Anomaly detection watches your monitor’s response times and learns what “normal” looks like over time. When it spots a pattern that doesn’t fit — a gradual slowdown, an unusual spike, or a trend heading in the wrong direction — it alerts you before the monitor goes fully down.
Why it matters
A monitor going from 200ms to 1800ms is a problem, even if it hasn’t crossed your timeout threshold yet. Traditional uptime monitoring only fires when something is completely broken. Anomaly detection catches the early warning signs.
How it works
The ML service builds a baseline model for each monitor using historical response time data. It accounts for patterns like:
- Daily traffic cycles (slower at night, faster during business hours)
- Weekly patterns (weekends vs. weekdays)
- Gradual growth trends
When new response time data arrives, it’s scored against the model. A significant deviation triggers an anomaly alert.
What you’ll see
When an anomaly is detected, you receive a notification — separate from a down alert — describing the nature of the deviation. The monitor detail page also shows:
- Trend line — whether response times are improving, stable, or degrading
- Predicted values — where response times are likely to go based on current trends
- Anomaly markers — points in the chart where unusual behavior was detected
Getting started
Anomaly detection runs automatically on all monitors once they have at least 7 days of check history. No configuration needed.
Plan availability: Anomaly detection is included on the Pro plan and above.