The fastest way to kill a monitoring tool is to let it spam people. A channel that lights up forty times on the first afternoon gets muted by lunch, and a muted monitor is worse than no monitor, because now you believe you are covered. So when we rebuilt AgentPing alerting, the design constraint was not "send more alerts". It was "earn every notification".
Here is what that looks like in practice.
One issue, not a stream
When a rule's condition becomes true, AgentPing opens a single issue and notifies your channels once. While that issue stays open, the same condition does not fire again. A run that keeps failing, an agent that stays quiet, a budget that stays over: each is one open issue, paged once, until it clears. When the condition does clear, the issue resolves itself and you get a single recovery note.
That is the whole contract: one ping when it breaks, one when it recovers. If you want a nudge while something stays broken, reminders are opt-in, and they only chase issues you have not acknowledged yet. Nothing fires on a loop.
Rules in plain language
You set rules for the things that actually matter, scoped to one agent or a whole team:
- an agent goes quiet
- runs start failing, or several fail in a row
- the share of healthy runs drops below your target
- quality drops, measured by your evaluation scores
- spending crosses a budget, or is on pace to
Every new account starts with one rule already switched on: the quiet-agent rule, routed to the owner's email. The single most common way an agent fails is that it simply stops, and now that is caught before you have configured anything at all.
The Issues page
Alerts land in your channels, but they also collect in one place. Issues in the sidebar answers a question every operator asks a dozen times a day: what is wrong right now, and what have I already dealt with? The nav badge carries the open count, so you get an ambient signal without going looking.
Open issues sort worst-first, and issues on the same agent collapse into one group. A struggling agent that trips three conditions at once reads as one problem with a count, not three separate fires you have to mentally reassemble. Acknowledge the group to quiet all of them at once.
The good-news state is the one we are proudest of: "No open issues. All agents operational." It should feel earned, because it is.
Acknowledge from the message
You should not have to open a dashboard to say "I've got this". Every notification carries a one-tap acknowledge link, signed so it needs no login. Tap it and the reminders stop; the issue stays open until the condition actually clears. Or snooze a rule for an hour, a day, or a week when you already know why something is noisy.
Where it goes
Email on every plan, including Free. Slack, Discord, Telegram, and signed webhooks on paid plans. Microsoft Teams and PagerDuty on Team and above. Each rule routes to the channels you pick, so staging can whisper to a side channel while production pages the people who need to wake up.
AgentPing observes your agents. It will never be the reason one stops, and it will never be the reason you stop reading your alerts.