Skip to main content
Alerts are set per agent in Agent Settings > Alerts. Conditions use your evals, red flags, data extraction fields, latency and call details, so set those up first. To deliver alerts to a webhook, add a webhook URL in Agent Settings > Webhooks.

​
What is an alert?

An alert is a notification Tuner sends when calls matching your conditions cross a threshold within a time window, for example “5 calls in 1 hour where Hallucination is FAIL”. Tuner checks your alerts after it finishes analyzing each call and notifies you by Email, Webhook or both.

​
Create an alert

1

Open the Alerts tab

Go to Agent Settings and open the Alerts tab. It lists your existing alerts with their channels, status and last triggered time.
2

Click Create Alert

Click Create Alert to open the alert form.
3

Name the alert and pick a severity

  • Alert name: a name for the alert, such as “Hallucination spike”.
  • Description (optional): a note for your team.
  • Severity: Info, Warning (default) or Critical. Severity tells your team how urgent the alert is.
4

Add trigger conditions

Under Trigger Condition, build each condition from three parts:
  • Select field: what to check. Fields are grouped by source (see Condition fields).
  • Op: how to compare. The operators offered depend on the field (see Operators).
  • Select value: the value to compare against. For a pass/fail eval, pick PASS or FAIL.
Click Add condition to add more, up to 10. With two or more conditions, choose Match: ALL (AND) to require every condition, or ANY (OR) to require at least one.
5

Set the threshold

Under Trigger when, choose Count or Percentage.
  • Count: enter a number of calls and pick a time window. The alert fires when that many matching calls occur within the window, for example “10 calls in 1 hour”.
  • Percentage: enter Percentage (%), Minimum calls and Time window. The alert fires when that share of calls match and at least the minimum number of calls match within the window. Percentage mode supports exactly one condition.
The time window is 5 mins, 15 mins, 1 hour (default) or 24 hours.Then pick a Cooldown: how long the alert waits after it triggers before it can trigger again. Choose 15 minutes (default), 1 hour, 4 hours or 24 hours. A longer cooldown sends fewer repeat notifications while calls keep matching. The summary under the panel updates to show the threshold and the cooldown.
6

Choose notification channels

Under Notification channels, select one or both:
  • Email: enter one or more addresses, separated by commas.
  • Webhook: Tuner sends an alert.fired event to your agent’s webhook. If the agent has no webhook URL yet, selecting it prompts you to configure one. See Webhooks for the payload.
Alert notification channels: Email and Webhook
7

Save the alert

Click Save alert. The alert appears in the list and is active.

​
Condition fields

The Select field list groups fields by source. A group only lists items that exist for this agent.

​
Operators

Is present and Is not present take no value. Text comparisons ignore case with every operator. Customer-Ended-Call matches customer-ended-call, and Contains hangup matches Customer_Hangup. A call with no ended reason matches Call Ended Reason only with Is not present or Not equal (!=).

​
Combine conditions

The dashboard combines a flat list of conditions, up to 10, with one Match: setting. ALL (AND) requires every condition, and ANY (OR) requires at least one. The API also accepts nested groups, which combine AND and OR in one alert. For example, an alert can fire when a call is a refund longer than 60 seconds, or when a call has a failed Hallucination eval.
  • A tree nests up to 10 levels deep and holds up to 50 conditions.
  • Percentage mode needs exactly one condition, so it doesn’t accept nested groups.
The dashboard can’t create nested groups. If you create an alert with nested groups through the API, you can read its conditions in the dashboard and change them only through the API. You can still edit the alert’s other fields, and saving keeps the conditions as they are.

​
When alerts fire

  • Tuner checks alerts after it analyzes a call. If any analysis step failed for that call, Tuner doesn’t check alerts for it.
  • If the call’s outcome has Stop Analysis turned on, Tuner checks alerts as soon as the outcome is saved. For that call:
    • Conditions on Evals, Data Fields and Call Intent don’t match, including Not equal (!=).
    • Percentage alerts on those fields leave the call out of the total.
    • Conditions on all other fields match as usual.
  • After an alert fires, it doesn’t fire again until its Cooldown has passed, even if more calls match.
  • Paused alerts are not checked.
  • If you rename an eval or data extraction field, alerts that use it keep working and show the new name.
  • Calls whose outcome has Stop Analysis turned on are still checked. Evals, data fields and intents Tuner skipped count as not analyzed, so a condition like Not equal (!=) on them doesn’t match.
  • If you delete an eval, data extraction field, voice metric or red flag that an alert uses, Tuner pauses the alert until you fix its condition. See Resume a paused alert.
  • Calls from a simulation or a Replay are not checked. They don’t count toward any alert’s number of calls or percentage.
In Percentage mode, the share is matching calls divided by the calls Tuner can check for the condition’s field. A call it can’t check never counts as a match, even with Not equal (!=). Tuner can check a call when:

​
Example alerts

Alert name: Hallucination spike
Severity: Critical
Condition: Hallucination Equals (=) FAIL
Trigger when: Count, 5 calls in 1 hourNotifies you when the agent gives inaccurate information on several calls.
Alert name: Identity not verified
Severity: Warning
Condition: Identity verification compliance Equals (=) FAIL
Trigger when: Percentage, 10% with Minimum calls 5, 1 hourUseful for agents that must verify identity before sharing sensitive information.
Alert name: Slow responses
Severity: Warning
Condition: P90 Latency Greater than (>) 2500
Trigger when: Count, 15 calls in 1 hourTells you when many calls have slow agent responses.

​
Manage alerts

Alerts management in Tuner
The Alerts tab lists each alert with its RULE NAME, CHANNELS, STATUS and LAST TRIGGERED time.
  • Pause or resume: use the switch in the STATUS column. The status shows Active or Paused.
  • Edit: click the edit icon, change the alert, then click Update Alert. The Cooldown shows the alert’s saved value. If the cooldown was set through the API to a value that isn’t one of the options, such as 30 minutes, it appears as an extra option and is kept unless you pick another. Conditions with nested groups are read-only.
  • Delete: click the delete icon and confirm. This can’t be undone.

​
Resume a paused alert

Tuner pauses an alert when an item its conditions use is deleted. The alert’s row shows the reason, for example Eval deleted. Edit to resume., and has no switch. If a paused red flag rule caused the pause, resume that rule first. To resume the alert, click Edit alert to resume and pick a condition that doesn’t use the deleted item. Then click Save and resume. If a condition still uses a deleted item or a red flag that no active rule applies, the alert stays paused. Otherwise it’s active again.

​
Best practices

  • Start with high-signal alerts. Alert on clear, critical problems only, so the important ones don’t get ignored.
  • Tune thresholds after you see real volume. Too low creates noise; too high misses issues. Percentage mode with a Minimum calls value helps when volume varies.
  • Route by urgency. Send critical alerts to a webhook connected to a channel your team watches. Email works for the rest.
  • Review regularly. Revisit your alerts when the agent changes.

​
Limits

Alerts are also available through the public API, in the Agent Settings group of the API Reference, starting with Create alert definition and List alert conditions.

​
Next steps

Red flags and labels

Define the rules you can alert on.

Webhooks

The alert.fired payload and how to set up your endpoint.