Connectivity is one observation

Checking in means the server has recently heard from the device. It is not an assertion that the device, its applications or every service are healthy. An online device can also have low disk space, a CPU incident or a selected service stopped.

After more than three minutes without a received check-in, an enrolled device can become Offline on evaluation. This reports the missing check-ins; it does not diagnose power, sleep, routing or a collector failure. Devices that never reported remain Waiting for first check-in.

Open the history

Open a machine for CPU usage, memory used and system-disk usage. Choose 1 hour, 24 hours or 7 days. The server retains at most one sample per device per minute for seven days. Longer views use averages: five-minute buckets for 24 hours and thirty-minute buckets for seven days.

Averaging can hide a short peak. Alert timing evaluates accepted check-ins independently of the chart's averaged points. Gaps remain gaps, not zero utilization. History begins when samples arrive; it is not backfilled from before enrollment.

Interpret CPU and memory

CPU uses changes between samples, so the first value after collector startup can be absent. Linux memory is total memory minus the kernel's estimate of available memory. Mac and Windows use platform-specific accounting. These are host-wide observations, not per-process diagnostics.

Use a fresh reading and a comparable time range when investigating. A retained reading from before an outage is historical evidence, not current utilization. Resource data alone does not identify the process responsible for a condition.

System disk and inventory

Disk reporting covers the system filesystem; it does not enumerate every volume. Low-disk incidents are distinct from connectivity. Inventory includes hostname, OS/version, kernel where available, architecture, logical CPU count, total memory and last boot. The boot timestamp comes from the device clock.

Inventory is basic device information, not a complete software inventory. It does not claim installed applications, serial numbers, user enumeration or file contents. The portal labels when facts were reported so an old value is not silently treated as fresh.

Common checks

  • No CPU after installation: wait for the second check-in and confirm a resource-capable agent.
  • Chart looks quiet but an alert opened: compare the incident's triggering value with raw timing, not only a long-range average.
  • Old disk value on an offline device: restore current check-ins before deciding space is healthy.
  • Unexpected inventory: confirm the device identity and timestamp; never copy enrollment between cloned machines.

See the Linux kernel /proc documentation for the underlying Linux measurement sources.