Fleet observability agent
Fleet metrics.
Alert-free fatigue.
rig-metrics classifies every agent-window against its own SLO thresholds and runs a rolling per-metric z-score detector for real anomalies — then compiles one deduplicated alert list. Silent on windows that are fine, so signal never drowns in noise.
rig-metrics · SLO classifier + anomaly detector · 10-agent fleet[higgsfield: gold-lit oscilloscope waveform on a dark control-room console, single anomaly spike highlighted, cinematic low-key lighting]
Why teams install
What rig-metrics earns back
Classification tracks ground truth, not vibes
97.53% SLO breach/compliant accuracy against the fleet’s hidden health regime, 10 agents × 150 windows.
‹3a153e7›Catches real anomalies, not everything
93.75% anomaly recall at a 3.93% false-positive rate — the alert-everything shortcut is explicitly rejected.
‹84200ec›High recall alone doesn’t pass
Switching to alert-everything hits 100% recall but 100% FPR — the false-positive-rate gate correctly fails it RED.
‹1d0ecd5›How it works
Three steps. No faith required.
01
Observe
Telemetry — calls, errors, latency, cost — flows in per agent, per window.
02
Classify
Fixed per-agent thresholds flag breach vs. compliant; a rolling z-score flags real anomalies.
03
Alert
One deduplicated, severity-ranked list — critical only when breach and anomaly coincide.
Proof band
One packet. Every claim accounted for.
packet rig/rig-metrics · evalobs v1 fleet 10 agents · 150 windows/agent · seed 42 result slo_accuracy 97.53% (≥ 95%) · anomaly_recall 93.75% (≥ 90%) · fpr 3.93% (≤ 5%) gate alert_everything mode REJECTED — fpr 100% fails the floor anchor PASS — all three gates independently recomputed ‹3a153e7› themis SEALED — merge gate armed
Every rig-metrics run emits a packet like this — hash-chained, verifier-signed, and queryable from your own dashboard. How ProofPackets work →
Buyers
Who runs rig-metrics
SRE Lead
agent-fleet platform team
Head of Reliability
regulated fintech
Solo Operator
multi-tenant agent SaaS
Objections
Answered before you ask
Isn’t this just Datadog with extra steps?
Dashboards show you numbers. rig-metrics ships a sealed verdict per window — SLO accuracy, anomaly recall, and false-positive rate are gated and hash-chained, not just charted.
Won’t the anomaly detector just alert on everything to be safe?
That’s the exact planted failure we test for: alert-everything hits 100% recall but 100% false-positive rate, and the FPR gate correctly rejects it. High recall alone proves nothing.
How do I know the 97.5% accuracy number is real?
It’s computed against the fleet’s hidden health regime — ground truth the detector never sees — across 10 agents × 150 windows, reproducible at seed 42.
Pricing
Start free. Pay when it pays.
Generalist ships today. Domain Expert adds vertical memory + priors. Custom is built to your stack.
Free
Priced after a short conversation
3 agents monitored, 7-day retention.
Team
Priced after a short conversation
Unlimited agents, 90-day retention, alert routing.
Pro
Priced after a short conversation
Dedicated fleet, custom SLO tiers, SLA.
Early install slots keep the terms they sign — team pricing is set after a short conversation.
