GuardAC reads the way a player moves their crosshair — angle, acceleration, jerk and mouse-sensitivity fingerprint — and a TCN neural network decides, in milliseconds, whether it's a human or an aimbot.
Cheats can fake a target, but not the physics of a human hand.
Per-tick yaw/pitch deltas plus their acceleration and jerk — the signature of real muscle movement vs. interpolated bot aim.
Real mice quantise to a sensitivity grid; aimbots compute in full float. The GCD error exposes the difference.
A causal Temporal Convolutional Network classifies whole combat windows, not single ticks — far harder to evade than thresholds.
A ping bucket feeds the model so laggy legit players aren't mistaken for cheaters.
Threshold tuned to a target false-positive rate, a per-player buffer, and an only-alert safety mode.
One Spigot/Paper/Folia plugin streams features here; this service returns a probability. No client mods.
Start free. Upgrade when your network grows. Pay via Telegram.
Point the GuardAC plugin at this service with your key.
Click “Get a free API key”, or buy a paid tier via Telegram. Copy the key — it's shown once.
plugins/GuardAC/config.ymlai: enabled: true server: "https://api.guardac.io" # только домен api-key: "YOUR_KEY_HERE" sequence: 40 # must match the trained model
Run /guard reload. Watch detections on your dashboard. Start in
only-alert mode, then enable punishments once you trust it.