AutoMod

Server AssistantApponly you can see this

AutoMod AI second-opinion

Borderline word-filter and spam hits get a quick AI confidence check. Hits at or above your threshold act automatically; anything below is left up and posted to staff for a one-tap decision. High-confidence hits (links, mention spam) always act immediately.

Status

Off · needs Premium

Auto-action threshold

70/100

Premium. Word-filter and spam hits are checked; word-filter checks are free.

Toggle on/off Set threshold ← Back

AI Review PREMIUM. A second opinion on filter hits. Every word-filter and spam match is sent to the AI for a confidence score: at or above your threshold it acts automatically; below it, the message stays up and staff get a Delete & warn / Allow prompt. The filter cannot rank its own certainty, so the AI is what decides whether a hit is clear-cut. Link and mention hits are never sent. Word-filter checks run on a dedicated free moderation endpoint and cost you nothing; spam checks use your AI allowance.

Server AssistantApponly you can see this

AI auto-action threshold

Hits at or above this score act automatically.

Threshold (0 to 100)

70

Save

The threshold form. Enter a number from 0 to 100. Higher means the AI must be more confident before acting on its own; anything below the line goes to staff to decide.

Server AssistantApponly you can see this

Self-trained AutoMod

The bot quietly learns from what your staff delete or flag, then proposes server-specific rules for your approval. Nothing is ever auto-enforced, you accept or reject each one.

Status

Learning

Samples collected

34 / 50 needed

Only message text is stored, locally, for 30 days. No author identities.

Toggle learning on/off Suggest rules now

Clear learning log ← Back

Self-training. With learning on, the bot remembers the content of messages your team deletes or flags via Message Report, clusters the patterns, and (when you press Suggest rules now) proposes custom filters for your server. Only message text is kept, locally, for 30 days, with no author identities; nothing is enforced until you accept it. Clear learning log wipes the samples.

Server AssistantApponly you can see this

Proposed rule 1 of 3

Your staff keep removing repeated invite-link spam. This rule would catch it automatically.

Regex pattern

(discord\.gg|discord\.com/invite)/\w+

Would have caught

• “join my server discord.gg/xyz”
• “discord.com/invite/abc free stuff”

Accept to add it to your custom filters, or Skip. Nothing is enforced until you accept.

Accept ⏭Skip

Reviewing proposals. Suggest rules now walks each proposed rule one at a time: the reason, the pattern, and real examples it would have caught. Accept adds it to your custom filters (live immediately); ⏭Skip discards it. You get a tally at the end.

Server AssistantApponly you can see this

Custom Filters

Your own patterns, on top of the curated packs.

2 custom filter(s)

\bbadword\b

spam\s*phrase

Select filters to remove…

Add filter ← Back

Custom Filters. Your own word/phrase patterns, on top of the packs. Add filter opens a form for a regex pattern (up to 200 characters); the bot rejects invalid ones, or ones that could hang on malicious input, with an explanation. The dropdown removes entries.

Server AssistantApponly you can see this

Add custom regex filter

Messages matching this pattern get actioned.

Regex pattern

e.g. \bbadword\b or spam\s*phrase

Save

The custom filter form. Enter one pattern. If it's invalid or could be slow on crafted input, the bot explains why and asks you to rewrite it, so a stray pattern can never wedge moderation.

Server AssistantApponly you can see this

Custom Link Blocks

Domains to block on top of the built-in lists.

1 custom blocked link(s)

badsite.com

Select links to remove…

Add link block ← Back

Link Blocks. Domains you never want posted, on top of the built-in IP-grabber and scam lists. Add link block takes a domain (e.g. badsite.com); paste a full URL and the bot strips it to the domain. The dropdown removes entries.

Server AssistantApponly you can see this

Punishment ladder

One step per line: count:action[:minutes]

Steps

3:timeout:10 5:timeout:60 10:ban

Save

Punishment Ladder. What happens as a member racks up warnings. Each line is count:action[:minutes]: e.g. 3:timeout:10 (10-min timeout at 3 warnings), then 5:timeout:60, then 10:ban. Actions are timeout, kick, ban or warn; minutes are optional. The default is the ladder shown.

Server AssistantApponly you can see this

Spam thresholds

Mention limits and repeat-message detection.

Max mentions per message

6

Anti-spam (yes/no), repeat-message detection

yes

Save

Spam Settings. Two controls: max mentions per message (over the limit counts as mention spam), and an anti-spam toggle (yes/no) that catches the same message repeated over and over.