Model Storage

DynamoGuard model weights are pulled from storage at runtime. Three patterns are accepted; pick the one that matches your environment's egress and storage posture.

Pattern	Requires	Best For
Direct download from the DynamoAI HuggingFace repository	Egress to HuggingFace from the cluster	Environments with permissive egress
In-cluster MinIO	PersistentVolume support (default StorageClass)	Air-gapped or egress-restricted environments
External object store	An object store the cluster can reach, populated with the model artifacts	Environments with an existing artifact store

Whichever pattern you choose, the model store remains read-only to the platform and must stay separate from the writable platform-artifact store; see the object storage dependency and the AccessDenied symptom that results from mixing them.

Model artifacts are shared with you at deployment time; your DynamoAI deployment engineer provides the transfer scripts for the mirroring patterns.