Q&A

A confidence-based, packet-level anti-cheat for Paper/Purpur that adds weighted evidence instead of banning on a single flag — 93 checks across combat, movement, packets, and world/economy protection.

Yes. Free gives you 10 core checks (KillAura, Reach, Speed, Flight, NoFall, Scaffold, FastBreak, AutoClicker, Timer) with the same confidence-based scoring, no payment or key required. Pro and Ultimate unlock all 93 checks and the full packet engine.

The packet engine is built on PacketEvents and runs alongside your server's normal event loop rather than replacing it. Timing-sensitive checks automatically suspend below a configurable TPS floor (17.5 by default) specifically so Sentinel never becomes the reason your server is laggy.

For Bedrock players connecting to your Java server through Geyser, yes — detected automatically, checks scaled to their input. Sentinel cannot install on a native Bedrock Dedicated Server or an Xbox-hosted Realm; no third-party anti-cheat can, since those platforms have no plugin system at all. Full detail in the Bedrock & Geyser docs.

Paper and Purpur, 1.21 and newer, on Java 21. Plain Spigot/CraftBukkit isn't supported — a few checks (including anti-xray) use Paper-only APIs. Fabric/Forge servers are a different plugin ecosystem entirely and aren't supported either.

Architecturally, not by luck. Every check adds weighted, decaying evidence to a hidden suspicion score rather than punishing on one flag. Kick, tempban, and ban have separate, escalating thresholds. Reaching the permanent-ban threshold additionally requires a second-opinion corroboration gate — two independent checks agreeing, or one check clearing a high-confidence bar — otherwise it's downgraded to a tempban for staff review instead.

Timing-sensitive checks suspend entirely below the configured TPS floor, widen their tolerances above a soft-ping threshold, and exempt outright above a hard-ping ceiling. Movement checks also grant grace after joining, teleporting, taking knockback, or dying — a lag spike or a respawn can't look like a hack.

Any check can run observe-only: it records evidence and a replay clip exactly as if live, but adds zero suspicion and can never cancel an action or punish. New or materially changed checks ship shadowed until they've been proven against real traffic. /ac shadow shows what's currently shadowed and how often it would have flagged.

Yes — touch and controller input genuinely differs from mouse and keyboard, so every violation a detected Bedrock player generates is scaled down (0.7× by default) before it reaches their suspicion score. This can only make Sentinel more lenient toward them, never more aggressive.

Every punishment comes with an appeal code and a full evidence trail. Ask the dashboard's Ask Sentinel chat to explain <player> for the exact, chronological timeline — time, check, detail, suspicion delta, running total, and the replay clip that covers it. It's read live from the plugin's own record, not summarized by the AI, so it's accurate enough to settle an appeal.

Drop Sentinel.jar into plugins/ and restart — no extra dependencies, PacketEvents is bundled. Host-specific click-by-click steps (Aternos, Exaroton, Apex, Pterodactyl- and Multicraft-panel hosts, your own VPS, Docker) are in the install guide.

No — the shipped defaults are conservative and safe out of the box. If your server has an unusual play style (lifesteal, hardcore, skyblock, prison, anarchy, and others), a matching config preset retunes lag tolerance and punishment pacing in one command without ever weakening detection.

Yes, per-check, per-world, or scoped to a cuboid region — useful for a minigame arena inside an otherwise-protected world. See checks.<id>.* in the configuration docs.

Free — 10 core checks, no key needed. Pro ($4.99/mo) — all 93 checks, client fingerprinting, player records, 1 server. Ultimate ($12.99/mo) — everything in Pro plus the live web dashboard, remote moderation, the AI assistant, Discord webhooks, and up to 10 servers. Pro and Ultimate share identical detection — Ultimate adds the tools to run it from anywhere.

Run /sentinel activate <your-key> in console after checkout. The key is bound to that server and unlocks instantly — no restart needed.

Yes, no phone call or email required — open the "Manage subscription" link in your license email for Stripe's self-serve billing portal. Cancelling stops future billing; you keep access for the rest of the period you already paid for.

If Sentinel doesn't work on your server and can't be made to after a good-faith support attempt, email us within 14 days of your first payment for a full refund. Monthly renewal charges aren't refunded once processed — cancel before the next renewal instead. Full terms on the Terms of Service page.

Watch live risk/detections from any browser, moderate remotely (kick/tempban/ban/unban/freeze/broadcast), toggle checks or safe config paths, and chat with an AI assistant grounded in your server's own data. The plugin dials out to the dashboard on a short timer — nothing ever connects in, so there's no inbound attack surface.

It's explicitly instructed never to, and reasons only over your server's real submitted data — but it's still a language model summarizing that data conversationally. For anything that needs to be exact (an appeal, a ban explanation), ask it to explain <player>, which is answered directly from the plugin's stored evidence log, not paraphrased.

No. Remote actions are a fixed whitelist (kick, tempban, ban, unban, freeze, notify, broadcast, toggle a check, a safe subset of config, manage banned words) re-validated on the plugin side. There is deliberately no "run any console command" action, so a leaked dashboard key can only perform bounded, reversible moderation.

In your own local SQLite database on your server. IPs never leave your machine on their own — the dashboard only ever sees what your server chooses to push to it, and IP addresses are masked there by default.

No. Checkout redirects entirely to Stripe, a PCI-DSS compliant processor. Card numbers never reach sentinel-anticheat.info and are never stored on it.

Stored as a salted SHA-256 hash by default — enough to detect "same IP," not enough to recover the address. The salt is generated locally and lives only in your plugin data folder, never in config.yml or git. The feature itself is purely informational and never bans, kicks, or adds suspicion on its own.