Monitoring and services
How live domains are checked every hour.
How monitoring works
Every live domain is checked once an hour against the records it is held to: the records a rail wrote, or, for a domain connected by hand or with Domain Connect one-click, the records verified before it went live.
- Drift is confirmed first. A record that looks missing or changed is
looked up again at the zone's own nameservers, with no resolver cache in
between. A lookup that does not answer is
inconclusiveand never counts as drift. - Missing records are put back when CustomDomain™ holds continuous access
(the zone it hosts, a kept sign-in or Domain Connect authorization, a
continuous provider account) and the
deployment has automatic repair switched on. The action is
auto_fixed, orfix_failedwhen it did not work or did not hold. - Only what disappeared, only what CustomDomain™ wrote. A repair writes back the record sets that went missing, from the records CustomDomain™ wrote for the domain; a kept Domain Connect authorization re-applies its template, the only write it allows. A domain whose records CustomDomain™ did not write is left for you.
- Only new drift. A repair runs only when the records disappeared after a check found them in place. Records that were already missing when monitoring first looked are left for you.
- Bounded. Repairs are paced and capped per hourly check and per provider
account, and every one is in the audit log. After 3 repairs of a domain
within 24 hours the monitor stops and reports
fix_failed: something keeps removing the records, and that needs a person. - A changed record is never overwritten. Pointing a record somewhere else
is often deliberate, so the action is
action_required. - Without continuous access, or where automatic repair is switched off,
drift is
action_required.repair.pausedistruewhen CustomDomain™ holds the access but repair is off, andrepair.detailsays what would repair it.
Read the state
GET /v1/connections/{id}/monitor:
{
"watched": true,
"every": "hour",
"last_checked_at": "2026-09-29T12:00:00Z",
"result": "drift",
"drift_since": "2026-09-29T11:00:00Z",
"drift": [ { "subdomain": "@", "type": "CNAME", "verdict": "record_missing", "expected": ["edge.customdomain.ai"] } ],
"action": { "kind": "auto_fixed", "description": "CustomDomain™ put the missing records back." },
"repair": { "automatic": true, "via": "provider_account", "detail": "..." },
"watching": [ { "name": "shop", "type": "CNAME", "value": "edge.customdomain.ai" } ],
"history": [ { "kind": "drift.found" }, { "kind": "drift.fixed" } ]
}GET /v1/monitor/domains (API key) lists the domains of the workspace with
their state, the ones needing attention first, a page at a time: limit
(200 by default, 1000 at most) and cursor, the previous page's
next_cursor, which is absent on the last page. summary counts every
domain, and auto_repair says whether the deployment puts missing records
back.
To check a domain right now instead of waiting for the hourly sweep, call
POST /v1/monitor:check with { "domain": "app.customer.com" }. Without a
baseline it compares live DNS with the connection's own records; send a
baseline array of { subdomain, type, values } to compare against records of
your choosing. It answers { domain, drift, records }, where each record has
subdomain, type, verdict, drift, expected and observed. It reads DNS
and changes nothing.
Monitoring is on for every connection unless you create it with
"monitor": false (Create a connection).
Alerts
domain.record_missing and domain.record_restored carry a drift object
with the record, the expected and observed values and what was done. Owners and
admins get one email per drift episode that needs them. Both are delivered where
MONITOR_ALERTS_ENABLED=1; the console always shows the state.
Services per domain
GET /v1/connections/{id}/services lists every CustomDomain™ service with its
state for the domain (active, pending, attention, paused for automatic
repair that is switched off, inactive): DNS
connect (with the rail that wrote the records and the access held), managed
DNS, edge serving, the HTTPS certificate, Power, monitoring, automatic repair,
redirects, email records and CAA. GET /v1/services:summary (API key) counts
them per application and for the workspace, and names the domains that need
attention (needs_attention).
Edge serving is attention for a live domain whose records point at the edge
when no origin is set for it (its own, or its application's default) and it
has no redirect: the edge answers such a domain with an error page, so it is
live, but has nothing to serve. Set an origin, or point the domain at your own
servers, and it is active again.