Skip to main content
Alerting lets you define conditions on call metrics and receive a notification when those conditions are met. Instead of manually watching the Analytics Dashboard, alerts notify you proactively when something goes wrong — a success rate drop, a cost spike, or an unusual call volume pattern. Access alerts at Monitor > Alerts.

​
Creating an alert

1

Open Alert creation

Go to Monitor > Alerts and click New Alert.
2

Name the alert

Enter a descriptive name. Good alert names include the metric and the condition: “Failed Call Rate > 15% (Production Agent)” rather than “Alert 1”.
3

Select the metric

Choose the metric to monitor. Available metrics:
MetricDescription
Failed call ratePercentage of calls that ended without the expected outcome
Average call durationMean call length in seconds
Call volumeNumber of calls in the window
Total costAccumulated TTS + LLM + transcription cost
LLM error ratePercentage of LLM calls that returned an error
Function error ratePercentage of function/tool calls that returned an error
4

Set the threshold and comparison

Define when the alert fires. Examples:
  • Failed call rate greater than 15%
  • Average call duration greater than 300 seconds
  • Total cost greater than $50
  • Call volume less than 10 (to detect when calls unexpectedly stop)
5

Set the evaluation window

Choose how much data the alert considers when evaluating. Options: last 15 minutes, last 1 hour, last 6 hours, last 24 hours. Shorter windows catch problems faster but produce more false positives from small sample sizes. Use 1-hour windows for most operational alerts.
6

Scope to an agent or number (optional)

By default, the alert evaluates across the entire workspace. Narrow it to a specific agent or inbound phone number if you want per-agent alerting.
7

Configure delivery

Choose how you want to receive the notification. Multiple delivery methods can be selected for a single alert.
8

Save

Click Save Alert. The alert begins evaluating immediately against the configured window.

​
Delivery methods

Email Enter one or more email addresses. When the alert fires, each address receives an email containing the alert name, the metric value that triggered it, the threshold, and a link to the Analytics Dashboard filtered to the relevant time range. Webhook Enter a webhook URL. When the alert fires, DialNexa sends a POST request to the URL with a JSON payload: 
{
  "event": "alert.triggered",
  "alert_id": "alert_xxxx",
  "alert_name": "Failed Call Rate > 15%",
  "metric": "failed_call_rate",
  "value": 0.21,
  "threshold": 0.15,
  "operator": "greater_than",
  "window": "1h",
  "agent_id": "agent_xxxx",
  "triggered_at": "2024-03-15T14:32:00Z"
}
The webhook is sent with a signature header for verification. See Webhooks for signature verification details.

​
Alert history

Monitor > Alerts > History shows a log of every alert trigger event, including the metric value at the time of trigger, the delivery status (email sent, webhook delivered), and whether the delivery succeeded. If a webhook delivery fails (non-2xx response or timeout), DialNexa retries up to 3 times with exponential backoff. Failed deliveries appear in the alert history with the HTTP status code returned.
Metric: Failed call rate
Threshold: Greater than 15%
Window: 1 hour
Why: A sudden increase in failed calls usually indicates a broken function/tool call, a prompt regression, or an upstream service outage. 15% is a reasonable threshold for most production agents — adjust based on your baseline.
Metric: Average call duration
Threshold: Greater than 300 seconds (or 1.5x your baseline)
Window: 1 hour
Why: Long calls often indicate the agent is looping, confused, or waiting on a slow function call. A duration spike with stable volume points to a behavioral regression.
Metric: Total cost
Threshold: Greater than your daily budget target
Window: 24 hours
Why: Prevents unexpected billing surprises. Set the threshold at 80-90% of your daily target to give yourself time to respond before hitting the hard limit.
Metric: Call volume
Threshold: Less than 5 calls
Window: 1 hour (during expected operating hours)
Why: If your agent normally receives 50+ calls per hour and volume drops near zero, something has broken upstream — a phone number issue, a routing misconfiguration, or a workspace error. This alert catches silent failures that do not show up in error rate metrics.
Metric: Function error rate
Threshold: Greater than 10%
Window: 30 minutes
Why: High function error rates indicate your integration endpoints are down, returning unexpected responses, or are being rate-limited. Catching this early prevents cascading call failures.