I’m checking the repo shape, the docs tree, and any existing AICODE- anchors first, then I’ll trace how local multi-listener mode is wired into the current Nix deployment.
I have the initial map. Next I’m reading the deployment docs and the Nix/service wiring, then I’ll trace how the app binds the control-plane and tenant listeners so the remote setup matches local mode.
The current deployment only provisions one socket and the app blindly takes the first activated listener, so remote mode collapses both surfaces into one port. I’m checking the go-systemd API in the local module cache before changing the service contract.
I found one extra wrinkle: the Nix service is still exporting BGN_ADDR, but the app now reads BGN_API_ADDR and BGN_VMS_ADDR. I’m running the test/build baseline now so I can separate pre-existing breakage from the socket/deploy fix.
I’ve isolated the main technical constraint: with two systemd listeners, the current code would lose the second FD because go-systemd/activation unsets the e