Datadog now integrates with Kubernetes Custom Resources that orchestrate AI inference (Dynamo, Kuberay and KServe) and surfaces them, and their controlled objects, as an Inference Workload in the Kubernetes Explorer and MCP.
The Inference Workload is a stable, aggregated representation of inference across a Kubernetes cluster. Model metadata, serving performance, controlling CRD logic, and the cluster’s disparate objects (pods, services, nodes, etc) are all linked into one single source of truth.
Datadog collects, attributes and analyzes data from every object and links them to the orchestration logic in the framework’s Custom Resource Definition.
To qualify for this preview you should be using Infrastructure Monitoring with Containers on supported inference frameworks.