Entities#

Entities are the underlying data objects that power the NeMo Platform. Models, datasets, jobs, configurations, and other resources are all represented as entities in the platform’s storage layer.

How Entities Work#

Each service in the platform—Customizer, Evaluator, Guardrails, and others—stores its resources as entities. This shared foundation provides consistent behavior across services:

  • Common metadata — All entities have standard fields like name, workspace, timestamps, and custom fields

  • Unified storage — Resources from different services live in the same entity store

  • Consistent operations — Create, read, update, and delete follow the same patterns

Workspace-Scoped Entities#

Entities are always scoped to a workspace. When you create or reference an entity (for example, a model, a fileset, or an evaluation job), it belongs to the workspace you are using in the CLI or SDK. Entity names are unique within a workspace, and access control is applied at the workspace level—so the same entity name can exist in different workspaces for different teams or projects.

Working with Entities#

In most cases, you interact with entities through service-specific APIs rather than a generic entity API. The platform CLI and Python SDK expose these through familiar commands and client methods:

  • Customizer — Create and manage fine-tuning jobs, model entities, and filesets. See Customizer overview and the customization job guides.

  • Evaluator — Create metrics, benchmarks, and evaluation jobs; retrieve results. See About Evaluating and the evaluation tutorials.

  • Models and inference — Register model entities, create deployments, and run inference through the gateway. See About Models and Inference and the inference tutorials.

These service APIs provide typed interfaces and validation specific to each resource type, built on top of the entity system. For raw HTTP access, see the API reference.