LumoAiDetector

LumoAiDetector An experimental, free anti-cheat that runs local machine learning inside Java. Instead of using pre-trained files or connecting to external Python APIs, you record combat data from your own players and build a model tailored to your specific community. This plugin is free and open-source because I want server owners to be able to test this approach without paying for closed systems. I do not provide a pre-trained model. You need to train your own, since a model trained on one setup is often useless on another due to different pings, DPIs, and PvP styles. How it works The plugin ignores random movement and watches only combat rotations. A data window reaches the model only when all criteria pass: * The player recently attacked or swung their hand. * A living target is nearby and inside their aim area. * The mouse moved enough across a 15-tick window. * Network timing stays within a normal range. Each valid window becomes 120 numeric features (15 ticks * 8 values: dx, dy, dt, v, a, j, err, derr). The local Smile Random Forest model then classifies this data as legit (0) or suspect (1). Core Features * Zero dependencies: Runs on Java 8+ without requiring NMS or ProtocolLib. Works on Spigot, Paper, Purpur, and Folia from version 1.8 to modern releases. * Async training: The training process runs in the background and will not freeze your server. * In-game management: Activate, deactivate, or restore deleted models from backups using admin commands. * Customization: Completely editable configs, messages, and threshold settings. Auto-punishments are supported but disabled by default. Commands