1 Considerations #
Deploying SUSE AI Factory in an air-gapped environment requires local artifact registries, a local model cache, and secure TLS endpoints. These local resources ensure that you can deploy and run AI applications securely without external internet connectivity. When planning your deployment, ensure that you consider the following:
OCI artifact repositories. In a connected environment, SUSE AI Factory pulls artifacts from external sources to deploy AI applications. These artifacts include Helm charts and container images. In an air-gapped environment, you must host these artifacts locally because external network access is prohibited. Mirror or import these artifacts to local OCI-compliant repositories. Configure the local repositories similarly to the external sources to ensure a seamless transition.
Model cache. AI applications can consume AI models from external sources, such as Hugging Face or NVIDIA NGC. AI models consist of binary files arranged in specific directory layouts. To use a model cache, store model data in Persistent Volume Claims (PVCs) attached to the application pods. For multi-node or multi-cluster deployments, the PVC provisioner must support distributed storage, such as NFS or Longhorn.
Endpoint security. Although you can use self-signed TLS certificates for local OCI-compliant repositories, we recommend using certificates signed by your organization’s certificate authority (CA).