Shepherd

AutoMod reads text. Scammers post screenshots.

Every crypto scam on Discord arrives as an image: fake exchange balances, forged verified posts, giveaway announcements, payment QR codes. There are no banned words to filter and no links to block, because the entire message lives inside the picture. Your AutoMod rules never see it.

Shepherd reads the image itself. When it recognises a scam it bans the account, clears the message, and posts the evidence in your log channel. It never waits for approval, which is the whole point: scams land at 3am, and that is exactly when nobody is watching.

━━━━━━━━━━━━━━━━━━━━

WHAT IT CATCHES

• Fake exchange, wallet and casino screenshots
• Giveaways and airdrops promising to double your money
• Forged verified posts impersonating well-known accounts
• Fabricated profit and withdrawal proofs
• Payment QR codes and wallet addresses
• Recruitment for Telegram and WhatsApp signal groups

━━━━━━━━━━━━━━━━━━━━

HOW IT DECIDES

Four stages, cheapest first.

1. Local filters — the roles and channels you exclude 2. Shared network — the poster was already caught on another protected server 3. Image fingerprint — this exact picture already has a verdict elsewhere 4. AI vision — only for images nobody has seen before

Scammers repost identical images across dozens of servers, so most blocks resolve in the first three stages at no cost to you. The first server to meet a new scam protects everyone else in the network.

━━━━━━━━━━━━━━━━━━━━

AN ASSISTANT FOR THE QUESTIONS YOU'VE ANSWERED A HUNDRED TIMES

Every community has ten questions asked every single day. On the Pro and Network plans, Shepherd answers them from a knowledge base you write, in a character you choose, in whatever language the member used.

It answers only from your entries. Ask it something you haven't written down and it says so, in an exact sentence you set, instead of inventing a confident wrong answer — which is what makes most FAQ bots worse than having none.

Write entries in the dashboard or paste an existing spreadsheet as CSV. The two halves are independent: run the assistant on its own if that's all you need.

━━━━━━━━━━━━━━━━━━━━

YOU STAY IN CONTROL

Choose the punishment — ban, kick, timeout, delete the message, or nothing at all — plus the log channel, the confidence threshold, and the roles and channels to skip. Everything is configured from a web dashboard after logging in with Discord.

Every report carries an Undo button. One click unbans the member and pulls that image out of the shared network, so the same mistake cannot repeat anywhere else. Server staff and anyone ranked above the bot are always reported but never sanctioned automatically — a moderation bot should not be able to ban the owner.

Prefer to watch before you trust it? Watch-only mode reports everything and touches nobody.

━━━━━━━━━━━━━━━━━━━━

COMMANDS

/shepherd — status of both modules, plan, usage and settings /scan <message_id> — check a specific message by hand

━━━━━━━━━━━━━━━━━━━━

PLANS

Free blocks every scam the shared network already recognises, without limits and without charge, because a database lookup costs us nothing. It does not read new images and never takes action on its own.

Guard · €4.99/month per server adds AI analysis of images nobody has seen before, and the automatic bans, kicks and timeouts.

Pro · €6.99 adds the FAQ assistant, checking of single images rather than pairs, CSV export and webhook alerts.

Network · €9.99 raises every limit and re-checks borderline calls with a stronger model.

No OpenAI key required. The analysis is included in the price.