Changelog¶
Release history for AI Monitor — ML-powered anomaly detection for CloudWatch metrics.
26h4c (Oct 2026)¶
- Cross-account metric discovery: the discovery wizard now enumerates metrics across all connected accounts in parallel and presents them as a unified, deduplicated catalog
- Discovery runs asynchronously with a brief on-demand refresh, so large multi-account fleets no longer block the editor
- Stale metrics age out automatically after 7 days
- Added a home glyph to the editor header title
Improvements¶
- Faster metric processing: reduced per-batch latency in the metric collector, preventing rare processing backlogs
- Clearer alerts for static-threshold breaches: when a value crosses a configured "Alert if above/below" limit, it is now labeled as a limit breach rather than shown as an anomaly score, and the AI explanation reflects the limit crossing instead of a statistical anomaly
- Expanded system-health visibility across the monitoring dashboard, the anomaly report, and the support bundle (queue backlog and dead-letter depth, Firehose delivery lag, Lambda liveness and throttle alarms, collector processing time, discovery totals, and AI circuit-breaker state)
- Claude Haiku 5.5 is now the default and recommended Amazon Bedrock model for AI explanations, tuning, discovery suggestions and report summaries (Haiku 4.5 remains selectable). Enable Haiku 5.5 model access in the Amazon Bedrock console of the account running AI Monitor; a model you chose yourself in Settings is kept
- AI Tune now gives guidance for subscriptions whose "Alert if above/below" limit already exists but fires too often, including switching queue-depth subscriptions to anomaly detection
- More reliable sign-in after a session expires: the Sign in button restarts a clean login instead of reloading a one-time callback page
26h4b (Oct 2026)¶
- Select the Amazon Bedrock model for AI explanations and suggestions (advanced+), with Test-before-apply and a retirement (end-of-life) warning
- Click the "AI Monitor" title to return to the Overview dashboard
26h4a (Oct 2026)¶
- Add a summary dashboard
- Add a subscription search
- Display units where required
26h4 (Oct 2026)¶
- Detection and alerting refinements
- Web application security hardening
26h3 (Sep 2026)¶
- Scheduled maintenance release
- Update AWS Metric Namespace Catalog
- AWS Qualified Software
26h2 (Sep 2026)¶
- MoM enhancements
- Move timezone selection to FAQ
- Now maintain local changes with Save, Undo and Diff
- Added support link to landing and fulfillment pages
- Change logout behavior
- Add AI SIEM checkbox for future integration with the AI SIEM product
26h2 (Aug 2026)¶
- On stack update, address the "Could not locate index-pattern-field" errors on dashboard visualizations requiring a manual index refresh
- AWS FTR Approved
26h2 (July 2026)¶
- Support for AWS Organizations — consolidate and centrally govern multiple AWS accounts
- Enhanced Correlation — now leverages specific Log Processor patterns when anomalies occur
- iPad layout fix — subscription list footer buttons now display correctly on tablet viewports
- Preview-based cloning — create new alerts based on wildcard search
- Popout panels — Baseline and Graph can now be popped out to a larger window
- Threshold enhancements — improved AI support for hard high/low threshold alerts
- FAQ — now includes alert recipient state, entitlement state, and Bedrock endpoint availability
- AI Tune v2 — tuning recommendations are now computed deterministically using rules-based logic (anomaly frequency, metric type, observed percentiles) rather than relying solely on the language model for numeric decisions. Bedrock is still used to generate the human-readable explanation, but threshold values, direction, and static alert recommendations are derived from your actual data and SLA-appropriate defaults.
26h1 (June 2026) — Initial Release¶
Anomaly Detection¶
- Random Cut Forest (RCF) scoring with per-day-of-week adaptive baselines
- Z-score anomaly scoring (0–10 scale) with configurable thresholds per subscription
- Direction filtering: alert on high-only, low-only, or both deviations
- Static threshold alerts (Alert if below / Alert if above) for absolute floor/ceiling safety nets
- Wildcard dimension support with per-resource independent scoring
- Wildcard batching optimization for high-dimension-count subscriptions (compact+ tiers)
- Anomaly trend tracking: worsening, improving, stable, new
- Configurable anomaly retention (per-subscription override on advanced+)
AI-Powered Features¶
- Natural language metric discovery via Amazon Bedrock (Claude Haiku)
- AI tuning suggestions based on anomaly history with Ctrl+click OpenSearch full-history mode
- AI-generated anomaly explanations (async batch with circuit breaker and 7-day cache)
- AI executive summary in weekly anomaly reports
- Smart defaults from AWS metrics catalog (auto-fill stat, direction, threshold, period)
- Next/Prev navigation for multiple AI suggestions with duplicate detection
Metric Discovery¶
- Discovery wizard with namespace/metric browser
- Quick-subscribe chips with pre-configured smart mappings (no AI call)
- AWS/TrustedAdvisor integration (RedFlagChecks, YellowFlagChecks, ServiceLimitUsage)
- AWS metrics catalog with 57+ namespaces, auto-updated via Bedrock
- Show-all-AWS-metrics toggle (active + catalog inactive)
- Double-click to subscribe from discovery list
- Return to expanded namespace after save/cancel
Subscription Editor¶
- Browser-based editor with split-panel layout and keyboard navigation
- Preview metric data (Shift+click = 24h, Ctrl+click = OpenSearch cross-account)
- Learned baseline visualization with per-day-of-week sparklines
- Version history with JSON diff and one-click rollback
- Clone, enable/disable all, double-click toggle
- Deep-link support (
?sub=id) for anomaly email links - Session expiry detection with 60-second keepalive ping
- Timezone selector with local time display throughout
Anomaly Management¶
- Recent anomalies panel with subscription and score filtering
- Anomaly detail view with CloudWatch console link, trend info, AI analysis
- Dismiss single, dismiss all for subscription (Shift+click skips confirmation)
- Orphan detection: deleted subscriptions shown as "(deleted)" in anomaly detail
- Red badge count on Anomalies toolbar button
Alerting & Reports¶
- SNS alerts with severity, trend, training status, CloudWatch link, and runbook URL
- Configurable email notifications with per-subscription notify threshold
- Cost context in billing anomaly alerts (today vs baseline, % change)
- Weekly anomaly report with AI executive summary, severity breakdown, cost analysis, system health
- Report scheduling (per day-of-week) with configurable lookback window
- On-demand report generation with auto-select in preview
- Editor deep-link in alert emails
Maintenance & Suppression¶
- Scheduled and ad-hoc maintenance windows (compact+ tiers)
- Per-subscription or global suppression with start/end SNS notifications
- Suppression metrics and reporting
Correlation¶
- Cross-metric correlation engine (compact+ tiers)
- Configurable correlation window (1–30 minutes) per subscription
- Correlated metrics included in anomaly alerts
Cross-Account¶
- Multi-account metric ingestion via Metric Stream + Firehose + cross-account S3
- Per-subscription account filtering (local, specific, all)
- Cross-account anomaly tracking and reporting
- Setup script for remote account configuration
Cost Monitoring¶
- Billing processor Lambda for daily cost aggregation by service
- Cost anomaly detection (spending spikes vs day-of-week baseline)
- Cost trends in weekly reports with 7-day average comparison
OpenSearch Integration¶
- Automatic dashboard import (7 dashboards, 20+ visualizations)
- Anomaly events indexed to OpenSearch for dashboard visibility
- Metric Explorer dashboard for ad-hoc investigation
- Re-import on every stack deploy via custom resource
Operations¶
- CloudWatch dashboard with detector duration, anomaly counts, AI metrics, system health
- CycleDuration EMF metric for detector performance monitoring
- Support bundle script for diagnostics collection
- SSO integration (SAML/OIDC) for enterprise tier
- RBAC: admin and viewer roles via Cognito groups
- Landing page with stack info, editor link, dashboards link
- Custom domain setup script
- Service quota check script
Tiers¶
- Five tiers: Basic, Essential, Compact, Advanced, Enterprise
- Per-Lambda memory, timeout, and concurrency configuration
- Marketplace entitlement integration with dynamic feature gating
- Entitlement-aware processing with grace period on expiry