Alerts
How alerting works
Before agent alerts existed, nothing told a pharmacy that automation had stopped. On one real day a QS1 session disappeared in the early evening, the queue stalled for about an hour and forty minutes, and it was found by someone noticing the screen — by which time roughly ninety processable documents had been pushed into manual review.
Agent alerts exist so that is an email, not a discovery.
Important: Alerts are off until you turn them on. There are no recipients by default. Setting them up is the single highest-value thing you can do after your first device starts — see Set up agent alerts.
Two triggers, deliberately both
| The agent reports a problem | The cloud notices silence | |
|---|---|---|
| Catches | errors, being stuck, stopping | crashes, killed processes, a powered-off machine |
| How fast | seconds | one check cycle, after the silence you configure |
| The email says | the agent's own sentence, verbatim | that there has been no contact for N minutes |
The first path is the agent telling on itself: it posts what happened and the email carries its own description, so you read "QS1 / PrimeCare is not running…" rather than a status code.
That path cannot work when the agent is dead or cut off from the internet — which is precisely why the second exists. Only the cloud can conclude that an agent has gone offline; the agent can never report its own crash. The two are complementary, not redundant.
The four things you can be alerted about
| Alert | Fires when | What it means for the pharmacy |
|---|---|---|
| Stopped with an error | The agent hit a fault it could not recover from and stopped. | Nothing will process until someone presses Start. |
| Stuck, needs attention | Still running but waiting on a person — QS1 isn't open, the screen can't be read, the cloud is unreachable, or the queue isn't moving. | Nothing processes until a person acts. The agent resumes by itself once the blocker clears. |
| Went offline | No contact while the schedule says it should be running. | The machine is off, asleep, or the agent crashed. |
| Finished a run | The run drained the queue, or someone pressed Stop. | Informational. Off by default, because it fires on every normal run. |
What alerts deliberately do not cover
Individual document outcomes. A single document going to manual review is normal and is the report email's job. Mixing the two makes both noisy and gets alert mail ignored — which defeats the point.
Where alerts are recorded
Every alert the system considers is written down, whether or not an email went out, together with the reason it was or wasn't sent. That ledger is what answers "the agent stopped — why didn't I hear about it?". See Why didn't I get an email?.