For the complete documentation index, see llms.txt. This page is also available as Markdown.

Nova Language Moderation

Nova Language Moderation lets you declare which language(s) each channel in your community speaks — and have Nova enforce it politely, automatically, in the background. Set #general to English-only, give #español to Spanish, let a support topic allow English + Chinese, or set one rule for the whole server/group. Nova then screens every message (including photo/video captions) and, when someone posts in a language that channel doesn't allow, responds the way you chose: a friendly redirect, a deletion, a mod alert — or any combination.

The killer detail is the redirect: if the message was in Spanish and another of your channels does allow Spanish, Nova's warning points the member there — "this channel is English-only — you're welcome to continue in #español!". No configuration needed; Nova derives it from your other channel rules. Wrong-language posts stop being moderation incidents and become signposts.

Unlike Guardian (which always protects the whole community), language rules are per-channel: each channel, thread, or Telegram forum topic can have its own allowed languages, with an optional community-wide default for everything else.


The Three Behaviors

You choose what happens on a wrong-language message — each is an independent toggle:

Behavior
What Nova does
Default

💬 Warn & redirect

Publicly replies to the author: which languages the channel speaks, what their message looked like, and where their language is welcome. The notice self-deletes after ~45 seconds, and the same member isn't re-warned in the same channel more than once every few minutes — the channel never fills with moderation noise.

On

🗑️ Auto-delete

Removes the violating message automatically (the warning then says it was removed and where to go instead).

Off

🚨 Alert mods

Posts a review card to your mod channel: author, detected language + confidence, the quoted message, and what action was taken.

Off


How Detection Works — and why it won't misfire

Language detection on chat messages is genuinely hard: real chat is short, full of slang, tickers, addresses, links and emoji. Nova uses a tiered detector designed so false positives are rare enough to trust with deletion:

  1. Writing-system check (instant). A message written dominantly in a script none of the channel's languages use — Cyrillic in an English-only channel, Chinese in a Spanish channel — is an unambiguous violation. This handles most real cases with zero cost.

  2. Same-script cases go to an AI check. Spanish in an English channel shares the Latin alphabet, so a fast language library takes a first look — but its opinion is never enforced directly. Anything suspicious is confirmed by an AI that reads the actual message with slang and context awareness before Nova acts.

What never gets flagged:

  • Short messages — "gm", "lol", "ok", one-word replies, emoji reactions.

  • Crypto/community slang — "wagmi", "lfg", "ser", "dyor", "bullish on stx ngl" are English, and Nova knows it.

  • Links, tickers, addresses, usernames, code blocks — stripped before detection.

  • Code-switching"thanks bro pero cuándo es el launch?" in an English channel is fine; mixing in an allowed language counts as allowed.

  • Quoted foreign phrases"what does 'привет' mean?" is a question, not a violation.

  • Moderators and admins — always exempt.

  • Allowlisted phrases — project names, catchphrases, or greetings your community tolerates.

AI Check Budget

The writing-system check (tier 1) is free and unlimited — it handles most real violations on its own. Only the AI confirmation for same-script cases draws from a daily budget set by your community's plan:

Free

Community PRO

AI language checks / day

50

500

When the day's budget is spent, script-based detection keeps running at full strength; ambiguous same-script cases are simply let through (never flagged without AI confirmation — the same when-in-doubt-stay-silent bias) until the budget resets at 00:00 UTC. Ask "@Nova language moderation status" to see today's usage vs the cap.


What Members See

A wrong-language post in an English-only channel (warn-only mode) gets a reply like:

With auto-delete on, the message disappears and the notice adapts:

The notice quietly deletes itself after ~45 seconds, keeping the channel clean.

What Mods See

With alert mods on, a card lands in your mod-review channel:


Set It Up — just talk to Nova

No dashboard, no wizard — configure it conversationally. (Setup is admin-only, in your server/group — not DMs.) @mention Nova or reply to it:

  1. Set a channel's language (run it in the channel, or name one):

    • "@Nova make this channel English-only"

  2. Or set a community-wide default:

    • "the whole server is English-only" — channels with their own rule keep it; everything else follows the default.

  3. Choose the behaviors:

    • "delete messages in other languages, and send the language alerts to #mod-room"

  4. Allowlist what's normal for your community:

    • "allowlist 'hasta la moon' — it's our meme"

  5. Check or change anytime:

    • "what language rules do we have?" · "turn off language moderation"

Bot permissions for deletion. Auto-delete needs Manage Messages on Discord, or admin with delete rights on Telegram. If a permission is missing, Nova degrades gracefully (warn/alert still work) and the mod card says exactly what failed.


Configuration Options

Option
What it does

Per-channel languages

The allowed language(s) for each channel / thread / forum topic

Community-wide default

A * rule covering every channel without its own rule

Warn & redirect

Public notice + pointer to the right channel (default on)

Auto-delete

Remove violating messages (default off)

Alert mods

Review cards to your mod channel (default off)

Mod-review channel

Where alert cards go (defaults to where you set it up)

Allowlist

Phrases that never flag, in any language

60+ languages supported — including English, Spanish, Portuguese, French, German, Italian, Dutch, Polish, Turkish, Indonesian, Vietnamese, Filipino, Russian, Ukrainian, Chinese, Japanese, Korean, Arabic, Hindi, Thai, Hebrew, Greek and more. Name them naturally ("english", "chinese") — Nova resolves them.


Privacy Note

Language detection happens in real time; ordinary messages in the right language are never stored. Only actual violations are written to the audit log — the detected language, confidence, action taken, and a short snippet for the mod card.


A rollout tip: multilingual communities typically pair a per-channel language map (#general-en / #español / #中文) with warn-only mode for a week, then turn on deletion once they've seen the calls.