Skip to content

Connection methods

Registering a cluster is admin only. Every cluster connects the same way: through the in-cluster agent, which reaches private clusters and leaves Seylo holding no credential for them. Adding a cluster offers nothing else.

Every other method needs two things a distributed fleet cannot reliably give:

  1. a network path from Seylo to the cluster’s API endpoint, and
  2. credentials for that cluster, held by Seylo.

Private and firewalled clusters have no inbound path, and their owners frequently will not hand over cloud credentials. The agent inverts the connection: it runs inside your cluster, dials out over WSS, and authenticates to Kubernetes with its own ServiceAccount.

The consequence worth repeating: there is no cloud credential or kubeconfig on our side to leak. See the trust model.

A cluster can name an image registry, optionally. With one that Seylo can list, the release form suggests image tags; without one, you type the tag, and the rollout checks it. On an agent-connected cluster Seylo usually cannot list the registry, because it holds no registry credential for it.

Field What it takes
Registry login server The registry’s host, for example registry.acme.example
Registry type Read from the host unless you choose one. ACR only for an Azure Container Registry host (such as *.azurecr.io), GAR only for a Google host (*-docker.pkg.dev or gcr.io), ECR, OCIR, and OCI for any other registry

Since 0.3.15:

  • ACR and GAR with any other host are refused: Registry type "acr" needs an Azure Container Registry host (*.azurecr.io) .... Choose OCI instead.
  • The Username env var and Password env var fields are not saved from the portal or the API. They name environment variables on the Seylo server itself, which only the server’s operator sets.
  • Seylo reaches a registry only at a public HTTPS address, does not follow its redirects, and quotes only the start of an error the registry returns.

Independently of the method, each cluster carries an allow or deny list of namespaces. It applies to everyone equally — it is cluster configuration, not a per-user permission. For a person-by-person limit, see where a role applies. For a boundary enforced by Kubernetes rather than by Seylo, use a namespace-scoped agent: see choose an RBAC posture.