Swappable Gateway scope: L4 network policy or external PDP interface (L7)?

[philippeabraxas-jpg]

Hi AX team,

I've been reading the roadmap and the Gateway docs. Two things caught my attention:

The roadmap mentions a "Swappable Google-Managed Gateway" — abstracting the data/control plane so clusters can swap between the built-in Substrate gateway and a Google-managed one.

The current Gateway spec is an explicit host/port egress allowlist.

My question: when you say "swappable", do you mean swappable only among Google-managed implementations, or is the intent to eventually support an external Policy Decision Point (PDP) — e.g., a webhook, OPA sidecar, or gRPC service — that the Gateway consults per outbound request?

Context for asking: I work on a deterministic action-level policy layer that judges what an agent does (not what it reasons), and exposes a PDP interface. If AX's Gateway is meant to be a PEP with a pluggable PDP, that's a natural integration seam. If not, knowing that early saves everyone time.

Two concrete sub-questions:

Is there a hook today (even experimental) to intercept an outbound request before it leaves the sandbox, beyond the static allowlist?

In the "Swappable Gateway" work, does the abstraction boundary sit at the network-policy level (host/port) or at the decision level (allow/deny per request)?

Happy to write a short design note on what an external PDP contract could look like if that's useful. No code, just the seam.