Scam filtering for Discord servers
The scam isn't a link.
It's a screenshot.
Most scam bots only match URLs, so a fake giveaway posted as an image walks straight past them. DONDA reads the picture itself, every word in it, and deletes the message.
Add DONDA to your serverFree, and scanning from the moment it joins. There's nothing to set up first.
OCR · Tesseract
- "OFFICIAL COMMUNITY EVENT"
- "$5,000 GIVEAWAY"
- "500 WINNERS TODAY ONLY"
- "CLAIM YOUR REWARD" flagged
- "CONNECT WALLET TO VERIFY" flagged
Links in message: 0
Verdict: scam · message removed
How it actually catches things
Five checks run on what gets posted. They don't depend on each other, and a scam only has to fail one.
-
OCR
It reads the image
Tesseract pulls the text back out of posted attachments, so a picture is judged on what it says. This is the part most scam bots don't do.
-
Image hash
Reposting it doesn't help
Report it once and DONDA knows it everywhere. Fuzzy hashing matches the picture, not the file.
-
Language
The wording gives it away
Keyword rules plus a Bayesian classifier trained on how these messages actually read. It runs locally on the bot, so nothing here goes to a third party.
-
Domain
One letter off still counts
Reputation checks, plus edit-distance matching against the domains people already trust. No blocklist entry needed.
-
QR
A QR code isn't a hiding spot
Codes buried in an image get decoded and checked like any other link.
On a hit, the message is deleted automatically. Repeat offenders can be softbanned too, off by default until you turn it on.
Protected the moment it joins
DONDA arrives with sensible defaults already switched on, so it's catching things before you've configured anything. There's no onboarding step to sit through.
When you want it tuned to your community, anyone with Manage Server can do it from inside Discord:
- /donda setup
- Pick the channel where catches get logged.
- /donda addkeyword
- Add phrasing specific to the scams your community gets.
- /donda adddomain
- Block a domain, or trust one your server actually uses.
- /donda config
- Toggle auto-moderation and softbans.
You don't have to wait on me to flip a switch for you.
What it doesn't see
You should want to know exactly what a bot like this looks at before you let it into your server.
What it reads
Attachments and messages that trip a detection rule: the image gets read, and anything text-like in it goes past the classifier and the domain checks.
What it never reads
Your ordinary conversation. Nothing that doesn't trip a rule gets parsed or stored, and none of it trains anything.
What I see
My own private dashboard shows which server, which channel, which rule fired, and when. Not the message. Not who sent it.
Your server is one screenshot away.
Add DONDA to your serverTakes about ten seconds: pick your server, confirm, and it's already scanning.