Best Discord moderation bots

Discord moderation tools come in four shapes: Discord's own AutoMod, an all-in-one bot with a moderation module, a dedicated moderation or security bot and a bot like YourBot, where moderation is one service among several. The table sums each one up, and the sections below say what each one documents for moderation.

Everything here about a bot other than YourBot comes from that bot's own website and documentation, checked on the date at the foot of this page, not from hands-on testing. We name the features each bot documents for itself and note where its documentation marks a moderation feature as premium, without quoting prices or limits. Bots change, so check a bot's own site before you decide.

Discord AutoMod

Discord's own auto moderation, available on every server and set up in Server Settings with no bot to invite. Its rules cover custom keywords (regex included), three lists of commonly flagged words (insults and slurs, sexual content and severe profanity), spam, mention spam and words in member profile names. Depending on the rule, it blocks the message, sends an alert to a channel you pick, times the member out or blocks their interactions. Because it runs inside Discord it stops a message before it is posted, which a bot's own filter cannot do: a bot only sees a message once it exists. Discord also has Raid Protection under Safety Setup, which alerts a channel you choose when it detects a raid and asks new joiners to complete a CAPTCHA for the next hour, and Activity Alerts that flag unusual spikes in joins or DMs. It posts alerts and writes audit log entries, but its documentation describes no numbered cases or appeals. If you run YourBot, Sentinel creates and edits Discord's keyword, spam and mention spam rules from its own dashboard and can post what they block to your mod log, so the two work together. Best if you want a first layer that stops obvious spam before anyone reads it.

YourBot (Sentinel)

YourBot is one bot whose features are services, and moderation is one of them. Sentinel is included free on every server: an automod rules engine, raid detection that turns raid mode on by itself, anti-nuke that can undo the damage, numbered cases, member appeals and member reports. Verification by button, captcha or web page sits in YourBot's Welcome service. The same bot also runs tickets, events, analytics and voice, and an open marketplace lets independent developers publish more. Best if you would rather run one bot than several.

Dyno

A modular bot with a web dashboard, running since August 2016. Its Automod filters bad words, spam, mass mentions, links, invites, known phishing links and more, and can delete, warn, mute or ban. The Moderation module adds a moderation log, cases, warnings with automatic punishment after a set number of warns and an appeals form for unbans and unmutes. The Action Log records server events, and Autoban can ban new members by account age or by words in their usernames. Dyno's FAQ says it does not currently offer a feature that stops moderators or administrators from nuking the server. Most settings live in the web dashboard, and moderators work through slash commands. Its premium page lists Automod as free for everyone, and a few extras such as custom automod responses are documented as premium. Best if you want a long-established all-in-one with a web dashboard (compare).

Carl-bot

A modular all-in-one bot featuring reaction roles, automod, logging and custom commands. Its automod covers message, attachment and mention spam, links, invites, bad words, caps and honeypot channels, and each rule can combine punishments from delete and warn up to timeout, kick and ban. Moderators get commands for bans, softbans, kicks, timeouts and warns, a modlog with each member's infraction history and event logging that can be split across five channels. Setup is through its dashboard and slash commands. A drama channel, where mods review flagged messages, and a few other extras are documented as premium. Best if reaction roles and logging sit at the center of your setup (compare).

MEE6

One of the most widely used all-in-one bots, with leveling, welcome messages, moderation and social alerts for Twitch, YouTube, Reddit and more. Its Moderator plugin filters bad words, invites, external links, excessive emojis and all-caps messages and gives moderators slash commands to ban, kick and mute. It also posts audit logs of moderation, message, member, role, voice, server and channel events to a channel you choose. MEE6's help center says the Moderator plugin needs MEE6 Premium, and its plan comparison marks moderation as not included on the free plan. Setup is through its web dashboard. Best if you want a familiar, mainstream all-in-one (compare).

Wick

A moderation and security bot built around anti-nuke and anti-raid protection, with a web dashboard. Its anti-nuke, on by default according to its setup guide, watches staff activity such as bans, kicks and channel and role creation or deletion, and quarantines an account that hits one of the limits you set. It also documents a heat-based automod that is on by default, join raid detection, a join gate for suspicious accounts, verification by button, captcha or web page, lockdowns, panic modes, moderation cases, logging and a web page where members appeal bans and timeouts. Join raid detection, panic modes and backups are among the features documented as premium. It is set up from its dashboard, which has a setup wizard, or with commands. Its own documentation says it takes setup rather than working out of the box. Best if your main worry is an attack on the server itself, from raiders or from a compromised staff account.

YAGPDB

Yet Another General Purpose Discord Bot: an open source bot under the MIT license, with official instructions for hosting it yourself. For moderation it has a Basic Automoderator (slowmode, mass mentions, server invites, links, banned words and banned websites), an Advanced Automoderator built from rulesets of triggers, conditions and effects, moderation commands and a modlog of kicks, bans, mutes and warnings. Automod rules can mute, kick or ban after a set number of violations in a set time. Feeds for Reddit, YouTube, Twitch and RSS, reaction role menus and custom commands sit alongside. Best if you want to host the bot yourself or build detailed automod rules.

Zeppelin

A private moderation bot built for large servers. Zeppelin says access is granted case by case. It covers automod (word filters and spam detection), cases with moderator notes and customizable server logs. All of its configuration is a YAML file you edit in its web dashboard rather than a page of toggles, and its source is public on GitHub under the Elastic License 2.0. Best if you run a large server, can get access and want configuration you can read as text.

What Sentinel actually does

Rules, not one fixed filter. 22 condition types look at what a message contains (words and phrases with an anti-evasion normalizer, spam rate, a heat score, copypasta, mention floods, caps, emoji, newline and sticker spam, attachment floods, file types, links, fresh domains and known phishing domains) and at who sent it (account age, avatar, name filters, links in profile names, staff impersonation, author roles, risk score and risk band). Each rule fires any of 13 actions (delete, warn, timeout, kick, ban, quarantine, alert a channel, DM the member, auto-rename, server mute, server deafen, voice disconnect and voice move) on one of 3 triggers: a message, a join or a nickname change. Rules match on any or all of their conditions, run in priority order in the channels you pick and skip anyone holding your mod roles or a role the rule exempts. Any action can wait until the member has a set number of active warns, and a rule that times out, kicks, bans or quarantines waits out a per-member cooldown (10 seconds by default) before it acts on the same member again. A honeypot channel, if you set one, bans whoever posts in it by default, staff and bots excepted.

Watch before you enforce. Any filter can run in Testing: it touches no one and notes in the mod log what it would have done. Setting Spam & scams to Testing puts every filter in Testing at once. The filter editor has a Try it box: paste a sample message and Sentinel says whether it would trigger and which checks matched. 16 filter packs ship, and the How strict? levels (Relaxed, Standard and Strict) bundle 2, 5 or 9 of them. The setup wizard installs your level in Testing unless you switch on "Act right away", and each pack remembers the filters it created, so one button removes exactly those. "Check old messages" runs your message filters over up to 2,000 messages a channel across up to 25 channels (filters built on spam rate, heat, copypasta or attachment floods excepted), with Report only or Delete matches.

Phishing links from shared lists. Two public feeds of known phishing domains, together tens of thousands of entries, refresh every 6 hours, and each server can block or allow domains of its own on top. A separate condition flags links to recently registered domains, within 30 days by default.

Raids. Sentinel counts joins in a rolling one-minute window, and when joins reach your threshold (10 a minute by default) it switches into raid mode by itself. Raid mode has three levels. Soft puts a 10 second slowmode on every text channel. Hard makes it 30 seconds and quarantines every new join by default, and it is where automatic detection starts. Lockdown raises Discord's verification level to its highest, stops @everyone sending messages in text channels and kicks new joins. Sentinel also looks for look-alike accounts joining together (near-identical names, created within a day of each other, no avatar, account IDs minutes apart), names the invite they used and, by default, revokes an invite that brought a cluster. It records each channel's slowmode and Send Messages setting and the verification level before it changes them, puts them back when it stands down and stands down after 30 minutes by default. /lockdown locks one channel, the whole server, one role, new joins or invites, with an optional timer, and /unlockdown puts back what it changed.

Anti-nuke, for the night a staff account is hijacked. Anti-nuke is on from the first time a server turns Sentinel on, with quarantine, restore and the hold, and a server that had Sentinel before gets a one-click button for the same settings. It has twelve watches, each with its own count and time window: channel deletion and creation, role deletion and creation, bans, kicks, webhook creation, webhook changes or deletion, a bot being added, a role gaining dangerous permissions, a member gaining a dangerous role and a change to the server name, vanity link or owner. Mass damage gets the response and pings your mod roles: by default the account loses every role Sentinel can take and gets the Quarantined role, and the roles come back when it is released. A bot doing the damage is kicked instead. You can switch the response to stripping roles, kicking, banning or an alert. A single risky change by an admin, such as adding a bot or handing out a dangerous role, is undone and reported while the admin keeps their roles, and a bot added by anyone but the owner, an extra owner, an exempt admin or an account you list is kicked. Whatever the response, dangerous permission changes are reverted and the server name is changed back by default, and Sentinel tries to put the vanity link back. An ownership transfer is reported but cannot be undone. With a punishing response, Sentinel also unbans the people that account banned in a mass ban, and anyone other than the owner, an extra owner or an exempt account who hands the account its roles back or lifts its punishment becomes a new incident. An optional panic mode, which also needs a punishing response, locks the server down and leaves Sentinel's commands to the server owner, extra owners and the accounts you exempt. With restore on, deleted channels and roles come back from the newest backup taken before the attack, with their permission settings, and their members get the roles back, apart from roles with dangerous permissions. The server owner's own account is not watched unless a recovery code froze it.

Backups of everything but messages, free. Every hour Sentinel backs up the server's settings and images, roles, every channel and category with its permissions, emojis and stickers, and every six hours its bans and, on servers of up to 25,000 members, who holds which role. It keeps the newest 24 backups and one a day for a week, plus up to 10 you take yourself. You can download a backup or restore it from the dashboard, choosing what comes back, and a safety backup is taken first so a restore can itself be undone.

Owners who cannot be locked out. Up to 10 extra owners can change every Sentinel setting, and anti-nuke never acts on them. The Discord server owner makes one-time recovery codes, and using one from another Discord account in the server makes that account an owner on YourBot and freezes every other owner account until it is unfrozen.

Discord's own AutoMod, managed from the same place. Sentinel creates and edits Discord's native rules from its dashboard: keyword lists with regex and an allow list, Discord's preset lists for profanity, sexual content and slurs, mention spam and spam. Once your server has a mod log channel, Sentinel posts the blocks and timeouts those rules make there by default, and it can also record them as cases.

A risk score with its reasons attached. Each member gets a 0 to 100 score built from their case history, a manual watchlist entry, a current quarantine, pending reports, denied appeals and recent automod hits, with a per-factor breakdown. Old cases decay: full weight for 30 days, half to 90 days, a quarter to a year, then nothing. The score falls into five bands, and each band above clean has its own response: a quiet watchlist entry, a mod log notice, a ping to your mod roles and, if you opt in, quarantine. Warns stop counting toward escalation after 30 days by default, and escalating timeouts, which the setup wizard turns on, double an automod timeout for each earlier timeout the member had within 30 days, up to 28 days.

Appeals that undo the punishment. Members appeal a case in Discord with /appeal. Anyone kicked or banned can appeal from a web page instead after signing in with Discord, and punishment DMs carry the link. Approving an appeal marks the case overturned, schedules the matching reversal (untimeout, unban, unquarantine or voice unmute) and DMs the member, with your note if you add one.

Reports and cases keep the receipts. Members report with a right-click on a message, with /report or by reacting with a report emoji you choose, and every report gets a triage score. A message report carries a jump link to the message, /report attaches the channel's recent conversation and an emoji report carries an excerpt of the message. Warns, timeouts, kicks, bans, quarantines and voice actions each create a numbered case, and when an automod rule acts on a message and creates one, it saves up to 11 recent messages from the channel as evidence. Bans, kicks, timeouts, server mutes and server deafens that moderators hand out in Discord itself become cases too, and open cases are flagged as stale after seven days. Server logs for members, channels, roles, server settings, voice and content post who changed what, before and after, to channels you pick. A message log keeps deleted and edited messages for as many days as you set.

Bans that travel. Servers owned by the same account can share bans, and a ban network shares them between servers with different owners: a server joins with the network's code, the network owner approves it and each server picks its own response.

The moderator's day. Purge with filters that never touch pinned messages, softban, temporary roles, /mass to ban up to 200 members or kick or time out up to 100 at once, Timeout, Quarantine, Ban and View cases on a right-click, DM templates in your own wording, a watchlist and member history with notes. Optional settings add a confirm prompt on destructive commands, anonymous moderation that credits the team instead of the moderator and a default timeout length. More detail is on the Sentinel service page.

If your current moderation bot is working, keep it. The useful test is whether you can answer three things about last week: who was actioned, why and what came of the appeal. Sentinel costs nothing extra to switch on and its filters can start in Testing, so it can sit beside what you already have while you compare. See pricing for what Premium adds.