Endpoint

An endpoint is an addressable access point for a service. In an HTTP API, developers commonly identify an operation by its request method and URL path. In managed model hosting, “endpoint” can also name a deployed serving resource that routes requests to model workers.

One address can represent many workers

Suppose an application sends a prediction request to a service's prediction path. A routing layer may send it to one of several workers, possibly using different approved model versions during a rollout. The visible endpoint need not identify one machine or one permanent model version.

SageMaker's deployment documentation illustrates hosted inference endpoints as well as inference modes that do not require the same interactive request pattern. An endpoint is an implementation access point, not the model artifact or the full API contract.

Reachability is only one check

The application still needs the expected request format, authentication, authorization, response interpretation, and timeout policy. A health endpoint answering successfully does not establish that prediction requests can retrieve their features or use the required model.

An endpoint may be private to a network or exposed to authorized external clients. Calling it an endpoint says nothing by itself about public accessibility or permission to use it.

Reference: Amazon SageMaker: Deployment for inference.