Skip to main content
Koperator is the operator that runs your self-hosted agents in your own Kubernetes cluster. You install it once, with one Helm command, and the Komodor Agentic Operation Platform (KAOP) shows the install as a deploy target. Developers then deploy, update, scale, stop and delete agents on that deploy target from KAOP, with no kubectl or helm of their own. Deploy targets are under Settings → Deploy targets when your account has them. If you do not see the page, ask your account team to turn it on.

Who owns what

Devops owns koperator: the install, the upgrades, and the guardrails it enforces in the cluster. Those guardrails live in your Helm values, so KAOP cannot change them. Developers own the agents. Who may deploy to which deploy target is an ordinary KAOP permission on that deploy target, so a grant on staging does not allow a deploy to production. The change policy of a deploy target is a KAOP setting that an owner or admin sets, and only an owner or admin approves a revision there.

How koperator works

  • It only dials out. Koperator registers with KAOP, sends a heartbeat, and holds a long-poll open for changes. Nothing connects in to your cluster.
  • A change applies in full or not at all. Koperator checks each agent spec against your guardrails first. If one check fails, it applies no part of the spec, the agent keeps its last applied state, and the developer sees the reason in KAOP.
  • It reports what happened. For each change, KAOP shows whether koperator applied it or why it failed.
  • It pins each image to its digest when the registry lets it read the digest, so a pod that restarts runs the same image.
  • It keeps the agents as it applied them. When another tool changes a field that koperator set, koperator applies its version again. See When something else changes an agent.
  • It runs two replicas. One is the leader and does the work; the other takes over if the leader stops. See High availability.
  • Running agents do not depend on it. When koperator is disconnected or stopped, the agents keep running and serving. A change in KAOP stays pending until koperator connects again.

What koperator lets you do

Install koperator

Create a deploy target and run the install command.

Configure koperator

Namespaces, guardrails, Secrets, network, proxy and KEDA.

Deploy agents to a deploy target

Deploy, update, revise, stop and delete agents from KAOP.

Adopt agents installed with Helm

Let KAOP manage agents you already run from a Helm release.

Upgrade koperator

Move to a new koperator version.

Uninstall koperator

Choose what happens to the agents when koperator goes away.

Argo CD and Flux

Run koperator under a GitOps tool, and handle drift.
To run an agent with no replica while it is idle, see Scale agents to zero.

Image security

Each koperator image is built for linux/amd64 and linux/arm64, scanned before it is published, and published only when the scan finds no fixable high or critical vulnerability. Each image carries an SPDX SBOM per architecture. The image and the chart are signed with keyless cosign, so you can verify them without a credential:
Run the same command against ghcr.io/komodorio/charts/koperator@sha256:<digest> for the chart. Report a vulnerability to security@komodor.io.