What problem are you facing?
The control-plane InferenceGateway has no way to set a hostname or configure TLS, and neither does ModelService. A model's endpoint comes back as an assigned IP in status.address, for example http://172.18.255.200/ml-team/qwen, so callers have to scrape the IP and talk plain HTTP. For a product whose core value is a single front door to the fleet, there is no way to give that door a stable name or terminate TLS on it, which makes it awkward to put in front of real clients or behind DNS.
How could Modelplane help solve your problem?
Let the InferenceGateway declare one or more hostnames and a TLS configuration, for example a listener hostname plus a certificate Secret reference following the Gateway API listener shape, and have ModelService publish status.address using that hostname. Then a stable, TLS-terminated URL fronts the fleet instead of an ephemeral cluster IP.
What problem are you facing?
The control-plane InferenceGateway has no way to set a hostname or configure TLS, and neither does ModelService. A model's endpoint comes back as an assigned IP in
status.address, for examplehttp://172.18.255.200/ml-team/qwen, so callers have to scrape the IP and talk plain HTTP. For a product whose core value is a single front door to the fleet, there is no way to give that door a stable name or terminate TLS on it, which makes it awkward to put in front of real clients or behind DNS.How could Modelplane help solve your problem?
Let the InferenceGateway declare one or more hostnames and a TLS configuration, for example a listener hostname plus a certificate Secret reference following the Gateway API listener shape, and have ModelService publish
status.addressusing that hostname. Then a stable, TLS-terminated URL fronts the fleet instead of an ephemeral cluster IP.