> ## Documentation Index
> Fetch the complete documentation index at: https://docs.vetox.io/llms.txt
> Use this file to discover all available pages before exploring further.

# AI Auto-Mod

> AI moderation that reads meaning rather than matching keywords — ten categories, escalating punishment and trust-based sampling. Black tier.

<Info>
  **Black tier only.** On any lower tier the page is replaced by the plans view. Custom Bot (`+`) does not grant AI access — it is tied to the tier alone.
</Info>

Standard [Auto-Moderation](/en/moderation/auto-mod) matches patterns. AI Auto-Moderation reads the message the way a moderator would: it understands sarcasm, coded language, misspellings meant to dodge filters, and content in any language your members write in. It also reads images.

<Note>
  Two switches must both be on: **AI Auto-Mod** in [Server Setup](/en/server-management/setup), and the settings on this page. With the module off, nothing scans and no warning appears here beyond an amber banner.
</Note>

## Start in Monitor mode

The **Operating mode** setting is the most important control on the page.

<CardGroup cols={2}>
  <Card title="Monitor" icon="eye">
    Violations are logged. **Nothing is acted on.** This is the default.
  </Card>

  <Card title="Enforce" icon="gavel">
    Configured actions are applied automatically.
  </Card>
</CardGroup>

<Warning>
  Monitor mode still costs tokens — every message is scanned, only the action is withheld. If you leave Monitor on without a log channel set, you pay for scans whose results nobody sees.
</Warning>

<Tip>
  Run Monitor for a few days with a log channel, read what it flagged, tune your sensitivity, then switch to Enforce. Skipping this step is the most common cause of complaints about false positives.
</Tip>

## Sensitivity

One slider, expressed as a percentage.

| Higher sensitivity         | Lower sensitivity                       |
| -------------------------- | --------------------------------------- |
| Acts on lower AI certainty | Acts only when the AI is very confident |
| Catches more               | Catches less                            |
| More false positives       | Fewer false positives                   |

A balanced starting point is **25–40%**.

<Note>
  Each category has its own sensitivity slider. **When a category's slider is set, it is the only one that applies** — the global value is a fallback for categories you have not tuned, not an additional floor.
</Note>

## Categories

Ten categories, each with its own switch, its own set of actions, and its own sensitivity.

| Category                            | On by default | Default action   |
| ----------------------------------- | ------------- | ---------------- |
| Harassment                          | —             | Delete           |
| Hate speech                         | —             | Delete           |
| Sexual content                      | —             | Delete           |
| **Sexual content involving minors** | **Yes**       | **Delete + Ban** |
| Violence and threats                | —             | Delete           |
| Self-harm                           | —             | Delete           |
| Illicit activity                    | —             | Delete           |
| Scams and phishing                  | —             | Delete           |
| Profanity                           | —             | Delete           |
| Advertising                         | —             | Delete           |

<Warning>
  Only **Sexual content involving minors** is enabled out of the box. Every other category is off until you turn it on — a freshly enabled AI Auto-Mod does almost nothing until you configure it.
</Warning>

<Warning>
  The sexual-content-involving-minors category is technically editable, but disabling it may breach Discord's Terms of Service. Leave it on.
</Warning>

### Age-gated channels

In a Discord age-restricted channel, the **Sexual content** category is suppressed. Every other category — including sexual content involving minors — still enforces normally.

### Actions

Pick any combination per category: Delete message, Warn user, Timeout, Mute, Kick, Ban, Alert.

Three combinations are refused:

* Kick **and** Ban together
* Alert **with** Ban
* Alert **with** Kick

Everything else stacks. Delete combines with anything.

<Note>
  **Mute is a text mute only.** It does not silence the member in voice channels.
</Note>

## Custom rules

Write your server's own rules in plain English. The AI applies them alongside the standard categories.

<Info>
  **25 rules**, Black tier only.
</Info>

Each rule has a name, a description of up to 1,000 characters, a single action, and an optional sensitivity that overrides the global value.

<Tip>
  Be specific about what is allowed as well as what is not. "No advertising other servers, but sharing a friend's art or stream link in #showcase is fine" performs far better than "no ads".
</Tip>

<Note>
  Rule text that looks like an attempt to manipulate the AI's instructions is rejected on save. If a rule refuses to save with a generic error, rephrase it as a plain description of the behaviour rather than as a command.
</Note>

<Warning>
  A custom rule that matches is checked against **its own** sensitivity and enforces independently of whether the equivalent category is enabled.
</Warning>

## Escalation

Off by default. When on, repeat offenders receive progressively stronger actions.

<Steps>
  <Step title="Add steps">
    Each step is an infraction count (1–99) plus an action, with a duration for timeout, mute and ban. Up to 20 steps.
  </Step>

  <Step title="Set the decay window">
    1 to 30 days, default 7. After this long, an infraction stops counting.
  </Step>
</Steps>

<Note>
  **The current message counts toward the total.** A step set to 1 infraction fires on the member's first offence, not their second.
</Note>

<Warning>
  A step's action **replaces** the category's punishment but keeps Delete and Alert if the category had them. Escalation escalates the punishment; it does not stop the message being removed.
</Warning>

<Note>
  A duration is capped at one year. The dashboard accepts values like 100,000 hours that the server then rejects with a generic save error — keep durations inside a year.
</Note>

## Exemptions

* **Exempt roles** — members with any of these are skipped entirely
* **Exempt channels** — messages here are never scanned
* **Exempt users** — up to 500, useful for bots

<Note>
  **Members with Administrator are never scanned**, whether or not you exempt them. This check runs before any token is spent.
</Note>

## Confirmation before kick and ban

On by default. Kick and ban actions are posted for a moderator to approve rather than applied immediately.

<Warning>
  **Confirmation does not hold back the deletion.** If the category's action set includes Delete, the message is removed the moment the violation is detected. Only the kick or ban waits for approval.
</Warning>

Confirmations expire after one hour. They are posted to your alert channel, falling back to the log channel, falling back to the channel the message was in.

## Trust Tiers

A token-saving measure. Members with a long clean record are scanned less often.

| Setting                           | Range                 | Default |
| --------------------------------- | --------------------- | ------- |
| Trust threshold                   | 5–1000 clean messages | 50      |
| Sample rate                       | 1 in 2–100            | 1 in 10 |
| Minimum days in server            | 0–365                 | 0       |
| Always scan links and attachments | —                     | On      |

Four safeguards mean a trusted member is never invisible:

1. **Discord invites, `@everyone`/`@here`, and 5+ mentions are always scanned** — no exceptions
2. A burst of 3 messages in 6 seconds forces a scan
3. Sampling never reaches zero — a trusted member is still scanned periodically
4. **Any violation resets trust to nothing instantly**

<Tip>
  On a busy server Trust Tiers can cut token spend substantially, because the members who post most are usually the ones who never break rules.
</Tip>

The **Trusted Members** list shows who currently qualifies. **Reduce** halves someone's score; **Remove** clears it entirely. Both take effect immediately — the save bar is not involved.

## Channels

* **Log channel** — every moderation event, with the matched category and confidence
* **Alert channel** — high-severity events and confirmation prompts

<Warning>
  With neither set, AI moderation runs and acts with no visible record anywhere.
</Warning>

## When the token balance runs out

AI Auto-Mod is fail-closed: if it cannot confirm there is budget, it does not scan.

| Balance              | Behaviour                                     |
| -------------------- | --------------------------------------------- |
| Below 100,000 tokens | A one-time "running low" notice is posted     |
| Below 3,000 tokens   | **Scanning stops.** A paused notice is posted |

Notices go to your alert channel, falling back to the log channel. The low-balance warning fires again if the balance recovers and drops a second time.

<Card title="AI Usage" icon="chart-line" horizontal href="/en/moderation/ai-usage">
  Track spend, see what was scanned, and top up.
</Card>

## What a scan costs

One message produces one AI call. A message with images produces one call for the text plus **one per image, up to four**.

<Tip>
  If image scanning is not important to your server, turning off **Scan images** is the single largest saving available.
</Tip>

## Username scanning

Display names and usernames are checked when a member joins and when they change their name. A Delete action is meaningless for a name, so it becomes an alert instead.

## Troubleshooting

<AccordionGroup>
  <Accordion title="Nothing is being caught">
    Work through these in order:

    1. Is AI Auto-Mod on in Server Setup?
    2. Have you enabled any categories? Only one is on by default.
    3. Are you in Monitor mode? It logs but never acts.
    4. Is your token balance above 3,000?
    5. Are the members involved Administrators, or on an exempt list?
  </Accordion>

  <Accordion title="Too many false positives">
    Lower the sensitivity on the specific category that is over-firing rather than the global slider. Return to Monitor mode while you tune.
  </Accordion>

  <Accordion title="Tokens are draining fast">
    Turn off image scanning, enable Trust Tiers, exempt high-traffic channels that do not need moderation, and disable categories you do not care about.
  </Accordion>

  <Accordion title="I turned it on but nothing at all happens, not even logs">
    A server with the module enabled but no configuration ever saved is completely silent. Open this page, set your categories and a log channel, and save once.
  </Accordion>

  <Accordion title="A member evaded a filter with creative spelling">
    That is exactly what AI Auto-Mod handles and standard Auto-Mod does not. Make sure the relevant category is enabled and its sensitivity is not set too low.
  </Accordion>
</AccordionGroup>

<CardGroup cols={2}>
  <Card title="Auto-Moderation" icon="shield-halved" href="/en/moderation/auto-mod">
    Pattern-based filtering, free on every tier.
  </Card>

  <Card title="AI Server Control" icon="wand-magic-sparkles" href="/en/moderation/ai-server-control">
    Manage the server by talking to the bot.
  </Card>
</CardGroup>