Start with the timeline

Record the last server-received check-in and the time the incident opened. Check whether the device ever reported. Waiting for first check-in, Offline and an unknown selected-service state mean different things.

Do not re-enroll a device simply because it is offline. Successful enrollment creates a durable identity. Replacing it can split history and hide the original problem.

Device is offline

  1. Confirm the device is awake and reachable through your normal administration path. A failed ping alone is not definitive; ICMP may be filtered.
  2. Check that the local monitoring service is running and enabled as intended.
  3. Read the companion's server-confirmed status, not only the operating system's process status.
  4. Inspect outbound HTTPS reachability to the configured application, certificate validation and the device clock.
  5. Review local logs for rejection or connection errors. Do not paste credentials or private state files into a ticket.

Network interruptions are retried. A revoked credential requires an administrator decision; it is not repaired by repeated retries. An unavailable application can affect many devices at once, so compare the incident timing across the fleet.

Device is online but a service is not

Check the exact selected name and the latest observation. Confirm the service exists and is expected to run continuously. On Windows use Service name rather than Display name. On Linux use a system .service unit, not a user unit or timer.

Unknown can mean access denied or an unsupported query. It is not evidence that a service stopped. If your administrator account sees Running but the collector sees Unknown, inspect query permissions rather than automatically elevating the agent.

Readings are missing

CPU needs a baseline sample after startup. Metrics and update reports can have different freshness rules. Historical chart gaps are retained intentionally. An older agent can check in successfully without supporting the newest report type.

Use Agent health to compare installed and available platform-specific versions. Use the application-requested agent updater after local helper setup, or the local verified update flow. Succeeded requires a check-in from the requested version; inspect the request result rather than assuming an installation finished.

The application address changed

If the administrator has confirmed a server-address move, an offline helper cannot receive a website update request from the old address. Open the new application’s platform download page and use its Copy repair command locally. The repair verifies the new destination with the existing CA, requires administrator authority and preserves identity and monitoring state. Do not invent a new origin, bypass TLS or re-enroll just to repair connectivity. Confirm fresh collector and helper check-ins afterward.

An alert was not sent

Check the rule source, enabled state, sustained duration, notification timing, maintenance and freshness. Review Notifications and attempt history. Muting offline does not mute resources or services. A canceled stage does not return when unmuted.

Only private development-inbox submission has been validated. Customer email is not configured. An accepted submission is not proof of inbox delivery, and an ambiguous attempt may need operator review.

Share useful evidence

Include the OS, agent version, approximate timestamps, visible state, expected behavior and redacted error text. Exclude enrollment codes, device credential files, private keys, personal account data and unrelated machine logs. Preserve the original record until an administrator has reviewed it.