Skip to main content

Create a monitor

A Continuous Eval monitor watches live traffic, scores a sample of it, and alerts you when quality slips. It lives at /online-evals.

Step 1 — Start from a query

The best monitors start from a real failure query. In the Logs explorer, build the query that defines the traffic you care about, then click Create draft monitor. A screenshot of the Logs explorer with the Create draft monitor button.

Step 2 — Define the sample and scorer

On the monitor draft page:
  1. Set the sample rate or filter that selects which traces to score.
  2. Pick the scorer (calibrated, ideally) that decides pass/fail.
  3. Set the alert threshold — for example, alert if the failure rate over the last 100 samples exceeds 5%.
A screenshot of the monitor draft with sample, scorer, and threshold fields.

Step 3 — Simulate on history

Before the monitor touches live traffic, simulate it against historical traces. This proves the scorer and threshold behave the way you expect on real data.
A monitor that has only been simulated is not active. It will not alert anyone until you explicitly activate it.

Step 4 — Save the version

Save the monitor version. Versions are immutable; changes create a new version that you simulate and activate again.

Up next

Simulate, then activate the monitor.