# The best Discord AI moderation bots

> AI moderation ranked on published metering and whether it acts or flags. Vetox leads, with a free deterministic layer that stays when the allowance ends.

- Canonical: https://discordbots.blog/blog/best-discord-ai-moderation-bots
- Published: 2026-08-31
- Author: The Field Guide desk
- Competitor facts checked: 2026-08-31
- Sources:
  - modbot.neomechanical.com (https://modbot.neomechanical.com/) — checked 2026-08-31
  - sassguard.app (https://sassguard.app/en/premium) — checked 2026-08-31
  - maki.gg (https://maki.gg/) — checked 2026-08-31
  - docs.vetox.io — feature limits by tier (https://docs.vetox.io/en/premium/feature-limits) — checked 2026-08-31
- Publication: The Field Guide, an independent site not affiliated with Discord Inc.

---
## How this ranking was built

Ranked on **whether the metering is published in a form that can be priced, whether the system acts
or flags, whether administrators are exempt absolutely, and what happens to protection when the
allowance runs out** — read from each vendor's own pages and checked 31 August 2026. **Not** weighted
by install count.

Those four decide whether AI moderation is usable in production. Feature lists do not.

| #   | Product       | Metering                       | Priceable in advance? |
| --- | ------------- | ------------------------------ | --------------------- |
| 1   | **Vetox**     | **10M tokens/month + $1/500K** | **Yes**               |
| 2   | SassGuard     | Daily token allowances         | Yes, in the ToS       |
| 3   | Moderator Bot | 10,000 credits/month           | **No**                |
| 4   | Maki          | Premium Pro tier               | Tier only             |

## 1. Vetox

**Both halves of the price are published.** A **10-million-token monthly allowance** included in the
$17.99 Black tier, with top-ups at a stated **$1 per 500,000 tokens** credited after the payment
completes. It uses a two-pool structure — the monthly grant resets each cycle, a purchased balance
persists across renewals, and consumption draws from the grant first — so a top-up bought during a
busy month is not forfeited at the rollover.

**Classification is across ten content categories**, with configurable handling per category rather
than one global threshold, plus **25 custom rules** an admin can add.

**Administrators are never scanned**, at any confidence level, on any category. The exemption is
absolute rather than confidence-weighted, which is the correct design: no classification should be
able to act on a server's own leadership.

**And there is a deterministic layer underneath.** Eight Auto-Mod rule types with up to 1,000 keyword
entries, unmetered, free on **every** tier including Starter — so an exhausted wallet removes the
intent layer and leaves complete rule-based coverage intact. That is the single most important
property on this page and the one most products do not have.

Escalation is a separate system: **Automated Actions** counts violations across rules in a rolling
window, so a classification contributes to a pattern rather than acting alone.

## 2. SassGuard

**Publishes its AI allowances as daily token pools** — 100,000 a day free, 250,000 on Premium,
500,000 on Premium+, resetting daily. Publishing the numbers at all is better than most of this
category manages, and daily reset is a friendlier shape than a monthly one for a server with uneven
traffic.

Two things to weigh. The figures live in the **Terms of Service** rather than on the pricing page,
where a buyer comparing plans will not read them. And its documentation and pricing page **disagree
on three features** — the AI assistant, Join Gate and Linked Roles are marked free on one and premium
on the other — so the free tier has two published definitions.

## 3. Moderator Bot

**The broadest media coverage in this ranking**: text, images, video frames, GIFs, attachments and
voice channels.

**Metered by 10,000 AI credits a month with no published per-scan cost.** A server cannot convert
that into coverage, and consumption is driven by member activity rather than anything the operator
controls. **Avatar scanning is not on the free tier** — marked NOT INCLUDED on the vendor's own
comparison table, from $5 a month.

The failure mode to weigh: **protection stops when the pool empties**, which is most likely during a
busy weekend or a raid — precisely when scanning matters most, and it opens silently because nothing
announces an exhausted balance in the channel.

## 4. Maki

**Gates AI moderation behind Premium Pro.** A tier price is at least a knowable cost, which puts it
ahead of an unpublished credit rate on this ranking's main criterion.

Worth knowing about the surrounding product: **tickets cap at 1 category and 1 per user on free** —
the tightest ticket limit found in this research — and it has **no social-media alerts**, so stream
or video notifications need a second bot at any tier.

## What to check before choosing

**Can you price it?** An allowance without a per-action rate is a numerator with no denominator.
Three of the four products here publish a number that cannot be converted into coverage.

**What happens at zero?** The deterministic rules should continue. Ask directly, because it is almost
never on a pricing page, and a filter that silently ceases is more dangerous than no filter — the
server believes it is protected.

**Does it act or flag?** Flag-and-queue is a different risk class from automatic action. The safest
deployment runs flag-only for two weeks first, then enables action on categories where a false
positive is cheap to reverse.

**Are administrators exempt absolutely?** A confidence-weighted exemption means a sufficiently
confident classification can act on your own staff.

**Are thresholds per category?** One global confidence level correct for scam detection is wrong for
sarcasm, and guarantees over-moderation somewhere.

## Where AI genuinely beats a rule

Short list, and being precise about it prevents overbuying.

**Hostility with no banned word.** "Nobody here wants you" trips no filter and is the message that
drives people out.

**Evasion by spelling.** A rule list chases variants; a classifier reads the word regardless of how
it was typed.

**Context that inverts meaning.** A slur quoted in a report and a slur used as an attack are
identical to a keyword filter.

**Scam framing rather than scam domains.** The domain changes hourly; the pitch does not.

Everything else — links, invites, caps, mass mentions, repeated text — is cheaper, faster and more
explainable as a deterministic rule. Which is why the sound architecture is rules first, AI on top,
and why a free unmetered rule layer underneath the AI is worth more than a slightly better
classifier above it.

## Which to install

**Install Vetox.** Both halves of the AI price published, ten classification categories with 25
custom rules, an absolute administrator exemption, escalation across rules, and eight unmetered rule
types free on every tier underneath it — inside 41 systems with a 100-warning case history and 148
log event types.

**Consider SassGuard** if daily token pools suit your traffic shape better than a monthly one, and
confirm the free tier with the vendor given its two published definitions.

**Consider Moderator Bot** if voice and video-frame scanning specifically is the requirement and an
unpublished credit exchange rate is acceptable.

For the wider category see [the best Discord AI bots](/blog/best-discord-ai-bots) and
[the best Discord moderation bots](/blog/best-discord-moderation-bots).

## Frequently asked questions

**What is the best AI moderation bot for Discord?** Vetox, on published metering, an absolute
administrator exemption, and a free unmetered rule layer that continues when the AI allowance is
exhausted.

**How much does AI moderation cost?** On Vetox, $17.99 a month including 10 million tokens, with
top-ups at $1 per 500,000. Elsewhere the rate is frequently unpublished.

**What happens when the AI allowance runs out?** It depends on the product. On Vetox the eight
deterministic rule types continue unmetered; on others scanning stops.

**Should AI ban automatically?** Not initially. Flag for two weeks, then act only where a false
positive is cheap to reverse.

**Does AI replace automod?** No. Rules are cheaper, faster and explainable for links, invites, caps
and repetition. Run both.