Most hosting panels treat security as an upsell. You get a basic firewall, and anything past it is a paid add-on or a third-party agent you install on every box yourself. Unicorn Shield is the opposite: seven security surfaces, built into the panel, covering the whole fleet, included at every tier including the free one.
This guide walks through what each surface does and when you would reach for it. For the hardening checklist that sits alongside it, read the security checklist.
One page, every server
Shield's seven tabs each query every server in the fleet through the panel's own cross-server API, so what you see is the whole estate rather than one host at a time. Where a tab offers an action, that action still runs on the specific server that saw the event.
That matters more than it sounds. The usual failure mode of per-server security tooling is that nobody looks at server six. A fleet-wide view means an attacker probing three of your hosts shows up as one row, not three unrelated incidents nobody correlates.
The ban daemon
Underneath the Firewall tab is upcp-shield, a small daemon that runs on every server. It is not a single process: OpenRC's supervisor keeps a parent alive, and that parent runs one log tailer per jail, each piping into its own copy of the same matcher. A jail pairs a log file with a filter and its own thresholds.
Four jails ship by default:
| Jail | Watches | Trigger | Ban |
|---|---|---|---|
sshd | system log | 5 in 10 min | 24 hours |
upcp-nginx-malicious | proxy access log | 5 in 10 min | 6 hours |
upcp-wordpress | proxy access log | 5 in 10 min | 6 hours |
upcp | panel API log | 25 in 10 min | 6 hours |
The panel's own API jail is deliberately the most forgiving, because a logged-in operator clicking around generates far more traffic than a scanner probing one path.
Bans are written to both SQLite, so the panel can show history, and the live nftables ruleset, so the kernel enforces them. Every five minutes the daemon snapshots the ban set to disk and replays it on startup, so a reboot does not reset hundreds of bans to zero.
Tuning without losing your changes on upgrade. The shipped jails live in shield.conf, which an update overwrites. Anything in conf.d/*.conf is parsed afterwards with last-wins semantics and is never touched, so that is where local changes belong. Disable a built-in jail, override one threshold, or add a jail of your own against any log file on the box.
The WAF
The WAF reads the proxy's access and error logs, classifies common probe shapes, and turns them into events you can act on. It opens on a fleet overview: headline numbers, a map of attacker countries, an attack timeline and the top offending IPs.
It classifies seventeen categories, from SQL injection, XSS, path traversal and SSRF through .env and config probing, webshell fetches, RCE attempts and log4shell, down to low-signal things like scanner user-agents and rate-limit refusals. Severity runs from low to critical, and a row takes the worst of its categories.
Three details worth knowing:
Probing filters are status-gated. The WordPress, phpMyAdmin and admin-console filters only match requests that returned a 4xx or 5xx. A legitimate admin visiting /wp-login.php and signing in successfully never appears.
Bulk actions are scoped. Select any number of rows and ban them on just the servers that saw the attack, or across the whole fleet. The selection is pruned to visible rows automatically, so you cannot act on something you cannot see, and already-banned pairs are skipped.
Repeat offenders escalate. Shield keeps ninety days of ban history per IP. A first WAF ban lasts 24 hours, a second 7 days, a third 30 days, and a fourth or later is permanent. Each row carries a tier chip so a repeat offender is obvious at a glance. Manual bans do not escalate: they stay at exactly the duration you picked.
Malware scanning
Every scan runs two engines back to back, Linux Malware Detect and then ClamAV, so a file has to clear both to be classed as clean. The Shield Malware tab is the fleet roll-up: one row per site with its last scan, the result and the quarantined file count. The per-file detail lives on each site's own Malware tab.
Scanner behavior is configurable fleet-wide under Settings > Malware, or per server. You can toggle the engines independently, limit scans to each tenant's public folder, exclude paths, cap the maximum file size, choose flag-only or quarantine, nice the scanner, cap how many containers scan in parallel, and set a per-container timeout. Auto-delete is deliberately not an option: quarantine is reversible and gives you a chance to restore a false positive.
A false positive can be whitelisted, and the entry records both the signature and the exact file it fired on, so whitelisting one image does not excuse the same signature elsewhere on the site.
The audit
Shield's Audit tab runs CIS-style checks against every server, scores each host, and offers a one-click fix for the common failures. Checks cover SSH configuration, SSL expiry across every panel-managed certificate, kernel sysctl flags, SUID surprises, file permissions on sensitive paths, disk pressure and clock skew.
The headline is a Fleet Score, which is the pass percentage across every check and every host, with a per-server tile so you can see which box is the weak link.
Some checks fix themselves in one click: stepping the clock back to NTP time, renewing a flagged certificate, setting PermitRootLogin, writing a sysctl value, running pending package upgrades, or chmod-ing a file back to the hardened mode the panel ships. Where a check has more than one sensible remediation, the fix dialog offers a choice and defaults to the stricter one.
Some deliberately do not. The permissive-elevation check warns on genuinely password-less sudo or doas rules but never auto-fixes, because whether to keep, scope or remove a rule is a judgment call. The reboot-required check warns when a new kernel is installed but not running, and leaves the reboot to you.
Vulnerabilities
The Vulnerabilities tab reads what each host has pending and cross-references it against that distribution's own security database plus OSV. On Alpine that is apk and Alpine's data; on Ubuntu it is apt and Ubuntu's. Either way you get CVE identifiers, severity scores and a count of what upgrading actually closes.
Group it per server to plan one host's patching, or by package to plan a rolling upgrade of one package across the fleet. Filter by severity, search a specific CVE, or export the list as CSV for ticketing.
This is the difference between "there are 40 pending updates" and "three of these close filed advisories and the rest are version bumps."
File integrity monitoring
FIM builds a baseline hash of the panel's own code, the system binaries behind it and the security configuration, then re-hashes every monitored path hourly and alerts on any difference. Changes to high-value paths, SSH and sudo configuration, SUID bits, anything under /usr/local/bin, are flagged critical.
Tenant files are deliberately not watched here. Application files inside a customer's container are the malware scanner's job; FIM is about the platform underneath them.
Each event can be accepted, which endorses the change as the new baseline, or investigated, which clears it from the list without updating the baseline. Grouping by file rather than by event lets you resolve "the panel pushed a new binary to every host" in one click.
Threat intelligence and process anomalies
The Threats tab drops connections to known-bad addresses at the firewall using nftables sets maintained from upstream feeds. Four are on by default: Spamhaus DROP and DROPv6, abuse.ch's Feodo Tracker, and ThreatFox. Three more are available but off: DShield, Blocklist.de and the Tor exit-node list. Each feed has a scope selector for whether its entries apply to outbound traffic, inbound, or both.
The other half of the tab scans running processes for post-exploit shapes: a shell holding a network socket spawned by PHP-FPM or nginx, curl or wget piped straight into a shell, or a cryptominer's stratum connection. Hits land in the same timeline as the threat-intel drops.
What this replaces
A typical hardened host runs fail2ban, a malware scanner, AIDE or Tripwire, a CIS benchmark script, something to track CVEs, and an ipset-based blocklist updater, each configured per server and each reporting separately. Shield is those six jobs, configured once, reporting across the fleet, with the actions wired to the panel that already knows which server owns which site.
It is also free at every tier, including the free one. There is no security edition.
Frequently asked questions
Is Shield a role I enable? No. The ban daemon is an OpenRC service on every host, not a role, so it is managed on the server rather than in the panel. The tabs read the panel API, so they report on a server whether or not the daemon is running on it.
Can I stop it banning something? Addresses that must never be banned go in an always-ignore list, one IP or CIDR per line. That merges into an allowlist that already covers Cloudflare's published ranges, verified search crawlers and your own fleet addresses.
Does any of this phone home? No. Threat-intel feeds are pulled from their upstream sources, and the signature databases update from theirs. Nothing about your sites leaves your servers.
Does it work on a single-server fleet? Yes. The tabs open on the fleet overview either way; with one server the fleet is one server.
Security that is on by default
Shield is included, enabled, and fleet-wide from the first install. Try the live demo or read the Shield documentation for the full reference.