❓ Server Assistant — Frequently Asked Questions

The full moderation toolkit is free forever — no card, no catch. Core moderation, AutoMod, anti-raid, DM verification, audit logs, warnings, the Customisation Hub, and zero-touch /autopilot setup never cost a thing.

The AI features (SAi, Message Report, Self-trained AutoMod, Pulse) include a 150,000-token free trial. After that, Premium ($7 USD/mo) keeps them going — but you're only charged once your trial tokens run out, so light-use servers may never pay. Got your own AI key? Premium BYOK is $3 USD/mo. See Pricing for the full breakdown.

You'll need Manage Server permission. Use the invite link, or find Server Assistant on Top.gg or discordbotlist.com. After authorising, the server owner or any admin with Manage Server runs /setup — and whoever invited the bot is DM'd the setup wizard automatically.

The server owner or any admin with the Manage Server permission can run /setup — the same permission Discord requires to add the bot in the first place, so whoever installed it can finish onboarding without waiting on the owner. /setup maps your roles to permission tiers; the role-to-tier mapping itself still gates the genuinely dangerous actions, so opening setup to Manage Server admins is safe while keeping the bot easy to get running.

About 60 seconds — and you may not even need to run /setup: when the bot joins it DMs the owner the setup wizard, pre-filled, ready to finish in DMs. However you start it, the bot auto-detects your channels and roles so you mostly just confirm: channels, roles, then a server-type + AI step — the server-type picker tunes AutoMod, the punishment ladder, and anti-raid to your kind of community (Gaming, Creator, Crypto, Education, NSFW, General…). In a hurry, /autopilot does it all in one command. Then /settings opens nine optional panels to fine-tune anything from embed colour to the AutoMod ladder.

Nothing — they're preserved. /setup only changes channel/role mappings and AI config. Warnings, notes, snippets, and audit logs are untouched.

No — the bot enforces Discord's role hierarchy: a mod can't action anyone whose role is at or above their own. The owner can action anyone except other owners.

Discord requires the bot's role to be higher than the target's highest role. Drag the bot's role up in Server Settings → Roles.

Yes, within 24 hours. Press the ↩️ Undo button on the action's confirmation message to reverse your most recent reversible action — bans, mutes, channel deletions, nickname changes.

Private to staff. The user gets a DM when warned, but there's no public-facing list. Use notes for purely internal observations that aren't formal warnings.

Repeated AutoMod violations auto-escalate. Defaults (Standard preset):

Auto-warnings	Action
3	10-minute timeout
5	1-hour timeout
10	Ban

Only AutoMod-issued warnings count toward escalation; manual staff warnings are tracked separately. Change via /settings → Quick Presets → Punishment Ladder (Gentle 5/10/20 · Standard 3/5/10 · Strict 2/3/5).

Filter packs use word boundaries to avoid the Scunthorpe problem — assassin, classic, Scunthorpe don't trigger. The default "Scams + slurs (gaming)" preset deliberately allows casual profanity. Strict mode adds it with higher false-positive risk — use only for child-friendly servers.

Yes. /automod → Add Filter for custom regex (validated; catastrophic-backtracking patterns auto-rejected). Add domains to the link blocklist for phishing sites.

No. Staff are completely exempt regardless of tier. Staff = anyone holding a role mapped via /setup or /settings → Role Tiers.

Both go on the record. Only AutoMod warnings count toward escalation thresholds (the punishment ladder). The audit log distinguishes the source.

Yes — DM explaining which rule fired and their warning count. If their DMs are closed, the skip is logged but removal + warning still go through.

Enable in /settings → 👥 Members → Verification. New joiners get a DM with a green Verify button. Clicking it grants the configured verified role, which gates channels via Discord's role permissions.

To set up those gated channels without editing permissions by hand, use the Auto-configure channels button in the same Verification panel: it proposes which channels stay visible to unverified members (your verify, rules, system and announcements channels), lets you keep extra ones visible, then on Apply hides everything else and opens it to your verified role. There's a one-tap Undo. It needs a verified role set and the bot to have Manage Channels + Manage Roles.

The failure is logged; the user stays unverified. They can see @everyone channels but not gated ones.

Currently it's fixed wording. Custom verification copy is on the roadmap.

/settings → 🔔 Notifications picks which staff roles are @-pinged per event:

• Anti-raid alerts — mass-join detection trips
• Dangerous-action approval requests — non-Owner needs sign-off
• AutoMod escalations — user crosses a punishment threshold (timeout / kick / ban)

Up to 10 roles per event; empty = silent. Routine single-message blocks don't ping (only escalations do).

These live under Instant alerts in the hub. The other half — Updates & digests — is where you set how often Pulse arrives (Off / Daily / Weekly / Monthly) and toggle Milestones and Announcements on or off.

Your call. In /settings → 🔔 Notifications → Updates & digests, set Pulse to Off / Daily / Weekly / Monthly. Any cadence you'd already set carries over automatically.

Server Assistant celebrates your community's wins in staff chat with a warm card — member-count landmarks (100, 500, 1k…), server anniversaries, and moderation milestones, framed as your achievement. Capped to at most one a day, and fully toggleable in /settings → 🔔 Notifications.

No. Monitoring staff-chat is free. The bot runs pattern matching (regex + keyword lookup) on every message at zero token cost. AI tokens are only spent when staff write a genuinely fuzzy request the pattern matcher can't resolve.

The decision chain on every staff-chat message:

Every staff-chat message
        │
        ▼
[is_bot_addressed]      ← pure string check (mentions, bot name, prefixes)
[has_actionable_command] ← regex vs FORBIDDEN/DANGEROUS/SAFE keyword sets
[looks_like_command]    ← imperative-verb match (make, show, find, run...)
        │
        ├── No gate matches
        │   → 💸 0 tokens. Silently ignored.
        │
        ├── Action keyword matched (warn / mute / kick / ban / purge / etc.)
        │   → 💸 0 tokens. Direct structured handler executes.
        │
        └── Bot addressed AND no clear action keyword
            → AI interprets intent
            → ~400–500 tokens per call (billed to your guild's allowance)

Examples that cost zero tokens:

Message	Why free
"brb getting coffee"	No gate matches → silently ignored
warn @user spam	Action keyword warn → structured handler
mute jen 1h	Action keyword mute → pure-code username resolver
purge since 1pm	Action keyword purge → time parser
info @user · stats · help	Keyword matches → direct handlers

That covers ~90%+ of routine moderation in staff-chat.

Examples that DO use AI tokens:

Message	Why AI is called
"@bot can you tell me who's been most active this week?"	No keyword + bot addressed → AI interprets
"bot, find me someone who joined yesterday and posted"	No keyword + bot addressed → AI interprets
"make me a summary of last night's drama"	Imperative verb (make) + no action keyword → AI interprets

Cost per fuzzy call: ~400–500 tokens (~$0.003 at Sonnet rates), billed to your server's AI allowance with a feature tag (nl_intent, nl_chat, nl_extract) so you can see exactly which interactions ate which tokens in /premium.

To drive NL spend to truly zero: the owner runs /privacy and toggles off "Natural-language commands in staff-chat". The bot then responds only to slash commands and explicit @mentions — no fuzzy AI parsing at all. Pattern-match action keywords still work via @bot warn @user style invocations.

Design philosophy: structured commands are free; the AI is an opt-in fallback for when you can't be bothered to remember the exact syntax. Most servers stay close to zero NL spend in normal use.

The address check is multi-tenant aware — it resolves the bot's actual nickname in your server at runtime, not a hardcoded name. So if you renamed the bot to "Pepper" or "Watchdog", any of these patterns trigger it:

• @mention — @Pepper mute @user spam
• Reply to a bot message — quoting/replying to me
• Command prefix — !warn @user or /info @user
• Bot's nickname in this guild — Pepper, mute @user or Watchdog: warn @user
• Bot's base username — ServerAssistant, ban @user
• Generic fallbacks — bot, mute @user or assistant, show stats
• Action keyword at the start of the message — warn @user spam (the action word leads, optionally behind one interjection like ok/yes or a vocative like bot,)

Why the start? Requiring the action keyword to lead matches how shorthand actually gets typed (warn @user, bot, ban @user, ok mute @user 1h), so casual chat like "can you help me move grass this weekend?" doesn't pick up move and respond. Need to address the bot mid-sentence? Just @mention it.

The "fuzzy logic" lives on the execution side (AI interprets what action you wanted), not addressing. Addressing is intentionally strict — match-by-name, @mention, or action-word-at-start — so the bot doesn't burn tokens whenever someone casually mentions an action word.

The free tier includes a 150,000 token lifetime allowance — enough to evaluate every AI feature. Core moderation (AutoMod, anti-raid, warnings, slash commands) is always free regardless of token usage.

When your trial tokens run out, subscribe to Premium ($7 USD/month) for 750K tokens/month — card only charged when tokens deplete. Or bring your own Anthropic/xAI/OpenAI key via /ai-config for Premium BYOK at $3/month. See Pricing for full details.

Not immediately. On subscribing to Premium ($7/mo), your card is saved but not charged. Billing starts only when your 150K free tokens are fully used — light-use servers may never hit that limit and never pay a cent despite being subscribed.

Premium BYOK ($3/mo) is charged immediately on subscribe, since you're paying for feature access rather than tokens.

Only what's needed for the request. Message Report sends ~20 messages of context; SAi sends your question plus your server's settings and recent event summary; /imagine sends just your text prompt. No user metadata, IDs, or persistent identifiers are transmitted. Full details in the Privacy Policy.

The default shared key uses Anthropic Claude (Sonnet) for all AI features — SAi, Message Report, Self-trained AutoMod proposals, and Bot Health Insurance checks.

Supply your own key via /ai-config to choose Anthropic, xAI Grok, or OpenAI. Image generation (/imagine) runs on the shared service — the free trial covers ~30 images, unlimited on Premium — or add your own DALL·E 3 / Stable Diffusion key via /ai-config for unmetered generation.

Yes — /ai-config → Skip AI disables all AI features for your server. /imagine and Message Report will decline gracefully. You can also use /privacy to disable individual AI features while keeping others active.

On the host's infrastructure, encrypted at rest. Per-server data is isolated. Full details in the Privacy Policy.

Your encrypted secrets (your own AI keys etc.) are wiped immediately. Operational data (warnings, notes, audit log) is retained briefly in case you re-add the bot. Request full manual deletion via /support.

Not yet. A premium-tier read-only API for your own server's data is on the roadmap.

One-shot export available on request via /support. A /export-server-data slash command is on the roadmap.

First-time global propagation can take up to 1 hour. If they're still missing after that, hard-refresh Discord (Ctrl+R).

• View Info / View Warnings = right-click a user. Message Report = right-click a message. Both under Apps ▸ Server Assistant.
• Hard-refresh Discord (Ctrl+R) — local cache may be stale.
• First-time global propagation can take up to 1 hour.

Most outages are transient — the bot auto-reconnects. If it's been down 5+ minutes, open a ticket via /support from another server you own that has the bot.

Discord may be rate-limited. Wait 60 seconds and try again. If still unresponsive, open a ticket via /support.

Run /support from any server that has the bot, tap Create a ticket, and send your question as a DM — you'll get a reply via DM. Full details here.

Open a ticket via /support and describe the idea — the more concrete the use-case, the better. Check the roadmap first to see if it's already planned.

No — the bot's source is closed. Documentation (this site, privacy, terms, changelog, roadmap) is public on GitHub. Bug reports and feature requests go through /support.

No — Server Assistant is a fully managed, hosted bot. Invite it and configure everything in Discord; nothing to install or maintain on your end.