quality in your server.