Settings & system
The Settings page — platform-wide configuration, launch-mode policy, security, hardening, backup, and governance defaults — and the System Check page that probes every dependency the control plane relies on.
Settings & system
Two operator pages sit at the base of god mode: Settings (/settings), the house
policy every tenant and server inherits, and System Check (/system), a live probe
of every connection the control plane depends on. Both require the settings.manage
privilege; without it each renders a locked state. See the
god-mode overview for how that gate works.
Settings
The Settings page is the policy of record for the platform. Most rows on it are
display-only — they state the current defaults rather than exposing a mutable toggle
— and a closing callout says so explicitly: mutating governance policy is gated behind
settings.manage and an append-only audit trail, with tenant-level overrides living
with each tenant.
The one live control on the page is Harden node in the Firewall section (below).
Launch-mode policy
The control plane presents one of three surfaces, narrowing what's exposed at each step:
| Mode | What it exposes |
|---|---|
| God (default) | Full super-admin surface — every section, every override, simulation and impersonation. Root-only. |
| Beta | Customer-facing surface with god-only sections hidden — for dogfooding the tenant experience. |
| Production | Locked-down surface — god tooling suppressed, standard RBAC fully enforced. |
Settings shows these as the policy of record; the live switch lives in the top
bar. Switching modes is written to the audit log with the acting admin and
IP, and the mode gates which godOnly nav items appear (see
god mode → session context).
Security
The Authentication & sessions card states the fail-closed security posture:
| Setting | Policy shown |
|---|---|
| MFA | Opt-in per admin. The page presents mandatory MFA as policy, but the control plane does not currently block a non-enrolled admin from signing in — enrollment is per-operator (TOTP) |
| Session TTL | Admin sessions expire 12 hours after sign-in (SESSION_TTL / SESSION_MAXAGE) — a fixed lifetime from issuance, not an idle timeout |
| Impersonation duration | Default 30 minutes, selectable when you start an act-as session; enforced server-side by an expiresAt on the impersonation token, which auto-revokes on expiry |
| Root super-admin | The root operator's email is the immutable root — cannot be demoted or removed, re-verified every request |
The impersonation default here is what backstops the per-session durations chosen on the Identity page's impersonation panel. MFA is presented as platform policy but is not enforced at sign-in today; each operator's own enrollment lives on Account → Security.
Firewall & hardening
The Host firewall card carries the page's only live action: Harden node. It
applies a ufw posture to a target node through the agent (v0.4.14+). The rows describe
what that posture is:
| Row | Policy |
|---|---|
| Default policy | Deny all inbound, allow outbound (the agent is pull-only). Applied and enabled atomically. |
| SSH safety | The chosen SSH port is always opened before the firewall is enabled, so the operator can never be locked out. |
| Default allowed ports | The managed proxy's 80/443 are open by default; add WireGuard (51820/udp) for mesh nodes and any app ports per node. |
This is the fleet-wide firewall control; the per-server, one-click hardening story (with the Security Score and armed auto-revert) is documented in Hardening your server.
Backup defaults
House policy applied to every newly captured image unless a tenant policy overrides it:
| Setting | Default |
|---|---|
| Default retention | "keep-all" — images are never pruned by count; lifecycle is driven by daily/weekly/monthly tiers, not deletion |
| Default storage backend | Cloudflare R2 (zero-egress object store) as the restic backend; tenants may bring their own |
| Auto restore-drill cadence | A managed restore drill runs monthly to the playground target, proving every image is recoverable |
| Encryption at rest | Every image is encrypted before it leaves the source host; per-(org, brand) keys, zero-knowledge custody |
Governance
The Audit & data residency card states the compliance posture:
| Setting | Policy |
|---|---|
| Audit retention | Every privileged action is retained for 13 months |
| Append-only enforcement | The audit log is write-once — entries can never be edited or deleted, including by the root super-admin |
| Data residency | Backups and metadata stay within US-East by default; per-tenant residency overrides are honored at capture time |
| Key custody | Encryption keys are held zero-knowledge — the control plane never sees plaintext tenant keys, even in god mode |
System Check
The System Check page probes every external dependency the control plane talks to
and reports a uniform status, so a single page can confirm the whole platform is wired
— instead of discovering a dead integration only when a feature silently no-ops. Each
probe is defensive (its own timeout and try/catch), so one broken dependency can't take
the page or the other checks down. The page is force-dynamic and re-runs on refresh.
At the top, a summary bar shows the overall verdict — All systems wired or N connections need attention — with counts by status. Each check uses one of four states:
| Status | Meaning |
|---|---|
| OK | Configured and reachable/healthy |
| Warn | Configured but degraded, or an optional thing that's fine left unset |
| Fail | Configured but broken/unreachable — needs attention |
| Off | Not configured (a known gap; the hint says what to set) |
Failing and off checks show a hint — the exact env var or step to turn it green.
The overall report is ok only when no check is failing.
Checks are grouped:
| Group | Probes |
|---|---|
| Core infrastructure | Database (Postgres), Migrations (with drift detection), Rate limiting / cache, Encryption key |
| Security & licensing | License signing key, License allow-list, Session secret, Web-SSH token secret, Google SSO |
| Fleet & agents | Agent fleet heartbeats, Hostinger VPS API, Hostinger provisioning, Coolify (legacy) |
| Networking & DNS | Cloudflare DNS, Public exposures (cloudflared), Web-SSH relay, DNS/uptime monitors, Transfer relay |
| Communications & alerts | Outbound email (SMTP), Alert channels |
| Billing & provisioning | Stripe payments (+ webhook verification) |
| AI & automation | Fleet Copilot (Claude), Browser worker |
A few checks are worth understanding:
- Database confirms Postgres is reachable and that core tables
(
admins,agents,jobs,deployments,exposures,tenants) are present. WhenDATABASE_URLis unset, the whole platform runs on in-memory demo data — this check reads Off and everything downstream is seeded, not live. - License signing key is a Fail when unset, because without it agents cannot activate (the activate endpoint returns 503).
- License allow-list (
HOSTSSH_LICENSE_KEYS) is a Warn when unset — any well-formed key may activate; setting it gates which keys are allowed to. - Coolify unconfigured reads Off by design — it's the vendor the platform is exiting, so "not wired" is the goal, not a failure.
- Cloudflare DNS actually calls the Cloudflare API to confirm the scoped
Zone:DNS:Edittoken is valid and reports how many zones are in scope.
Related
- God mode (admin overview) — the RBAC and session model these pages sit on
- Identity & access — where the impersonation ceiling and MFA enforcement described here are exercised
- Hardening your server — the per-server, one-click hardening surface
- Audit log — the append-only trail these governance defaults reference
- Licensing, seats & transfers — the license model the signing key and allow-list guard