Custodian is layered defense for Discord: it filters scam links and shuts down raids and takeovers at the gate, contains the scammer and bot accounts that slip through, and gives your mods a profile on every member, with every action written to a tamper-evident ledger. Built for real moderators, never run on Administrator.
Minimal permissions, requested by name. Set up in one command.
Link filter, dehoist, malware scan, join-risk, clustering and cross-server reputation all feed one Threat Assessment. Your mods stop asking "which system fired?" and see one risk read, the evidence behind every signal, and one decision. The same card shows up in Discord and here on the dashboard.
* individuating signal -- points at a specific bad actor
Ban -- Overwhelming evidence -- remove them.
Example only. Tap any signal to see why it fired.
Custodian is built in layers: stop the obvious threats on the way in, contain what gets through, and keep a complete, verifiable record. Each system does one job and passes the next the evidence.
The moment a join-flood hits, the server locks down on its own: verification goes up, DMs pause, and the arrivals are quarantined or kicked. It then lifts automatically once the wave passes. No mod has to be awake for it.
.antiraid on.antiraid lockdownReal-time removal of scam and phishing links: known scam domains, brand typo-squats and homoglyphs, and IP-grabbers. Deletes and logs the attempt (evidence preserved), can contain the poster, and syncs to native AutoMod.
.linkfilter on.linkfilter automod syncA real, offline antivirus. Native Discord has nothing like it. It reads attachment bytes to catch a disguised executable (a “photo.png” that is really an .exe), archive payloads and zip bombs, password-protected archives, macro docs, malware links, and known-bad files by SHA-256. Confident hits are quarantined to Restricted (never a ban) and logged with the file's hash, never the file. Train it on samples and it auto-updates from free Discord-scam feeds.
.av.av scan.av trainStops a takeover in progress. It tracks each actor's dangerous audit-log actions (tight limits on channel, role, webhook and bot destruction), and on a breach strips their roles, contains or bans them, and DMs the owner. Trusts no one but the owner and your whitelist.
.antinuke on.antinuke whitelist @roleStrip a suspect account of its roles, restrict it to a holding area, and capture the messages that gave it away as evidence. Linked accounts are grouped so a whole scam ring is handled together, then each completes a timed verification (or one-click Easy Verify) before it can return.
.contain @user.release.contain clusterLearns your server's own impersonation signals (staff names, roles, channel topics) and quietly renames or automatically contains the “Binance Support” and leading-symbol hoisters that prey on members below a role floor you set. Trusted roles are never touched.
.dehoist.dehoist scan/reportBan, kick, timeout, warn, purge, slowmode and channel lock, offered as prefix commands and native slash commands, rendered in Discord's Components V2 UI. Everything a mod team reaches for, in one place.
.ban.timeout.purge.lockDiscord has no real malware scanning. Custodian does, fully offline and self-hosted, so no member's file is ever uploaded to a third party. It reads the bytes, catches what a filename hides, and quarantines the sender to Restricted while the rest of the server never sees it.
Containment is only half the job. The other half is knowing who you're dealing with and being able to show your work afterward.
.contain profile @user is the primary lookup: identity and account age, membership and the invite they joined with, last-seen and active channels, and a Risk panel of reports, containments and filter hits, pulling in cluster links and the member's full activity trail. An on-demand .contain secscan audits the server's permissions, channels and verification for the gaps scammers exploit.
.contain profile.contain activity.contain secscanEvery action is categorised and hash-chained, so the audit trail can be proven intact, not quietly edited after the fact. Per-category routing, a timeline, stats, and high-severity alerts when something serious happens.
.modlog.modlog timeline.modlog statsNo dashboard onboarding maze. Add the bot, run one setup command, and tune the rest from a settings panel inside Discord.
The invite asks only for the permissions the bot actually uses: View Audit Log, Manage Roles, Moderate Members, and the like. Never Administrator.
.initOne guided command builds the Contained role, the tank category and channels, the escape-room thread, the moderator channel, and routed mod-log channels.
Open .settings for a live Components V2 panel (thresholds, account-age gating, dehoist mode, anti-raid and anti-nuke, retention), then run .contain secscan to audit the server for the gaps scammers look for.
A few of the communities using Custodian right now. See the full, security-ranked list in the finder.
Setup takes a couple of minutes. Most members will never see the scam attempts it stops.
