Skip to main content

​
Automated Pattern Detection

Continuously analyze your memory store to surface recurring patterns, anomalies, and risks — before they become incidents.

​
Pattern Types

The Pattern Detector scans memory entries and decision trace data for four pattern types:
PatternWhat it findsSeverity
Recurring failuresEntries that repeatedly appear in incident causal chainsWarning / Critical
Hot entitiesEntities with disproportionate write/outcome activity vs. averageInfo
Stale clustersGroups of old entries with no outcome links that may need pruningWarning
Confidence driftEntries whose confidence diverges significantly from their entity averageInfo

​
Alert Manager

Configurable rules that fire when matching patterns are detected: 
from amfs_patterns import PatternDetector, AlertManager, AlertRule

detector = PatternDetector(incident_threshold=2, stale_days=30)
report = detector.analyze(entries, outcome_data=outcomes)

manager = AlertManager()
manager.add_rule(AlertRule(
    name="Critical recurring failures",
    pattern_type="recurring_failure",
    min_severity="critical",
    cooldown_minutes=60,
))
manager.on_alert(lambda eval: send_slack_notification(eval))

evaluations = manager.evaluate(report)
Features: severity filtering, entity-path scoping, cooldown-based suppression (prevent alert fatigue), and callback registration for routing to Slack, PagerDuty, email, or custom systems.