Docs
Vanta catches scams, abuse and spam, tells banter from the real thing, warns before it punishes, and learns from your mods. Every command works as v!name or /vanta name.
Getting started
Add it and set it up in a minute.
Commands
Every command, with examples.
The 13 rules
What Vanta looks for.
Add Vanta to your server
Put its role above the roles it moderates
Pick a mod log channel: v!modlog #mod-log
Look over the settings: v!settings
v!test your message shows exactly what Vanta would do with a message, and punishes nobody.v!mode dry and it will only say what it would have done. Or try it in our test server first.View Channels, Send Messages, Send Messages in Threads, Embed Links, Read Message History, Manage Messages and Timeout Members. Kick Members and Ban Members only if you let it kick or ban. Vanta's settings panel lists anything it's missing under “Needs attention”.
Vanta isn't a word filter. Each of the 13 rules is a question in plain English, and every message is scored against all of them at once, as a chance from 0 to 100% that it breaks that rule. Before that, disguises are undone: leetspeak, look-alike letters, s p a c e d out words and slurs typed backwards.
Each rule has two lines:
| Score | What happens |
|---|---|
| Under the flag line | Nothing. Most messages score close to 0%. |
| Flag line (60%) | Reported to your mod log for a person to look at. The message stays up. |
| Punish line (85%) | The rule's punishment happens: delete, warn, time out and so on. |
Each rule can be relaxed (both lines higher, fewest false alarms), normal or strict (both lines lower, catches borderline messages). Change it with v!sensitivity slur strict or in v!settings.
When a score lands just under or between the lines, Vanta looks again at just those rules, this time with the last few messages in the channel and the message being replied to. “you're dead when I find you” after game chat is left alone; after a real argument, it isn't. The mod log shows how the scores moved.
Anyone with Manage Messages is exempt unless you run v!mods check. Explicit content isn't checked in age-restricted channels.
Each rule has one. Set it with v!set rule punishment or in v!settings.
| Punishment | What happens |
|---|---|
| Off | Not checked at all |
| Flag for mods | Leave it up, report it to the mod log |
| Delete | Delete the message, nothing else |
| Auto punish | Delete, then punish by severity and record |
| Delete + timeout | Delete and time the author out |
| Delete + kick | Delete and kick the author |
| Delete + ban | Delete and ban the author |
The default for the serious rules. The message is deleted, and the member gets a warning the first time. Repeat it and each offence climbs the ladder. Only serious messages start higher (1 hour), and extreme ones, like a real threat, start at a week.
Dashed steps are above the default ceiling: Vanta never kicks or bans by itself unless you raise it with v!ceiling.
Every offence adds to a member's record, and its weight halves over time: every 7 days with v!memory short, 3 weeks with normal, 2 months with long. Nothing is deleted; old offences simply stop counting, so someone who behaves gets a clean slate.
Every one can be switched off or given a different punishment. These are the defaults.
| Rule | What it catches | Default |
|---|---|---|
| Slur | Slurs against a group, including leetspeak, reversed, abbreviated and sound-alike spellings. Ordinary swearing is ignored. | Auto punish |
| Hate speech | Hatred of a group of people even without a slur: calling them subhuman, saying they deserve harm. | Auto punish |
| Severe abuse | Vicious personal attacks. Trash talk, mild insults and banter between friends are ignored. | Auto punish |
| Obfuscated text | Text scrambled to dodge a filter when the hidden word cannot be worked out. | Flag for mods |
| Threat of violence | Threats of real-world violence. In-game threats and obvious jokes between friends are ignored. | Auto punish |
| Self-harm risk | Someone who may be in crisis. A person needs to see this, so Vanta can never punish it.Only ever reported | Flag for mods |
| Doxxing | Posting someone else's private details: address, phone number, real name, workplace. | Delete |
| Explicit content | Graphic sexual content. Never checked in age-restricted channels.Skipped in NSFW channels | Delete |
| Sexual harassment | Unwanted sexual advances at another member: asking for nudes, sexual comments about them. | Auto punish |
| Sexualizing minors | Anything sexualizing under-18s, or grooming.Acts on weaker signals | Auto punish |
| Scams & phishing | Fake Nitro links, fake staff asking for logins, crypto schemes. Usually a hijacked account. | Auto punish |
| Spam & ads | Unsolicited server invites, shop and follower ads, attention-grabbing repeated text. | Delete |
| Illegal trade | Offers to buy or sell drugs, weapons, stolen accounts, card data or hacking services. | Delete |
Everything Vanta does to a member is saved as a case with an id that counts up in your server: V-c1, V-c2 and so on. The id is on the mod log report and under the notice the member sees, so “why was I muted?” can point at one case.
| Command | What it does |
|---|---|
v!cases | The ten latest cases in your server. |
v!cases @member | One member's record. |
v!case V-c12 | One case in full: the message, what happened and why. |
v!delcase V-c12 | Delete a case. It stops counting toward their next punishment straight away. |
v!reset @member | Delete a member's whole history, after asking. |
Your moderators' calls make Vanta better at your server specifically.
| Do this | And Vanta |
|---|---|
| Wrong call on a report | Undoes what it can (lifts the timeout, reverses a ban), stops the case counting against the member, and treats messages like it as fine in your server. |
| Right call on a flag | Treats messages like it as ones to act on in your server. |
| React 🚩 to a message it missed | Catches messages like it from then on. The 🚩 is taken back off so nobody sees who reported it, and the reporter gets a DM confirming it. |
The buttons are on every report in the mod log and need Manage Messages. v!examples lists everything your mods have taught Vanta, and v!delexample 3 removes one. Each server's lessons stay in that server.
Hacked accounts post pictures, “claim your free crypto”, a fake MrBeast giveaway, a casino bonus page, because text filters can't read them. Vanta reads the text in every image and asks the same model whether it's a crypto, casino or influencer scam.
A post like that almost always means the account was taken over, so Vanta deletes it, mutes the account for 3 days, and DMs the owner how to secure it: change their password and remove apps they don't recognise. It doesn't count against their record. When Vanta isn't sure, it only flags the image for your mods.
Each caught image's fingerprint is saved, so a repost is recognised instantly in any server running Vanta, without being read again. This follows your Scams & phishing rule: switch that off and images aren't checked.
Every command works as v!name or /vanta name. Words in <angle brackets> are required, [square brackets] optional, and a|b means one of them. Commands marked Manage Server are for your staff.
Everything Vanta can be set to do, on three tabs: the rules and their punishments, automatic punishments, and general switches like the mod log. Changes save the moment you click.
Also /vanta settings · v!config, v!panel, v!setup
Every punishment and every flag is reported here with the message, the reason and its case id. Pick a channel only your moderators can see.
v!modlog #mod-logv!modlogAlso /vanta modlog
Dry run is for trying Vanta out: it replies with what it would have done and reports it to the mod log, but deletes nothing and punishes nobody.
v!mode dryv!mode liveAlso /vanta mode
By default anyone with the Manage Messages permission is exempt. Choose check to hold moderators to the same rules.
v!mods checkv!mods exemptAlso /vanta mods
An ignored channel is skipped entirely: no rules, no flood limit, no pictures read. An ignored role does the same for every member who holds it. Useful for a venting channel, a bot-command channel, or a trusted staff role.
v!ignore #off-topicv!ignore @Moderatorv!ignore listv!ignore remove #off-topicAlso /vanta ignore
With a rule, puts that one rule back to its default. With nothing, resets every setting in this server, after asking. With a member, it leaves the settings alone and deletes every case that member has here instead, after asking, so their next offence counts as their first.
v!reset @member · Delete a member's whole history (asks first)
v!reset slurv!resetv!reset @someoneAlso /vanta reset
Sets the punishment for one category. off ignores it, flag only tells your mods, delete removes the message, auto warns first and then escalates on repeat offences, timeout mutes for a set length, kick and ban remove the member. A timeout takes a length: 10m, 2h, 1d, 1w (28d at most).
v!set slur autov!set threat timeout 1dv!set spam_ads deletev!set self_harm flagAlso /vanta set
Relaxed only acts when Vanta is very sure, so it has the fewest false alarms. Strict also catches borderline messages. Normal is the default. Less sure matches are still flagged to your mods.
v!sensitivity slur strictv!sensitivity spam_ads relaxedAlso /vanta sensitivity
When a moderator marks a report as a wrong call, the message is saved as fine for this server. A right call on a flagged message, or a 🚩 reaction from a moderator on a message Vanta missed, saves it as one to catch. The newest of each are shown to Vanta with every message, so it judges messages like them the same way.
Also /vanta examples · v!taught, v!learned
Takes the number shown by examples. Vanta stops using it straight away.
v!delexample 3Also /vanta delexample · v!deleteexample, v!forget
Rules set to auto warn first, then climb with each repeat offence: 10m, 1h, 1d, 1w, kick, ban. The ceiling is where that climb stops. The default is timeouts only.
v!ceiling timeoutv!ceiling kickAlso /vanta ceiling
Each offence makes a member's next automatic punishment harsher. Its weight halves every 7 days (short), 3 weeks (normal) or 2 months (long). Nothing is deleted: old cases simply stop counting.
v!memory longAlso /vanta memory
Every time Vanta deals with someone it saves a case with an id like V-c12: what was said, what happened and why. With a member, shows their latest cases and how many they have in total.
v!casesv!cases @someoneAlso /vanta cases · v!history, v!record
One case in full: the message, what happened and why
v!case V-c12Also /vanta case
Removes a case from the member's record. It stops counting toward their next automatic punishment straight away. Other case ids do not change.
v!delcase V-c12Also /vanta delcase · v!deletecase, v!removecase
Runs a message through the rules with your current settings and shows how close it came to each one. Nothing is recorded.
v!test you are so bad at this game lmaov!test free nitro, claim it at discord-gift.xyzAlso /vanta test
Every command, or one explained with examples
v!helpv!help setAlso /vanta help · v!commands
No. It judges what a message means, not which words are in it, and looks at the conversation when it's unsure. If it ever gets one wrong, a Wrong call on the report undoes it and teaches it.
Yes. v!mode dry makes it only describe what it would have done, and v!test shows its verdict on any message you type.
React 🚩 to the message. Vanta catches messages like it in your server from then on.
Its role is below theirs, they own the server, or it's missing Timeout Members. The mod log says which, and v!settings lists missing permissions.
Cases, your settings and what your mods taught it. Messages are checked when they're sent, not stored beyond the cases. The privacy policy has the details.
Anything else? Ask in our Discord, or read the privacy policy.