Alerts

Be told by email, Slack or a signed webhook when call failures, latency, spend, credit or a campaign cross a limit you set.

Alerts watch a few numbers about your calls and tell you when one crosses a limit, so you hear about a problem before your customers do. Owners and admins set them up under Settings, Alerts.

Rules are always saved, but they are only checked, and only send anything, when alerts are switched on for your account. The page tells you when they are not.

What you can watch

Rule	Fires when	Notes
Call failure rate	`failed / (completed + failed)`, as a percentage, rises above your limit	Over the last N minutes. Calls still in progress are not counted. Not judged below the minimum number of finished calls (default 10), so one failed call out of two is not a 50% emergency.
Average response latency	The average of each call's end-to-end latency rises above your limit, in milliseconds	Over the last N minutes, same minimum.
Spend	What calls cost in the last N minutes rises above your limit, in dollars	Counted when the call ends and is priced.
Credit balance	Your prepaid balance falls below your limit, in dollars	Not available (not judged) when billing is not switched on for your account. This is in addition to the built-in low-balance email.
Campaign stalled	A running campaign has calls due, its calling hours open, and has placed no call for longer than your limit, in minutes	A campaign waiting for its hours to start, between retries, paused or finished is not stalled.

The numbers are measured the same way as on the analytics page, so you can check one against the other. A rule can be limited to one agent. Check it now shows what a rule would read right now and whether it would fire.

When you are told, and how often

Each rule is in one of two states, OK or firing, and you hear about changes, not about every check:

Fired: the first time it crosses the limit.
Reminder: if it is still over the limit, at most once per cooldown (default 60 minutes).
Resolved: once, when it is back within the limit.
If it clears and crosses again inside the cooldown, the second crossing is held back (and recorded as such in the history). A value hovering around the limit sends one message, not one every few minutes. When the cooldown ends and it is still over, you get a normal "fired" message.
If there is not enough data to judge (too few calls), nothing changes. A rule never resolves because traffic stopped.

Where it goes

Make channels and pick them on each rule. Send test sends a clearly labelled message through a channel and tells you how it went.

Email. To the workspace's confirmed owners and admins and/or confirmed members you list. Alerts cannot be sent to anyone outside the workspace, and a person who leaves it stops receiving them.
Slack. An incoming-webhook address (Mattermost and other tools that accept {"text": ...} work too). The address is a secret: it is stored, never shown again, and only its host and last four characters appear on the page.

Signed webhook. A JSON POST to your server, signed exactly like event webhooks (X-Dialgem-Signature) with a secret shown once when you create the channel:

json

{
  "type": "alert.firing",
  "id": "…",
  "created_at": "2026-11-03T12:00:00Z",
  "organization": { "id": "acme", "name": "Acme" },
  "rule": { "id": "…", "name": "Failures spiking", "metric": "failure_rate", "window_minutes": 15, "threshold": 10, "agent_id": null },
  "value": 23.5,
  "message": "Call failure rate is 23.5% over the last 15 minutes, above your limit of 10%."
}

type is alert.firing, alert.reminder, alert.resolved or alert.test. Deliveries are retried with backoff if your server is down; a response that can never succeed (a 404, for example) is recorded and not retried.

The page lists recent alerts with whether each channel received them (kept for 90 days). Every change to a rule or channel is in the audit log.

Limits

Up to 25 rules and 10 channels per workspace; checks run about once a minute, so an alert can be up to a minute or two late.
Alerts are not available to API keys; they are managed from the dashboard.
Only the numbers above. There is no per-number filter beyond the agent, and no quiet hours.