v0.11.0Open BetaJuly 2, 2026
Meet the Warden — your server, protected
The whole safety engine arrives — anti-raid, anti-nuke, one rule builder, backups, verification, and a cross-server trust network.
Lumei has always been the engine that grows your community. Today it starts protecting it too. The Warden is the safety half of Lumei — the guard on your front gate, the watch over your most-trusted accounts, and a memory that never loses a case — and it starts working the moment you invite the bot, on sensible defaults you never have to touch.
The idea is simple: a server should be safe out of the box and endlessly adjustable when you're curious. So every protection is on from the first minute behind a curated preset, and every setting is one click away on a new Warden page in your dashboard, which opens with a plain 0–100 read of how protected you are and what to turn on next.
And the line we won't cross: the pieces that keep a server safe are free. The paywalls sit on scale and polish — higher limits, more restore points, deeper analytics — never on protection itself.
The perimeter — Anti-raid watches your front gate for coordinated joiners; anti-nuke watches your trusted accounts — and guards the bots and webhooks most tools leave open.
One rulebook — A visual when-this, if-that, do-this rule builder feeding a single escalation ladder, so a warning here and an auto-timeout there finally add up to one story.
Recover from anything — Structural backups of your channels, roles, and permissions — restored in one click when something goes wrong, and the restore itself can be undone.
The Smart Gate — A verification gate that tightens itself the moment a raid starts: a friendly button on a calm day, a real challenge when things get loud.
The mod desk — Cases, private notes, a full member dossier, member reports, private modmail, and one-click appeals — the record that remembers everything.
A trust network — Opt in, and an account caught raiding one Lumei server arrives at the next already flagged for your review — hashed, human-confirmed, and only ever advisory.
Everything in this release
New
- Anti-raid — a join gate and join-spike watch that catches a coordinated wave before anyone can post, with a reflex for mass-mention floods.
- Anti-nuke — per-person limits on dangerous actions like mass channel or role deletes, with an instant quarantine when a trusted account goes rogue — and it covers bots and webhooks, not just people.
- The rule builder — one visual canvas to say "when this, and if that, then do this," in place of a dozen disconnected toggle tabs. It reads real regex, and look-alike or leetspeak text is caught by default.
- Escalation and Standing — every warning, automatic action, and note feeds one points-based ladder that forgives over time, so a member who slips once and then behaves is let off the hook on their own.
- Verification with a Smart Gate — a one-tap gate for new members that quietly steps up to a captcha when your raid signal climbs, and steps back down when it's calm.
- Backups and one-click restore — scheduled snapshots of your server's structure, restored in a single click, with a fresh snapshot taken first so the restore is itself reversible.
- Cases and dossiers — every action becomes a numbered case on a member's record, with private staff notes and one screen that holds their whole history.
- Member reports — your members flag a bad message or user straight into a private mod queue, and the reporter stays anonymous.
- Modmail — a member DMs the bot and a private thread opens with your staff, who reply as the bot: one calm channel instead of five pinged moderators.
- Appeals — a punished member gets a clear DM and a one-click appeal that reverses the action the moment you approve it.
- The moderator toolkit — the full set a mod reaches for, from ban and kick to timeout, mute, purge, and lock, each tunable from the dashboard.
- Role permissions — give a trusted mod the toolkit without giving them server-wide Administrator, and keep the dangerous powers with only the roles you choose.
- The trust network — opt in to a shared, privacy-respecting signal, so an account confirmed as a raider on one server arrives pre-flagged on the next. It's off until you turn it on.
- Dossier export — download a member's full record as a spreadsheet for your own files, a Premium reach-back.
Improved
- The Warden dashboard now reads like a document — every setting is a plain sentence with a visible label, not a wall of unmarked switches — and opens on a security score that shows its own math.
Known issues
- AI-native moderation isn't live yet — context-aware toxicity and image scanning are reserved, and the ladder already holds their place, but they aren't built.
- The trust network only helps once servers opt in, both to share and to receive — by design, nothing is shared until you choose to.
- During a raid the gate can weigh signals like a brand-new account or one flagged elsewhere, but that read is advisory — new accounts, shared homes, and VPNs can look alike, so it routes a likely evader to a challenge or to your review, never an automatic ban.
- Role permissions govern what a mod can do in your server; opening the dashboard itself still requires Manage Server for now.
What's next for the Warden
AI-native moderation — a confidence-scored detector that plugs into the same rule builder — is the reserved next chapter, and there's more delegation and analytics on the way. The roadmap has the wider picture, and what you tell us still sets the order.