Hanzo Visor: Cloud Infrastructure and VM Management

Running AI at scale requires compute infrastructure that can grow from a single GPU to a fleet of H100s without changing how you manage it. Hanzo Visor provides that unified control plane.

What Visor Manages

Virtual machines. Provision, configure, and manage VMs across compute tiers — from cost-optimized CPU instances for inference to GPU clusters for training. SSH access, cloud-init configuration, and lifecycle management from a single interface.

Kubernetes clusters. Visor deploys and manages Kubernetes clusters on Hanzo's infrastructure. Scale the control plane and worker nodes independently. Persistent volumes backed by Hanzo's distributed storage.

Networks and security groups. Define network topology, configure ingress rules, and manage private networking between services. VPC-style isolation for multi-tenant deployments.

Storage. Block storage for VMs, object storage for model weights and datasets, and network-attached volumes for shared workloads. Replicated across availability zones.

Built for AI Workloads

Visor includes profiles optimized for common AI deployment patterns:

• Inference clusters: Auto-scaled GPU pools with load balancing across model replicas
• Training jobs: Spot instance scheduling for cost-efficient training runs
• Embedding pipelines: CPU-optimized instances for batch embedding at scale
• Fine-tuning environments: Persistent GPU instances with attached NVMe storage

The Visor API integrates with Hanzo's MLOps tooling — model deployments, experiment tracking, and serving infrastructure — making compute management part of the AI development workflow rather than a separate operations concern.

Available at github.com/hanzoai/visor.