How moderation works

Watching starts the moment you hire it by picking its name, not at install. From then on it watches your members' messages in every channel it can see, except its own private channel.

It judges text and images. To keep it out of a channel, remove its access with Discord's channel permissions.

• Voice chat isn't heard or judged.

A flag usually reaches you within a minute.

It judges against your rules, your server's culture and any calibrations you've taught it. Banter, trash talk and swearing that fit your community aren't violations. It also flags a few serious safety issues on every server, whatever your rules say, such as sexual content involving minors, credible threats of violence and doxxing.

It's built to avoid over-flagging, so when it's unsure it stays quiet. Messages it doesn't flag aren't stored anywhere. They only count toward the totals in your reports.

When something crosses a line, it posts a flag card in its private channel. Each card shows:

• The title: "Flagged" plus a severity of Low, Medium or High, so you can see at a glance how serious it is.
• Who posted it and in which channel, with the message quoted (images are listed by file name).
• Why: a one-line explanation of the problem.
• Rule: the rule it broke, in your wording. If no rule of yours covers it, this reads "Core safety flag (no specific server rule)".
• Repeat offender: shown only if the member has been flagged before, with their flag number and how earlier flags were handled (for example: flag #3, 1 warned, 0 deleted, 0 timed out, 1 dismissed).
• A footer reminding you that it's in observer mode: the call is yours, and it doesn't act unless you click.

Each card has five buttons. When someone clicks Warn, Delete message, Timeout (1h) or Not a problem, the action is recorded with who clicked it, the buttons disappear and the footer changes to the outcome, such as "Warned by" and their username.

• Warn: replies to the flagged message, mentioning the member: "heads up: that message isn't allowed here. Please keep it within the server rules." If the message is gone, it posts the warning in that channel instead.
• Delete message: deletes the flagged message. Needs the Manage Messages permission. Without it, only you see a note saying so, and the card stays open.
• Timeout (1h): times the member out for one hour. Needs Moderate Members. It can't time out the server owner, an admin, or anyone whose highest role is at or above its own. If it can't, only you see why, nothing is recorded and the card stays open.
• Not a problem: closes the flag with no action against the member. Your employee then offers to stop flagging that kind of thing, and only does so if you confirm.
• Jump: takes you to the original message.

Every flag and every action your team takes builds a member's record, and your employee takes relevant history into account when it judges.

Staff can check anyone's record with /history, and log a warning or a private note with /warn or /note. See Commands.

Calibrate it by telling it in its private channel, in plain words. For example: "OOC trash talk is fine here" or "stop flagging swearing between regulars". It proposes a calibration with Confirm and Cancel buttons, and once you confirm, it judges with that in mind from then on. A calibration can relax how it treats language on your server, but it can't switch off the safety checks above.

To change what's actually against the rules, edit the rulebook instead. See Your rules.

• It never warns, deletes, times out, kicks or bans without a human clicking first.
• It doesn't block messages before they're posted. It reads messages after they appear.
• It never calls out a member in public on its own. Its judgments go to you, privately.
• Kicks and bans aren't on the flag card. You can ask for one on a recently flagged member in its private channel, and it proposes it for you to confirm. See Talking to your employee.

Each server has a monthly reading capacity. As yours gets close, your employee gives you a heads-up in its private channel and keeps watching as normal.

At the limit, it tells you and stops reading new messages until capacity resets at the start of the next month. While paused, it doesn't flag anything or answer @mentions in public channels. Its private channel keeps working, so you can still talk to it and act on earlier flags. Nothing it has logged is lost.