> ## Documentation Index
> Fetch the complete documentation index at: https://docs.myprotect-bot.fr/llms.txt
> Use this file to discover all available pages before exploring further.

# Vision AI & Anti-Scam Protection

> Deep visual and OCR inspection of image attachments.

# Vision AI & Anti-Scam Protection

Modern threat actors have evolved beyond text-based spam: they distribute malware, session hijackers, and phishing portals through **image attachments**, embedded QR codes, and fake Discord Nitro promotions that bypass standard keyword filters. MyProtect deploys a dedicated asynchronous **Vision AI pipeline** to neutralize visual threats before users can click them.

***

## Visual Phishing Vectors

> [card] Fake Discord Nitro / Steam Gifts
>
> Scammers post promotional banners promising free subscriptions that mirror Discord's native UI. MyProtect analyzes layout geometry and brand imitation to catch phishing designs instantly.

> [card] Malicious QR Codes
>
> Attackers post Discord login QR codes disguised as verification or giveaway steps. Scanning them compromises the victim's token. MyProtect decodes and blocks all unauthorized login QR codes.

> [card] Crypto & AirDrop Lures
>
> Fraudulent investment schemes and fake token air-drops featuring prominent web3 logos and spoofed exchange domains are identified via specialized OCR models.

> [card] NSFW & Shock Images
>
> Adult content, gore, and disruptive spam images are classified and removed in compliance with Discord's Community Guidelines.

***

## Processing Architecture

To maintain zero impact on Discord message delivery speeds, image scanning runs asynchronously through a dedicated worker cluster:

```text theme={null}
[ User uploads image in #chat ]
              │
              ▼
   [ Gateway: messageCreate ]
              │
              ▼
   [ Enqueue to BullMQ Worker ]
   (divy-worker / myp-worker)
              │
     ┌────────┴────────┐
     ▼                 ▼
[ Tesseract OCR ] [ Vision Classifier ]
     │                 │
     └────────┬────────┘
              ▼
  [ Malicious Score > 0.85? ]
              │
             YES
              ▼
    [ Delete Message (Discord API) ]
    [ Timeout / Kick Offender ]
    [ Post Forensic Alert to Staff ]
```

1. When an attachment is posted, the bot immediately dispatches the image URL to the worker cluster via BullMQ.
2. The worker extracts text via high-throughput OCR and runs structural classification checks.
3. If an evasion pattern or phishing domain is identified, the message is deleted in under 800 milliseconds and the user is placed in temporary timeout to prevent further damage.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.