Find the workload that fails next — while it still looks healthy.
Every running workload is scored continuously against 200+ rules. 49 correlation engines check its config against live telemetry — something a manifest scanner cannot do.
one helm install · get, list, watch only · first findings in minutes
Overview
A pod can be Running and still be one node drain away from an outage. Runtimez scores every workload on runtime health, spec hygiene, resource sizing, security, change history and add-ons, and checks the config against what the workload actually does. Each finding comes with its evidence, the rule that fired and a fix.
Every score shows its working
Open any workload and see each factor that moved the score, with a recommendation next to it.
- Security, reliability and sizing factors add up to one score, so one ranking covers all of them.
- Survival verdicts for maintenance, AZ failure and node failure, each with a plain-English reason.
- Honest gaps: when there is no observed latency, it says a probe cannot be compared instead of guessing.
One drawer for everything about a workload
Metrics, logs, events, containers, changes, history and profiles — without leaving the page.
- CPU and memory against requests and limits, with current, average, max and p95.
- Kubernetes events quoted verbatim, such as a
FailedSchedulingnaming the anti-affinity rule that blocked it. - Mesh and add-on routes for the workload, and a deep link into root-cause analysis.
Know whether the cluster can actually scale
NodePools with limits vs. usage, NodeClaims, disruption budgets and 24-hour activity, next to Cluster Autoscaler node groups.
- Pod provisioning p50 / p90, launch failures and pending pods in one strip.
- State per NodePool: not ready, at limit, degraded, drift blocked, drifting or healthy.
- 10 Karpenter and 9 Cluster Autoscaler rules, including nodes that fail to register and stale scaling loops.
Answers in the mesh's own words
Runtimez reads Envoy response flags per edge and Istio's own objects, so a 503 is traced to its real cause.
- Fault injection left on, broken DestinationRules, mTLS mismatches and routes with no subset — 24 Istio rules.
- Traffic counters that a sidecar restart cannot inflate.
- “Do not scale this workload”: tells you when the failing requests never reached it.
A service graph built from real calls
Traces, logs and metrics are used as evidence — not as another dashboard to watch.
- The graph comes from observed calls, with calls, error rate and p99 on every node and edge.
- Zero-code instrumentation: OTel SDKs injected by policy with no operator, and eBPF traces as a fallback.
- Log and trace sampling with a shadow mode, so you see what a policy would drop before it drops it.
Your next OOM and your bill, on one screen
Requests and limits compared with observed p50 / p95 usage for every container.
- Verdicts per workload: under-provisioned, unset, over-provisioned, good or insufficient data.
- A suggested request for each container, based on seven days of pod-level usage.
- Reclaimable CPU and memory totalled for the cluster.
More in Production Reliability
Everything else that ships in this product today.
Config vs. telemetry
HPA denominator mismatch, a PDB that blocks its own drain, probe timeout vs. real p95, grace period vs. in-flight requests, limits cut below observed usage.
Dependency blast radius
Uses observed call volume to show who else goes down with a workload.
Autoscaler verdicts, verbatim
Surfaces FailedGetResourceMetric and similar HPA errors before the autoscaler silently stops scaling.
GitOps · Argo CD
Every Application with sync and health state. NOT_SYNCING is explained in 31 deterministic rungs read from Argo CD's own messages; 39 rules.
Ingress · Traefik
Every route with Traefik's router verdict, resolved backends and live request counters; 44 rules.
Telemetry cost controls
Log drop and sample rules with live preview, severity normalization, dedupe and a metrics cardinality view.
How it compares
They check config against best practice. Runtimez checks config against your live telemetry.
They tell you what broke. Runtimez tells you what will break, and why it broke, with citations. It uses your telemetry as evidence rather than replacing it.
Is this a replacement for Datadog or Grafana?
No. Runtimez uses your existing telemetry (Prometheus, OpenTelemetry, metrics-server) as evidence for risk verdicts and root cause. It is not trying to be your dashboarding tool.
What does a correlation engine check?
Rules that need both config and live telemetry: for example, a liveness probe timeout compared with the workload's real p95, or a memory limit compared with observed peak usage.
Which add-ons get their own workspace?
Argo CD (GitOps), Traefik (ingress), Karpenter and Cluster Autoscaler (autoscaling) and Istio (service mesh). Each appears automatically when the add-on is detected.
How often are workloads re-scored?
Continuously. Inventory syncs are pinned and every answer quotes the sync it came from, so you never act on stale data.
Related reading
One agent, the rest of the platform
Every product runs off the same read-only sweep and feeds the same ranked queue.
See it on your own cluster in under an hour.
Free for your first cluster. Read-only by default. Uninstall is one helm command.














