KServe predictors add support for custom model container images

KServe predictors can now serve models from custom container images, so any model that packages as a container can run on an OpenEverest KServe instance — not only the models from the built-in catalog.

For proprietary or fine-tuned models that are not in the catalog, point the predictor at your own image: the deployment gets the same predictor topology, monitoring, and traffic management as catalog models.

Custom model images are available in provider-kserve v0.4.0. To learn more, visit the release notes.