A flat "alert me if CPU goes above 90%" rule is a blunt instrument — it misses slow-building problems and false-alarms on normal daily patterns. PulseServe's Alert Thresholds tab supports several smarter detection styles alongside the classic fixed threshold, so different kinds of problems can be caught the way they actually show up.
Detection styles available
- Fixed threshold — the classic "alert if this value crosses a line" rule.
- Adapts to normal — learns a metric's usual range and alerts when it drifts unusually far from that, instead of a hardcoded number.
- Catches a trend early — flags a metric heading in the wrong direction before it actually crosses a hard limit, so there's time to act before a breach.
- Accounts for daily/weekly patterns — so an expected Monday-morning traffic spike doesn't false-alarm the way a flat threshold would.
- Catches sudden shifts — flags an abrupt change in behavior, not just a gradual drift.
- Requires multiple signals together — only fires when several related things degrade at once (e.g. latency and error rate both worsening), cutting down on noise from any one flaky signal.
- Error-budget style — for teams tracking service-level objectives, alerts based on how fast your error budget is burning rather than a single data point.
Managing rules
From the Alert Thresholds tab, an admin can add, edit, enable/disable, or delete rules, organized by category (application, infrastructure, database). About 30 common rules — covering error rates, latency, CPU, memory, disk, database connections, and more — ship ready to go, and several more advanced examples ship pre-built but switched off, ready to enable when needed.
Predictive Warnings
Independent of the rules you configure, the Predictive Anomaly Agent continuously scans your connected systems and raises three kinds of warning: a resource trending toward exhaustion (with an estimated time until it becomes a real problem), a sudden surge in errors, and an upcoming change scheduled against a system that's already under strain. Click Run Predictive Anomaly Scan on the MELT page to trigger a scan on demand, or check the Predictive Warnings tab for what's already been found.
Example
Disk usage on a database server has been climbing steadily for days — nowhere near a hardcoded "90% full" alert yet, but the trend is unmistakable. A predictive warning fires days in advance with an estimated time until it actually fills up, giving the team time to schedule a resize instead of firefighting an outage.