What changed, in plain English.

The fleet page now opens with a "needs attention" list: every device with a problem, the reason in plain text ("disk 89% (F:)", "offline — last seen 3h ago"), most urgent first, one click from the device itself. No more scanning site cards hunting for the yellow cell — if the band isn't there, everything is healthy. And every device now answers the question its uptime raises: a new "last boot" line says why the machine last restarted — a planned Windows update, a person, or an unexpected loss of power. That last one matters: a dirty shutdown also appears in the attention list for its first day, because a server that lost power is a server worth a second look.

For about three hours early this morning, Arqos sent "offline" alerts for machines that were running fine. No monitored system was ever down: an update had idle computers reporting an activity number so close to zero our database refused to store it, and the whole check-in was rejected with it. To Arqos, everyone went silent at once. We found the exact error in the logs, fixed it the same morning, and every false alarm closed itself. The structural change: a device's "I'm alive" signal is now recorded before — and independent of — its measurements. A bad statistic can degrade a chart; it can never impersonate an outage again.

Hypervisors, switches, and storage appliances don't allow software installs — so any Arqos agent at the same site now quietly polls them for you (ping and port checks, every minute). There's nothing to assign or babysit: agents share the work as a pool, and if one goes away another takes over on its own. If the site ever loses all eyes, you get one honest "monitoring blind" notice instead of a wall of false alarms.

You can route notifications by site and severity, every incident stays in one email thread from first alert to all-clear, and anyone can acknowledge with a single click — no sign-in needed. If a critical alert sits unacknowledged, Arqos walks it up the chain you define, one contact at a time.

Rebooting a server on purpose? Snooze its alert emails for an hour, four hours, or until 8am — right from the alert email itself. Detection never pauses and the record never has gaps; only the emails go quiet. If the machine is still down when the snooze expires, you get one final alert.

A disconnected mapped drive or a stuck Windows subsystem could previously freeze an agent mid-collection — alive as a process, silent as a reporter. Every system call the agent makes now has a deadline, and a watchdog restarts the agent cleanly if a collection cycle ever stalls.

Devices are now grouped by site with problems sorted to the top, healthy fleets collapse into a one-line "all quiet" strip that shows your last incident and how fast it recovered, and a quick switcher (⌘K) jumps you to any device by name.

The agent has no remote-control channel — not a hidden one, not a disabled one. It can read health numbers and send them out; it cannot receive commands, run scripts, or update itself without your administrator acting first. That's an architecture decision, not a policy, and it's why a compromise of our service can't become a compromise of your network.

One command installs the agent on a server, and from then on it checks in every minute over an encrypted outbound connection. If a machine goes silent, you get an email within about two minutes — and when it comes back, the alert closes itself and tells you. Setup is an afternoon; the day-to-day is silence.