Skip to main content
EnterpriseAI Platform

Architecture overview

HAMi Enterprise runs inside Kubernetes and provides a closed loop across GPU discovery, scheduling extensions, admission control, runtime isolation, and metrics collection.

HAMi AI Platform adds a platform control plane for multi-cluster lifecycle management, tenant quotas, permissions, alerts, resource views, and operations data.

Layers​

  1. Device and node layer: GPU drivers, device plugins, and heterogeneous accelerator adapters.
  2. Scheduling and admission layer: scheduling policies, resource validation, and workload mutation.
  3. Runtime layer: memory isolation, compute allocation, and high-performance modes.
  4. Platform layer: clusters, tenants, quotas, alerts, audit, and self-service entry points.