Sage Infrastructure Solutions Group IncSage Infrastructure Solutions Group IncLoading…

Sage Infrastructure Solutions Group Inc

Kubernetes access

A Kubernetes cluster enrols much like a host, with an agent that runs inside it. Keyreeve then decides each API request before it reaches the API server, acting in the cluster as the person — their identity and groups — and nothing more.

Enrol a cluster

Create a cluster enrollment token under Admin → Kubernetes clusters, then apply the agent manifest the page gives you. Unless the token approves clusters automatically, the cluster then waits on that page for an administrator to approve it. The agent keeps the RBAC bindings Keyreeve manages and carries Keyreeve's requests into the cluster.

Who may do what is set under Admin → Kubernetes access: rules grant namespaces (by name or label), pods by label, or the whole cluster to identity-provider groups, with a role. Keyreeve's own namespace can never be granted. Temporary access works for clusters too.

Use kubectl

Sign in with keyreeve login first, then:

keyreeve kube ls                    # the clusters you can reach
keyreeve kube login my-cluster      # write a kubeconfig context
kubectl get pods

keyreeve kube login takes --kubeconfig to choose the file, --namespace (-n) to set the context's default namespace, and --no-switch to leave your current context alone.

In the browser and the desktop app

The Kubernetes page in the web interface and the Kubernetes view of the desktop app list the clusters, namespaces and pods your rules allow. From there you can read a pod's log, open a recorded shell in a pod (choosing the shell the image has), and open a kubectl shell. The desktop app can also forward a local port to a pod, where port forwarding is allowed.

What Keyreeve does with a request

  • It decides before the API server sees the request, and re-checks long-running ones every ten seconds.
  • exec and attach sessions are recorded the same way SSH sessions are. WebSocket is the only protocol allowed (kubectl 1.30 or newer); SPDY exec is refused.
  • A successful write is recorded with its body, masked first: a Secret's values are never stored, container environment values are masked, and anything credential-shaped is masked wherever it appears, including inside a CRD.

The kubectl shell

Where a rule allows it (it is off by default), Keyreeve opens a kubectl shell from the browser or the desktop app: a pod created for that session alone, with no service account, running as non-root with a read-only root filesystem, capabilities dropped and a network policy that keeps it off the cluster network. It is deleted when the session ends, and recorded from end to end.

AI in Kubernetes shells

As on hosts, a person can let an AI assistant type into a pod or kubectl shell they have open, where AI is allowed for that cluster (per cluster, under Admin → Kubernetes clusters) and by their access rule. Every AI input is marked in the recording and the audit log.

Access that did not come through Keyreeve

The agent reports role bindings made by hand in the cluster that grant a Keyreeve identity — the shape of a break-glass grant someone made themselves — and Keyreeve-managed bindings that were changed or deleted underneath it.

Beyond that, Keyreeve can take the cluster's own audit log: point the API server's audit webhook at Keyreeve and reaching the API server another way is reported rather than invisible.

[!NOTE] The webhook is additive. A Kubernetes API server writes its audit file and posts to a webhook at the same time, so an audit pipeline you already collect into your own SIEM keeps working exactly as it did.

The generated policy is deliberately narrow: system components, Keyreeve's own agent, health endpoints and watches are dropped, and what remains is at Metadata level, so no request or response bodies — and therefore no secrets — leave the cluster. Set it up with the Audit log button on the cluster's row under Admin → Kubernetes clusters.