Argus now reports only the moderation actions Discord actually accepted, verification cannot leave members in a half-finished state, and server exports and deletion cover every current guild record.
Fixed
Moderation actions no longer block one another. If Discord cannot delete a message, Argus still applies the configured warning, timeout and direct message, and the log names anything it could not do.
Warning thresholds no longer claim success when Discord refused the punishment. Permission and role-hierarchy failures stay retryable for the next warning.
Verification finishes as one consistent operation. Argus clears only that member's verification role and deadline; if Discord or storage rejects a step, roles are rolled back so the member can safely retry.
Audit-log events use one attribution result. Anti-nuke, external moderation records and server logs now agree on who performed a ban, kick, channel change or role change.
Improved
Member exports now cover every current place their account can appear. That includes scheduled actions, persisted restrictions, configuration activity, moderator activity and references inside server backups, with other members' details redacted.
Removing Argus now removes all guild-scoped records after the 30-day recovery window. Persisted restrictions and configuration audit history are included in the purge.
The privacy notice now matches the product. It explains case evidence, reports, scheduled restrictions, retention periods and automatic deletion after Argus leaves a server.
Added
Every release now runs the complete automated test suite and production website build in CI. Generated release files must also be committed before a build can pass.
Every moderation action is now a case you can read back, reverse and appeal. Argus watches who is doing the damage rather than only what is being said — raids at the door, nukes by the people already inside — and it finally tells you when something needs a person right now.
📁 Cases — one record per incident
Every action opens a case with a number. A case carries the reason, what Argus did, the messages that caused it, moderator notes and the appeal — instead of the same incident scattering across a warning, a log embed and an appeal that only a moderator's memory connected. Case numbers appear in log embeds, in /history and in the new dashboard tab.
The evidence is kept. Offending messages are snapshotted at the moment of action, because they are usually deleted seconds later. A case you read next month still shows what actually happened.
Overturning is a real outcome, not a label. Reverse a decision and that case's warnings stop counting toward punishment thresholds immediately — so a mistake stops pushing somebody toward a ban. Nothing is deleted: /history still shows the warning, marked as voided.
Appeals close the loop. Submitting an appeal marks the case as appealed; approving one overturns it and cancels any scheduled unban.
Raid protection could only see messages. A raid was invisible until it started typing, and the only answer was locking down the whole server — silencing every innocent member too.
Now Argus watches the joins themselves: a minimum account age, and wave detection (N joins inside a window). A member caught by both rules gets the harsher action once, never two punishments for one join.
It tells you what the wave has in common — accounts created in the same window, missing avatars, names sharing a stem once the digits are stripped — as evidence in the log embed.
Quarantine instead of lockdown. Hold the wave with a role and let everyone else carry on. The release is scheduled, so it still happens after a restart.
Requires Discord's Server Members Intent, which your host enables — it is off by default and the gate stays inert until then.
🗂️ Per-channel and per-role settings
Profiles let you be stricter in one place, not just quieter. Until now every filter was server-wide with an ignore list — you could turn a filter off in one channel but never turn it up.
A profile names the channels, categories and roles it applies to, and overrides any subset of your settings. Profiles apply in order, so the strictest goes last, and the applied profile is named in the log embed so nobody has to guess why the same message was judged differently in two channels.
📥 Moderator queue and member reports
Members can report what Argus cannot classify — /report, or right-click a message → Apps → Report Message. Reports are always private to the reporter.
One place for everything Argus noticed but didn't act on: reports, dry-run AI hits and members the join gate is holding, with bulk handling and one-click action.
Dry run finally keeps its findings. It used to compute a full verdict, post an embed and store nothing, so "what would this threshold have caught last week?" was unanswerable. Now every dry-run hit is in the queue with its score and reasoning.
⏳ Temporary bans that actually end
Bans and quarantines can now be temporary. Give a ban a duration from the action modal or the dashboard and the unban is scheduled — durably, so it survives a restart, unlike the timers this replaces.
A permanent ban cancels any pending release, so a stale timer can never quietly lift a ban nobody meant to lift. The same applies to a manual unban, an undo, and an approved appeal.
The dashboard shows everything Argus is going to do later, and lets you cancel it.
🙍 Names, trust and privacy
Usernames and nicknames are filtered. A slur in a nickname shows on every message that member sends and in every mention, and no message filter ever touched it. Argus renames them by default — proportionate where removing the member is not.
Trust scaling. A brand-new account and a two-year member no longer face the same AI threshold: account age, time in the server and roles shift it by up to 12 points either way. Off by default, and the shift is shown in every log embed.
Data retention is yours to set. Choose how long each kind of record is kept, export everything Argus holds about a member with /dataexport, and if Argus is removed from a server its data is purged automatically after 30 days.
Configuration changes are audited — who turned what off, and when.
🛑 Anti-Nuke — stop the person, not the server
Raid protection watched the server; anti-nuke watches people. It counts destructive actions per moderator — channel and role deletions, mass bans and kicks, webhook creation, dangerous permission grants, vanity URL changes — and steps in when one person does too many too fast.
The response is aimed at whoever is doing it. Argus strips their roles, which stops a nuke where it stands: the account still exists but the permissions it was using are gone. Locking every channel punishes your whole membership and leaves the attacker's permissions untouched.
Each kind of action is counted separately, because deleting three channels and banning three people are different signals — and the defaults sit well above what a busy moderator looks like on a normal day.
Argus never punishes a guess. If it cannot see who performed an action it logs and does nothing. The server owner is always exempt, and you can exempt anyone else.
Needs the View Audit Log permission. Off by default.
🔍 Argus can finally see who did what
Moderation done by hand now appears in your records. A moderator who right-clicks someone and hits Ban used to leave no case, no audit row and no analytics entry — so your busiest moderators were the least visible in the numbers. Bans, unbans, kicks and timeouts performed anywhere in Discord are now recorded with a case, attributed to whoever performed them.
"Who deleted that message?" now has an answer in the message log, and it can tell the difference between somebody else deleting it and the author deleting their own.
A manual unban cancels a scheduled one, so a temporary ban lifted early can never come back to lift a later, deliberate one.
🚦 Verification, and quarantines that stick
Members can prove themselves. The join gate could hold or remove somebody but gave them no way back — a false positive waited until a moderator noticed. Now there is a verification panel: one click, or a short question, with an optional deadline before a kick.
A quarantine now survives leaving and rejoining. Discord drops every role when someone leaves, so a held raider could simply rejoin clean while Argus's release timer counted down against nobody. Restrictions are re-applied the moment they come back, and cleared properly when the hold genuinely ends.
Friction where it is needed. Verification can apply only to accounts younger than an age you choose, so established members walk straight in.
There is deliberately no image captcha — those are solved trivially now and are hostile to anyone using a screen reader. What stops a raid wave is requiring a step their tooling does not script.
🔔 Alerts, purge and file filtering
Argus can ping you. Everything it reported was an embed in a log channel, which is the right home for a record and the wrong one for a raid at 3am. Raids, nukes and join waves now mention an alert role, with a severity setting so the role does not become noise people mute.
/purge has arrived — delete recent messages, optionally only from one member, containing certain text, or only bots, links, attachments or embeds. Pinned messages are left alone, and Argus tells you when messages were too old for Discord to bulk-delete rather than quietly skipping them.
Executables are blocked. Attachment scanning only ever looked at images, so .exe, .apk and .scr files passed through untouched. The new file filter catches them by extension — including disguises like invoice.pdf.exe. Archives such as .zip are deliberately left alone by default.
📜 Server log — everything that isn't a message
Joins, with how old the account is. Leaves, with the roles they held — information Discord destroys the moment somebody leaves, and the thing that tells you whether a departure matters.
Role changes with who granted them, nickname changes, channels created and deleted, and optional voice join/leave/move logging.
Every event is individually switchable, because "log everything" on a large server produces a channel nobody can read.
⚡ Staying online when the database has a bad day
A brief database hiccup used to take Argus offline entirely. If the connection to the database wobbled for even half a minute, Argus assumed another server had taken over, shut itself down, and restarted — straight back into the same problem, over and over, until the database recovered. On a single-server setup there was never another server to hand over to, so this was pure downtime for no benefit.
Argus now rides it out. A short outage no longer stops the bot; moderation keeps running, and it reconnects on its own the moment the database comes back. It only ever steps aside if a second server is genuinely online and able to take over — the guarantee that two copies never run at once is unchanged.
Starting up without a database no longer fails. Argus comes up, serves the dashboard and status page, and waits — retrying quietly in the background instead of restarting in a loop.
A single slow response is no longer treated as a failure. It now takes several consecutive failures, over a longer window, before Argus concludes anything is actually wrong.
🐛 Fixes
The dashboard could show you stale settings. Saving from the dashboard wrote to the database without refreshing the settings that process was holding, so later reads on the same page — a manual backup, the warning counts — could use pre-save values. A server configured entirely from the dashboard, with no control panel, was never refreshed at all.
The global timeout duration never applied. It was accepted by the dashboard and read throughout the bot, but was missing from the database schema, so every save of it was silently discarded and the built-in default was used instead.
A raid lockdown never survived a restart. The expiry time was written to a field that did not exist in the schema and was therefore dropped, so the recovery pass on startup found nothing to restore.
Warning counts now ignore voided warnings everywhere — the bot, the dashboard and /history all agree on the number that actually counts.
Threads ignored your channel settings. A profile set up for a channel silently stopped applying inside that channel's threads, because Discord reports a thread's parent as the channel rather than the category. Threads now follow the channel they live in.
A scheduled action could burn its retries on nothing. When the desired outcome had already happened — the member was already unbanned, or had left — Argus failed to recognise it and retried five times before giving up.
Warnings can now stop counting toward your punishment thresholds after a period you choose — so a member warned twice last year isn't one message away from a ban.
⏳ Warning Decay
Pick a window, warnings age out of it. Set a number of days and only warnings from inside that window count toward the thresholds on your Punishments page. A member who was warned twice eight months ago starts fresh, instead of carrying it forever.
Nothing is deleted. Decay changes what *counts*, not what's *recorded* — /history still shows every warning ever issued, with the full timeline of actions, appeals and notes alongside it. A decayed warning is context a moderator can still read; it just no longer pushes anyone toward a ban.
Thresholds still only fire once. A threshold that has already been served isn't re-triggered when an old warning ages out and a new one lands — you won't see somebody kicked twice for crossing "3 warnings" once. Once their slate is genuinely clear, the threshold arms again.
The dashboard shows both numbers. The Warnings page displays the count that actually counts, with the lifetime total beside it — so "3 of 5" tells you at a glance that two have expired.
Default: Off — nothing changes until you turn it on. Configure it at the bottom of the dashboard's Punishments page.
🐛 Fixes
A punishment threshold could be silently skipped. When two warnings landed at the same instant — a spam burst tripping two filters at once — both could read the same warning count, and the threshold in between was never triggered at all. Thresholds are now matched by "at or above", so a busy moment can't slip past one.
A punishment that fails is retried. If Argus couldn't carry out a threshold action — missing permissions, a role above its own — that threshold used to be considered served anyway. It's now left armed for the next warning.
Warning lookups are indexed. The warning count runs on every auto-mod warning, and on large servers it was scanning the whole collection each time.
The Context-Aware filter now reads all 28 languages Argus already filtered for profanity — detecting each message's language automatically, and understanding what "you" means in every one of them.
🌍 Multilingual Context-Aware Filtering
28 languages, detected per message: English, Spanish, French, German, Italian, Portuguese, Russian, Dutch, Polish, Turkish, Japanese, Chinese, Korean, Arabic, Persian, Hindi, Thai, Swedish, Danish, Norwegian, Finnish, Czech, Hungarian, Filipino, Esperanto, Kabyle, Klingon, and Québécois French. No configuration — Argus works out which language each message is in and applies that language's rules.
It knows who "you" is in every language: The context engine's whole premise is telling *“this game is stupid”* from *“you are stupid”* — which requires knowing that "you" is du, tu, ты, お前 or mo. Every language ships with its own second-person pronouns, negation words and joking markers, so a directed insult is recognised as directed no matter what it's written in.
Severity is per language, not translated: German *Idiot* is milder than English *idiot*; Spanish *gilipollas* is stronger; Japanese 死ね sits at the top of the scale. Each language's terms are weighted on their own terms rather than machine-translated from English.
Mixed-language servers handled: English is always analysed alongside the detected language, because English profanity travels into every other language's chat. Cross-language collisions are guarded — Dutch *hoe* ("how") is no longer read as an English slur, and Turkish *top* ("ball") is no longer read as one either.
Bypass detection works per script: Word-boundary matching now understands Cyrillic, Arabic, Devanagari and Hangul, and languages that glue endings onto words (Turkish *aptal*→*aptalsın*, Finnish *vittu*→*vittuun*) or use no spaces at all (Japanese, Chinese, Thai, Korean) are matched appropriately.
Still fast, still local: Roughly 0.07 ms per message — about 15% more than the single-language version, for 28× the coverage. No third-party API, no message content leaves your bot.
🐛 Fixes
Russian and Greek messages are no longer penalised. Ordinary text in those languages was being mangled into Latin lookalikes and flagged as a deliberate bypass attempt, adding a hidden penalty to every message. Argus now folds lookalike characters only where scripts are genuinely mixed — which is what an actual homoglyph attack looks like — so writing normally in your own language costs nothing.
A word shared between languages no longer counts twice. *“idiot”* is a word in six of the supported languages, and a single use was being scored once per language.
A filter that reads the room. Argus now scores messages for toxicity and harassment using natural-language analysis — catching the tricks that slip past word lists, without punishing people for saying a play was insane.
🧠 Context-Aware / AI Filter
Scores, not blocklists: Every message gets a 0–100 hostility score instead of a yes/no keyword match. You set the threshold — lenient, balanced, strict, or a custom number — and Argus only acts above it.
It knows who you're talking to: The same word lands very differently depending on aim. “this game is stupid” is venting; “you are stupid” is directed at a person. Argus weighs second-person pronouns, mentions and replies, and quietly discounts self-deprecation, quoted speech, negation (“I'd never call you that”) and joking markers.
Bypass techniques don't work: Zero-width spaces, homoglyphs (Cyrillic а for Latin a), leetspeak, fullwidth and fancy-font letters, zalgo, stretched words and letter-by-letter spacing are all folded back before analysis — and because hiding a word is itself evidence of intent, obfuscation raises the score rather than escaping it.
Harmless slang stays harmless: Built-in guards for the Scunthorpe family (classic, assassin, Dickens, analysis), praise collocations (“fucking brilliant”, “sick play”) and fixed idioms (“holy shit”, “hell yeah”, “killing it”). Your existing Allow List still overrides everything.
Six categories, individually switchable: Toxicity, Harassment, Hate Speech, Threats, Self-Harm Encouragement and Sexual Harassment. Slurs, threats and self-harm are never softened by tone — no amount of “lol” excuses them.
Catches split-up messages: Abuse spread across several quick messages is scored against the 15-second buffer, not just one line at a time.
Repeat offenders escalate: One bad moment scores lower than the fourth in three minutes.
Shows its work: Every action logs the score, the exact terms matched, their weights and which context modifiers applied — so moderators can see why, not just what.
Try before you enforce: Turn on Dry Run and Argus logs what it would have flagged while changing nothing. The dashboard's new AI Filter tab also has a Live Tester — paste any message and see the score, the signals and the verdict against your live settings.
Runs locally: No third-party API, no message content leaves your bot, no extra dependency — roughly 0.06 ms per message.
Configure it from the Discord control panel's new AI Filter button or the dashboard's AI Filter tab. Default: Off — enable it when you're ready to tune it.
Keep a record of what your members edit and delete — Argus now logs edited, deleted, and bulk-deleted messages straight to your log channel.
📝 Message Audit Log
Edit & delete tracking: When a member edits a message, Argus logs the before and after; when one is deleted, it logs the original content, author, and channel — so evidence isn't lost when someone cleans up after themselves.
Bulk deletes too: Channel purges are logged with a summary and a full text transcript of the messages Argus still had cached.
Your channel, your rules: Logs go to your existing log channel by default, or pick a dedicated one. Toggle edits, deletions, and bulk deletions independently from the Discord control panel or the new Message Log tab in the dashboard.
Quiet by design: Argus skips its own messages, ignored channels, and anything it removed via auto-mod (already logged) — so the audit log stays signal, not noise.
Default: Off — enable per-server when you want the paper trail.
Look up anyone's full moderation record without leaving Discord — warnings, actions, appeals, and private moderator notes, all in one timeline.
🗂️ Member History & Notes
/history command: One command shows a user's complete record in this server — warnings, every auto-mod and manual action, ban appeals, and notes — merged into a single reverse-chronological timeline with summary counts.
Private Moderator Notes: Leave context for the next moderator with /note add — notes are visible only to your mod team, never to the user. Review them with /note list and clean up with /note remove.
Nothing to enable: History draws on the moderation data Argus already keeps, so it works out of the box — Moderate Members permission is all that's required.
Durable by design: Moderator notes are kept indefinitely; auto-mod action records still age out after 180 days, and /history makes that distinction clear.
A new dashboard tab that turns your moderation into numbers — see how much your filters are catching, which fire most, who your repeat offenders are, and how appeals are trending.
📊 Mod Analytics
New Analytics Tab: A dedicated dashboard view with summary cards, an actions-over-time chart, and breakdowns — switchable between the last 7 and 30 days.
Full Audit Trail: Every moderation event is now recorded — auto-mod deletes/warns/timeouts plus manual warns, timeouts, kicks, and bans from the log buttons, the action modal, and the dashboard.
Know Your Filters: See which filters fire most and how actions break down by type, so you can tune your configuration with real data.
Leaderboards: Top offenders and top moderators at a glance, plus an appeal funnel showing approval rate.
Privacy-minded: Records auto-expire after 180 days. Analytics start accumulating from this update forward.
Banned users can now appeal their ban, and your moderators can review and decide — from Discord or the web dashboard, with a full audit trail.
⚖️ Ban Appeals
Two Ways to Appeal: When a user is banned, Argus DMs them a notice with an Appeal button and a link to a dedicated web appeal page — so they can still appeal even if their DMs are closed.
Dual Review: Approve or deny appeals from a Discord review channel with one-click buttons, or from the new Appeals tab in the web dashboard. Both stay in sync.
Auto-Unban on Approve: Approving an appeal lifts the ban automatically and notifies the user; denials can include a reason that's shared with them.
Abuse Protection: One open appeal per user at a time, a configurable re-appeal cooldown after a denial, and server-side verification that the user is actually banned before an appeal is accepted.
Configurable: Set the review channel, re-appeal cooldown, decision DMs, and auto-unban behavior from the dashboard's Appeals tab.
Default: Off — enable per-server when you're ready to accept appeals.
Argus can now snapshot your entire server and restore it later — a built-in safety net against accidental deletions, nukes, and botched reorganizations.
📦 Server Backups
Full Snapshots: Capture roles, categories, channels, and every permission overwrite, plus server settings (name, icon, banner, verification level, content filter, AFK & system channels), custom emojis & stickers, the ban list, and member roles & nicknames.
One-Click Restore: Rebuild your server from any saved backup. Restoring is server-owner-only and always requires explicit confirmation, since it deletes current roles and channels before recreating them.
Web Dashboard: A new Backups tab lets you create, browse, restore, and delete backups — with live progress updates streamed as a restore runs.
Slash Command: Manage everything from Discord with /backup create, list, info, load, delete, and interval.
Automatic Backups: Opt in to scheduled backups every X hours (minimum 24h); Argus keeps the most recent backups per server and prunes the rest.
Resilient Restores: Rate-limit aware, and anything Argus can't recreate (e.g. roles above its own) is safely skipped and reported instead of failing the whole restore.
Default: Off — create backups on demand or enable scheduled backups per server.
Argus can now dynamically manage channel slowmode based on real-time message activity — keeping conversations readable without manual intervention.
🐢 Smart Auto Slowmode
Automatic Adjustment: When a channel gets busy, slowmode steps up gradually through Discord's valid intervals. When activity drops and stays low for the configured cooldown, slowmode relaxes back down.
Fully Customizable: Set the minimum and maximum slowmode range, the message-rate threshold that triggers an increase, the time window for measurement, and how long the channel must stay quiet before relaxing.
Per-Channel Overrides: Configure different thresholds, slowmode ranges, or cooldowns for individual channels — or apply one global policy everywhere.
Smart Logging: Each adjustment is logged to the server's log channel with the channel name, old → new slowmode value, and the message rate that triggered it (rate-limited to avoid spam).
Control Panel: Toggle and configure via the new Smart Slowmode button in the Discord control panel.
Permission Checks: When enabling, Argus verifies it has ManageChannels and warns if it's missing.
Graceful Cleanup: Channel-specific state is cleaned up when channels are deleted, and stale tracking data is periodically pruned.
Sharper word detection that stops innocent words from being caught by the NSFW filter.
🐛 False Positive Fixes
Fixed a bug where normal words like "analysis" or "analisis" could be incorrectly flagged as NSFW because they contain the letters "a-n-a-l" in sequence. Argus now only flags words when they appear as whole, standalone terms — not when they happen to be part of a longer, innocent word.
A full security audit and hardening pass to keep your servers safer.
🔒 Security Update
Permission checks tightened: Moderation buttons in the log channel now require the correct Discord permissions for each action — Kick Members for kicks, Ban Members for bans, and Moderate Members for timeouts. Previously any moderator with Manage Messages could use all actions.
Config protections: Only server admins with Manage Server permission can now change Argus's settings via the control panel buttons. Regular users can no longer toggle filters or modify the configuration.
Dashboard security improved: The web dashboard now includes rate limiting, cross-site request protection, and more secure session handling.
Link filter privacy: A potential leak of the Google Safe Browsing API key through web server access logs has been fixed.
🛡️ Moderation Improvements
Allow List refined: Adding a word to the Allow List now only exempts that specific word from being flagged — it no longer disables ALL filters for messages containing it.
Role hierarchy respected: Moderators can no longer warn, mute, kick, or ban users at or above their own role level.
Image scanning more reliable: NSFW image detection now queues scans if the AI model is still loading after a restart, instead of silently skipping them.
⚡ Reliability
Crash recovery for lockdowns: If the bot restarts during a raid lockdown, it now remembers and resumes the lockdown rather than leaving it partially applied.
Memory improvements: Periodic cleanup of unused tracking data prevents slowdowns on busy servers over time.
Argus now supports an Allow List, letting you specify words or phrases that will never trigger the auto-mod, even if they contain flagged terms.
✅ Global Allow List
Bypass Everything: Allowed words bypass profanity, NSFW, and custom word filters entirely.
Cross-Platform Support: Manage your list directly from the new "Allow List" button in the Discord terminal or via the dedicated Allow List tab in the web dashboard.
Exact Match Logic: Prevents common false positives like "bass" or "assassin" from being caught by the word filters.
Argus now detects profanity across 29 languages, with a dedicated engine for scripts that don't use spaces between words.
🌐 Multilingual Profanity Filtering
Detects profanity across 29 languages, including Arabic, Spanish, French, German, Portuguese, Russian, Korean, Hindi, Polish, Dutch, Turkish, and more.
Japanese, Thai, and Chinese are now supported with a dedicated detection engine built specifically for scripts that don't use spaces between words.
⚡ False Positive Fixes
Fixed a bug where normal phrases like "was it" could incorrectly trigger the profanity filter.
Fixed a bug where short strings inside URLs (e.g. YouTube links) could incorrectly match against the word filter.