Перейти к содержимому
Mineforgian

OverWatch-ML

OverWatch-ML is a next-generation cheat detection system designed for modern Minecraft servers.

Загрузки
55
Подписчики
0
Обновлён
18 марта 2026 г.
Лицензия
MIT

Опубликован 3 марта 2026 г.

OverWatch-ML Banner

OverWatch-ML

Next-Generation Machine Learning Anti-Cheat for Minecraft

Modrinth GitHub Discord License

Paper Java ProtocolLib


OverWatch-ML is a next-generation anti-cheat for Paper 1.21+ that combines classical heuristic analysis with trained Machine Learning models to catch what traditional anti-cheats miss.

Most anti-cheats look at individual packets. OverWatch-ML looks at behavior — building a persistent profile for every player and analyzing patterns across Combat, Movement, Xray, and World Interactions using dedicated ML models.


How It Works

At its core is a Detection Orchestrator — a centralized engine that weighs signals from internal ML analysis alongside data from external anti-cheats (Grim, Vulcan, Matrix, Spartan). No single flag triggers a punishment. The system reasons about the full picture.

Player Actions → Behavioral Profiler → ML Models → Orchestrator → Verdict
                                                 ↑
                                    External AC Scores (Grim, Vulcan...)

Features

🧠 Machine Learning Detection

Every player builds a persistent behavioral profile over time. CPS, movement speed, mining efficiency, and dozens of other signals are tracked continuously and fed into dedicated ML models for Combat, Movement, Xray, and World Interactions.

📡 Detection Orchestrator

A centralized "brain" that coordinates internal ML analysis with violation scores from external anti-cheats. Intelligent signal weighting means fewer false positives and smarter verdicts.

💻 Staff Control Panel

A fully GUI-driven management interface. Monitor flagged players, stream live behavioral data, and issue punishments — no config editing required.

🔔 Real-time Staff Alerts

Instant notifications with one-click teleportation to the suspect. Your staff acts fast, every time.

🛡️ Smart Mode Detection

Built-in bypass logic for Creative and Spectator mode players. Zero false positives from your own team.

⚡ Zero TPS Impact

All ML inference runs fully asynchronously on Java 21. Heavy computation never touches the main thread.


Integrations

OverWatch-ML is designed to complement your existing anti-cheat stack, not replace it.

GrimAC Vulcan Matrix Spartan


Requirements

Requirement Version
Server Software Paper / Spigot 1.21+
Java 21 or newer
Dependency ProtocolLib

Commands & Permissions

Command Description
/owml Open the Staff Control Panel
/owml player <name> View a player's ML profile
/owml staff Open staff management menu
Permission Access
owml.staff Staff menu and live alerts
owml.admin Full configuration and ML system access

Compatibility

Minecraft Version Status
1.21.x ✅ Fully Supported

Developed with ❤️ by the OverWatch Team · Powered by Machine Learning

Support on Ko-fi Join Discord

Ченджлог

2.2.1Релиз1.21.9, 1.21.10, 1.21.11 · 18 марта 2026 г.

[2.2.1]

Fixed

  • XRAY False Positives: Improved XRAY detection to properly account for tool efficiency (efficiency enchantment) and Haste potion effects. Players with Efficiency V pickaxes and Haste II are no longer falsely flagged.
  • TPS Measurement: Fixed critical TPS calculation error that showed incorrect values (1.0 TPS) due to custom calculation issues. Now uses reliable Bukkit.getTPS() API.

Features

  • Flag-Only Mode: Added ml.flag-only config option. When enabled, ML detections only notify staff without applying automatic punishments. Useful for testing or servers that want manual punishment review.

Changed

  • Config Cleanup: Removed unused configuration sections (web admin panel, advanced-detection, online-learning, performance-monitoring, audit-logging) and added missing options. Config is now cleaner and more organized.
2.2.0Релиз1.21.9, 1.21.10, 1.21.11 · 17 марта 2026 г.

[2.2.0]

Fixed

  • Duplicate Initialization: Fixed multiple objects being created twice during plugin startup in OverWatchML.java.
  • Thread Safety: Replaced HashMap with ConcurrentHashMap in DecoyManager for thread-safe operations.
  • Debug Performance: Removed duplicate debug checks in BlockListener that were executed on every block break.

Optimized

  • Performance Monitoring: Changed PerformanceMonitor interval from every tick to every 20 ticks (1 second), reducing overhead by 95%.
  • Detection Throttling: Added cooldown (2 seconds) to DetectionOrchestrator to prevent excessive detection runs on high-activity servers.
  • GUI Auto-Refresh: Removed auto-refresh task in GUIManager that was causing unnecessary load with many players.
  • MainHubGUI Caching: Implemented 2-second caching for TPS, ML stats, and alert data to reduce refresh overhead by 80%.
  • Metrics Collection: Consolidated 4 separate scheduler tasks in RealtimeMetricsCollector into one task (every 10 seconds), reducing overhead by 75%.

Removed

  • Web Admin Panel: Completely removed the web admin panel and related commands for better security and reduced dependencies.

Комментарии

Загружаем…