[Rylon]

Summary

The API's NetworkPolicy allows ingress to the API port from any source, so any pod in the cluster can reach it. Callers still need valid credentials, but there's no way to narrow access only to the application that uses it, as the chart already does for every other component.

This adds an optional networkPolicy.api.ingressFrom, following the existing metrics.sandboxRunner.ingressFrom.

If empty (the default) we preserve the current behaviour, allowing access from any source. When set, we add the entries to the from section of the API's ingress rule, for example:

networkPolicy:
  api:
    ingressFrom:
      - namespaceSelector:
          matchLabels:
            kubernetes.io/metadata.name: librechat
        podSelector:
          matchLabels:
            app.kubernetes.io/name: librechat

One thing to watch out for when using this: the ingress controller (with ingress.enabled) and Prometheus (with metrics.enabled) will both need to be added to ingressFrom when they are enabled, because they reach the API on the same port.

Testing

• helm template with values-local.yaml renders identically before and after this change.
• With the values above, the API policy's ingress rule gets the expected from block.