FAQ
What roles exist, and what can each one do?
admin (global, full access), maintainer (scoped to one or more assigned
clusters; can operate deployments, evaluation, simulation, storage,
observability and most of Model Service on those clusters), and end-user
(self-scoped: manage your own API tokens, call the inference gateway, view
your own usage). See Auth and RBAC for the
full permission catalog.
As a maintainer, can I change a model service's selection policy?
No — choosing the selection policy (random, round_robin, ...) for a
model service is admin-only. You can still adjust a provider's own
parameters under whatever policy is currently set (for example weight,
once a weighted policy is available), and you can publish, edit or remove
your own cluster's providers. See Model services.
Why can't I see another team's cluster on the Model Services page?
Model Service data (groups, members, routers) is filtered by cluster scope like every other page. If you're a maintainer, you only see clusters you're bound to; ask an admin to bind you to a cluster if you believe you should have access.
Does Lens sit in the inference request path?
No. The Lens UI/BFF/Python backend are the control plane only. Inference traffic to a published model service goes directly from the caller to the cluster's shared Gateway, which routes through the model's HTTPRoute to its InferencePool/EPP and then to the backend pod — Lens is never on that request's critical path.
How is Model Service usage measured?
Usage is derived from the gateway's access logs (asynchronously, so it never blocks a request) and can be filtered by cluster, model and user on the Usage tab. See Observability.
Where do I report a bug or ask for a feature?
Use the repository's issue tracker, or see Contributing for the expected workflow.