All featuresAI Spam Protection

AI Spam Protection

Catches unsolicited bulk messages by what they say, in any language — not by how fast they arrive.

Overview

Rate limits catch floods. They cannot catch the message that arrives once, at conversational speed, and is still spam — the unsolicited pitch, the copy-pasted offer, the same template a dozen servers received this week. AI Spam Protection reads what a message says instead of counting how many arrived, so it sees that message. It runs on the same analysis as Ad Protection but is its own switch, because wanting advertising policed and wanting this policed are two different decisions.

How it works

Read, then judge

The message text, any images it carries, the links it resolves to and the channel's own recent conversation are read together. A verdict is only reached on concrete evidence in the message, never on a member being new or unfamiliar.

Language-independent

A multilingual model embeds the message in a shared space, so the same pitch in Thai, Turkish or Russian matches the template already learned in English. Nothing here is a keyword list.

Learns from your own moderators

A block your team reverses is remembered. Sylon stops matching that message and anything close to it in your server, and near-identical spam it did catch reinforces the pattern for next time.

Key capabilities

Content-based detection, independent of message rate

Works in every language, including scripts with no word spacing

Reads text inside images and resolves shortened links

Its own punishment, threshold and log channel

Moderator reversals train the detector for your server

Exempt roles, members and channels

Setup

1.Open the dashboard, go to the AI Spam Protection tab and enable it.

2.Choose how many detections a member may accumulate before the punishment applies.

3.Pick the punishment — a timeout is the safest place to start — and set its duration.

4.Set a log channel, and exempt any roles, members or channels that legitimately post promotional content.

How it works in detail

Why content and rate are different questions

A member who posts five messages in three seconds is doing something a counter can see. A member who posts one message offering to sell accounts is doing something only a reader can see. Sylon has always had the counter; this is the reader, and separating them means you can have either without the other.

Evidence, not suspicion

A verdict requires something concrete in the message: an external link, an invite to another server, a QR code, promotional text read out of an image, or a recognisable pitch template. A new account, a low trust score or an unfamiliar name is never evidence on its own, and Sylon refuses to act on those alone.

The channel decides what is off-topic

Sylon reads what the channel is for and what its members are actually saying before judging whether a message fits. A room where people openly look for each other is a room where looking for someone is on-topic, whatever language its name is written in.

Every decision is reversible

Blocks are logged with the reason, the confidence and the evidence behind them. Your moderators can undo one in a click, and that undo is the strongest signal Sylon has: it stops matching that content in your server from then on.

Settings

Violations before punishmentHow many detections a member may accumulate before the punishment is applied.

PunishmentThe action taken against the user when the limit is reached. Timeout temporarily restricts, Kick removes from server, Ban permanently bans.

Timeout Duration (min)How long a user will be timed out when caught spamming. Only applies when Timeout is selected.

Log ChannelThe channel where the bot will send log messages for this feature.

Also apply to botsOff by default: bot messages are ignored entirely. Turn it on when a raid is being run through bots — the usual way around anti-spam, which only counts people. Sylon never judges itself, and a bot listed under exempt roles or exempt users is still exempt. Note that Administrator does not exempt a bot here the way it exempts a person: name the bots you trust in the exempt lists.

Exempt RolesMembers holding any of these roles will be exempt from this protection.

Exempt MembersSpecific members this system completely ignores, in addition to the exempt roles.

Exempt ChannelsMessages in these channels will not be checked by this protection.

Good to know

Sylon needs Manage Messages to delete, and Moderate Members, Kick Members or Ban Members depending on the punishment you choose.

This shares its analysis with Ad Protection, so enabling both costs no extra processing.

Exempt roles, members and channels are checked before anything is analysed, so exempted traffic is never judged at all.

Frequently asked questions

How is this different from Anti-Spam?

Anti-Spam is a rate limit: so many messages in so many seconds. It never reads a word. AI Spam Protection reads the message and can flag a single one. Running both is normal; running either alone is a valid choice.

Do I need Ad Protection enabled too?

No. They are separate switches on purpose. Ad Protection handles advertisements, scams and recruitment; this handles unsolicited bulk content. Enabling one never enables the other.

What happens if it gets one wrong?

Every block carries an undo button and a false-positive report. Undoing it restores the member and teaches Sylon not to make that call again in your server. Starting with a timeout rather than a ban is the safer setting while you are judging the results.

Related features

AI Advertisement Protection

Spam Protection

Harassment Protection