Production Alerts

Alerts let you define threshold-based rules that run on a recurring schedule and notify your team when a metric on your traces, spans, or threads crosses a configured value. Every rule appears on the Alerts page (under Configure in the Observatory's side navigation) with its severity, what it watches, how often it runs, when it last triggered, and a status strip showing how often it has triggered over the last 30 runs.

Besides threshold rules, the page lists rules that fire on what Drift detects: a Regression or Anomaly on a trace's versions, or a Signal spike in a classifier's labels. Search, filter by severity, or sort by severity, name, or last triggered from the toolbar.

Create an Alert

1. Navigate to Alerts
2. Click New Alert
3. Enter a Name and an optional Description
4. Click Create — the alert is saved in a disabled state and you land on the alert detail page to finish configuring it

Configure an Alert

1. Configure the alert event

   Choose what to measure:

   • Data Model — Trace, Span, or Thread
   • Aggregation — the metric to evaluate; options depend on the data model (e.g. trace count, error rate, average latency, token cost)

2. Customize advanced filters (optional)

   Narrow the data the alert evaluates using the same filter controls available in the Observatory — environment, tags, metadata, and more. Leave empty to monitor all data for the selected model.

3. Set alert conditions

   • Threshold — a direction (Above or Below) and a numeric value; the alert fires when the aggregated metric crosses this boundary during a scheduled run
   • Frequency — how often the rule evaluates

   Frequency	Interval covered
   Every 30 minutes	Previous 30 minutes
   Hourly	Previous hour
   Daily	Previous day
   Weekly	Previous week
   Monthly	Previous month

Alert Status

Each alert row shows a status strip — a series of bars representing the last 30 evaluation runs, colored to show whether each run triggered. The row header shows the overall triggered percentage and the timestamps of the first and last run in the strip. Click the strip to open the full alert log with per-run details.

Test, Pause, and Delete

• Try alert — immediately evaluates the rule against current data without waiting for the next scheduled run. Useful for verifying your configuration before enabling.
• Pause / Resume — toggle an alert between active and paused. Paused alerts skip notifications but retain their configuration.
• Delete — permanently removes the alert rule and its history.

Last updated on October 7, 2026