Skip to content

Roles and permissions

Seylo has four roles. They rank viewer < releaser < approver, and admin implies all three.

Role Can do
viewer Read: dashboard and matrix, history, requests, resources, environment comparison
releaser Everything a viewer can, plus: submit releases, create namespaces, add components, propose manifest and environment edits
approver Everything a releaser can, plus: approve or reject pending requests, and view the audit log
admin Everything above, plus: register and edit clusters, install and revoke agents, configure single sign-on, manage users and policy

Billing manager (billing_manager) sits beside them, outside the ranking. It lets a person see and pay on the Billing page and nothing else; it satisfies no other role. Give it to your finance team without giving them access to your clusters. See plans and billing.

A user holds a set of roles, not one role. The set is the union of everything granted by your identity provider mapping, your organisation’s base roles, and any explicit per-user grant.

A user who authenticated successfully but whose identity provider granted no mapped role is signed in and authorised for nothing. Every action returns 403.

This is not a bug and it is not a misconfiguration by itself — it is exactly what a correctly configured directory does for someone who has not been granted access yet. Seylo says so explicitly rather than leaving you to discover it: test login reports “no roles” as a finding, with the reason.

When approvals are enabled — the default once you are using an identity provider — the mutating actions do not apply immediately. A releaser submitting a release creates a pending request; nothing touches the cluster.

  • releaser submits → the request is pending.
  • approver or admin approves → the same engine code runs, with the original requester preserved as the initiator. The request becomes applied or failed.
  • approver or admin rejects → rejected, nothing applied.

With approvals disabled, mutations apply immediately — but still write audit rows, so there is an audit trail either way.

Where roles apply, and what they do not control

Section titled “Where roles apply, and what they do not control”

A role says what someone may do. Where they may do it is set by grants, and by two boundaries outside Seylo roles altogether. Knowing which is which saves an argument later.

By default a role applies on every cluster and in every namespace: a releaser is a releaser everywhere your organisation has clusters. An admin can narrow that on Admin → Access → Where roles apply:

  1. Add grants. A grant names a person, a role, a cluster and the namespaces the role applies in — “releaser in dev and qa on cluster eu-1” is one grant.
  2. Press Turn scoping on. Grants do nothing until scoping is on, so you can prepare them first.

The rules that are easy to miss:

  • A grant only narrows. It never gives anyone a role they do not already hold from your identity provider or their role assignment, so an admin cannot use grants to raise anybody’s access, including their own.
  • Someone with no grants is not scoped. With scoping on, their roles still apply everywhere. The page shows how many role-holders are still unscoped. To give someone no access, take the role away.
  • Grants are allow-only. There are no deny rules. The cluster’s namespace deny list applies first, to everyone.
  • A higher role covers a lower one. An approver grant lets its holder both submit and approve in those namespaces.
  • Grants limit what a person sees, too. A scoped person sees only their namespaces: in namespace lists, environment variables and ConfigMaps, the resource browser and its YAML, schedules, deployments and insights. Opening another namespace is refused with the reason. Cluster-wide objects, such as nodes, are not limited by grants.
  • Approvals follow grants. A scoped person’s approval queue lists only requests in their namespaces, and approving or rejecting needs an approver grant in the request’s namespace. A request with no single namespace (a cluster-wide one) needs a grant over every namespace of its cluster.

To let someone deploy to dev but not to production, give them a grant for the dev namespaces and turn scoping on.

Two more limits apply to everyone, whatever their role and grants:

  • the cluster’s namespace allow/deny policy, and
  • the agent’s RBAC posture in that cluster, enforced by Kubernetes itself — see choose an RBAC posture.

The agent’s own Kubernetes Role is the final word

Section titled “The agent’s own Kubernetes Role is the final word”

Whatever your Seylo role says, the agent can only do what its Kubernetes Role/ClusterRole permits. Seylo reads that grant back and shows it to you, so the real boundary is visible before something is refused rather than after.

admin is a role inside your organisation: full control of your clusters, agents, schedules, approvals and SSO, and no knowledge that any other organisation exists.

A platform administrator is a Seylo-side identity that can act across organisations. It is deliberately not a role claim — role claims are minted by your directory, and a customer who can edit their own app roles must not be able to grant themselves the platform. Every platform action is written into both audit trails, so if we act on your data, you can see it in your own audit export.