What it does
Komodor's Agentic Operations Platform (KAOP) packages AI SRE workflows for investigating and managing production operations on Kubernetes. Its agent catalog separates single-resource RCA, cluster-level investigation that can correlate Grafana telemetry and GPU signals, and fleet-wide questions across an account's clusters. This makes Komodor an operational investigation layer for teams already collecting signals and alerts, not a metrics store or replacement for their existing monitoring system. The distinction matters: its Kubernetes RCA agent is documented as read-only apart from a constrained diagnostic exec grant, and it reports findings rather than changing the cluster.
Choose the investigation scope after an alert
Komodor's Kubernetes catalog offers agents at different scopes. Kubernetes RCA investigates one resource in one cluster; Cluster Investigator can examine a cluster with Grafana telemetry and GPU health signals; Klaudia Investigator handles questions spanning the clusters available to the Komodor account. The first two are self-hosted and limited to the cluster where they run, so installing one instance does not give it fleet-wide visibility.
An illustrative workflow, not hands-on tested: after an existing monitor flags an unhealthy Deployment, use the resource-level agent to gather evidence about that workload, or choose the cluster investigator when the question needs connected Grafana context. Review the agent's run evidence before taking action. The Kubernetes RCA documentation explicitly says the agent changes nothing; it is a diagnosis step, not alert detection or an automatic remediation system.
The deployment boundary is important. The self-hosted RCA agent reads cluster resources, excludes Secrets by default, and grants pods/exec for a fixed set of diagnostics; that permission can be narrowed or removed. It has no external metrics-vendor or cloud-provider context, so it should state when the cause is outside what it can observe rather than imply a complete diagnosis.
Komodor lists Starter and Enterprise as custom-priced plans; Starter combines a platform fee with AI tokens, while Enterprise adds token and invocation allowances. The pricing page offers a trial but does not state a public price. Keep the tools that detect incidents and collect telemetry in place, then use Komodor for Kubernetes-focused investigation and operational workflows.
Strengths
- Offers separate Kubernetes agents for single-resource RCA, one-cluster investigation, and questions across an account's cluster fleet.
- Cluster Investigator can correlate Kubernetes context with Grafana telemetry and GPU health signals.
- The self-hosted Kubernetes RCA workflow records evidence and is documented as making no changes to the cluster.
Watch out for
- Kubernetes RCA is self-hosted, runs one instance per cluster, and does not see external metrics-vendor or cloud-provider data.
- The diagnostic agent's `pods/exec` grant is cluster-wide by default, although its documentation describes ways to narrow or remove it.
- Secrets are excluded from the Kubernetes RCA agent's default read access, so some evidence such as Helm release history is unavailable unless the configuration is changed.
Best for
- Kubernetes platform and SRE teams that already have alerting and telemetry but need help investigating resource- or cluster-level incidents.
- Teams that want governed, recorded AI-assisted operational investigations rather than an additional metrics collection backend.
Sources
- Kubernetes - Komodor - KAOP (Oct 26, 2026)
- Pricing and Plans | Komodor (Oct 26, 2026)
- Komodor: The Agentic Operations Platform for Production (Oct 26, 2026)