YadduGuard is an advanced AI-powered chat moderation plugin for Velocity proxy servers. It uses cutting-edge AI models to detect toxic messages, Hindi/Hinglish gaaliyan, English abuses, harassment, and hate speech in real-time — and automatically punishes offenders with escalating consequences.
AI Providers Supported
Groq (LLaMA 3.3 70B) — Free, ultra-fast
Google Gemini 2.0 Flash — Free tier available
OpenAI GPT-4o Mini — Premium option
Auto-fallback: if one provider fails, automatically switches to next
How It Works
Every chat message passes through a 3-layer system:
QuickFilter — 200+ normalized gaali patterns checked instantly (no API call)
Message Cache — repeated toxic messages blocked from cache (no API call)
AI Check — only suspicious messages sent to AI for deep analysis
This means minimal API usage while maximum detection accuracy.
Escalating Punishments
Severity 3-4 → Soft Mute (5 minutes)
Severity 5-6 → Kick
Severity 7-8 → Temp Ban
Severity 9-10 → IP Ban
3+ violations in a session → automatic escalation
Smart Detection
Understands context — "mc" alone is fine, "mc tu" is toxic
Detects leet speak — b s d k, bh0$d1k, b.s.d.k all caught
Hindi, Hinglish, and English detection
Works across all backend servers simultaneously
Admin Commands
Powered by YadduNetwork
AI Providers SupportedGroq (LLaMA 3.3 70B) — Free, ultra-fast
Google Gemini 2.0 Flash — Free tier available
OpenAI GPT-4o Mini — Premium option
Auto-fallback: if one provider fails, automatically switches to next
How It WorksEvery chat message passes through a 3-layer system:
QuickFilter — 200+ normalized gaali patterns checked instantly (no API call)
Message Cache — repeated toxic messages blocked from cache (no API call)
AI Check — only suspicious messages sent to AI for deep analysis
This means minimal API usage while maximum detection accuracy.
Escalating PunishmentsSeverity 3-4 → Soft Mute (5 minutes)
Severity 5-6 → Kick
Severity 7-8 → Temp Ban
Severity 9-10 → IP Ban
3+ violations in a session → automatic escalation
Smart DetectionUnderstands context — "mc" alone is fine, "mc tu" is toxic
Detects leet speak — b s d k, bh0$d1k, b.s.d.k all caught
Hindi, Hinglish, and English detection
Works across all backend servers simultaneously
Admin Commands
Code:
/yg status — View plugin status
Code:
/yg reload — Reload config live
Code:
/yg test — Test AI on any message
Code:
/yg unmute — Remove soft mute
Code:
/yg violations — View player historyPowered by YadduNetwork
