Run agent-driven Amazon SageMaker HyperPod operations with InstantStart
aws ·
Amazon SageMaker HyperPod InstantStart is an open-source control plane that manages multi-stage HyperPod operations such as cluster bootstrap, capacity, training, inference, and storage through both a web UI and an AI agent interface. It runs as a single container in the user's AWS account, calling AWS and Kubernetes APIs without being in the data path. The system separates concerns into four layers—infrastructure, capacity and resilience, workloads and data, and interfaces—while relying on AWS-managed HyperPod capabilities for health monitoring and recovery. Its MCP tools wrap the backend REST APIs, ensuring the agent and console share identical validation logic. The agent can create a cluster from a single command, polling long-running operations to completion and asking users only decision-grade choices like zone, instance type, and capacity. The design emphasizes retryable, staged workflows so failures in later stages do not roll back earlier successes, and it codifies operational rules into the control-plane API for dependable automation.
aws