← All products
Infrastructure & Operations

Parinita Maestro

Kubernetes lifecycle for multi-silicon AI edge — 909+ clusters, 101 POPs, zero-touch.

Maestro is the Kubernetes lifecycle platform purpose-built for Parinita's nine-plane infrastructure. It manages 909+ RKE2 clusters (plus K3s on edge planes) under 11 immutable per-plane cluster profiles — the first K8s management platform purpose-built for multi-silicon AI edge.

Talk to us See the platform
Capabilities

What it does

  • Per-plane cluster profiles

    Each plane (Gaudi 3, Blackwell, EPYC Turin, Sierra Forest, NVMe, AmpereOne, etc.) has a versioned, immutable cluster definition pinning resources, GPU plugins, kernel params, and network policies.

  • Zero-touch bootstrap

    From Instrument node registration to seats available in Orchestra — RKE2 bootstrap, device plugin install, registry join, all under 12 minutes per node.

  • Precise northbound API

    gRPC/REST consumed exclusively by Orchestra — list clusters with capacity and health, provision a namespace + ResourceQuota for a seat purchase, deprovision on churn, query node health.

  • Self-healing operations

    A/B image distribution with auto-rollback, self-healing node replacement, certificate rotation, etcd snapshot/restore, continuous profile drift detection — all autonomous.

  • No external dependency

    Maestro's control plane runs on the infrastructure it manages. Even at a regional outage, local Maestro keeps reconciling.

How it works

Maestro’s core innovation is per-plane cluster profiles: each plane has a versioned, immutable definition pinning resource limits, GPU device plugins, kernel parameters, and network policies appropriate for that silicon and workload class. A profile change triggers a controlled canary rollout — never a manual reconfiguration of hundreds of clusters.

On first boot, the Instrument runtime agent registers each node with Maestro. Maestro validates the silicon ID and POP, selects the correct cluster profile, bootstraps RKE2 (or K3s on Plane 7), installs the right GPU/NPU device plugins, joins the registry, and signals Orchestra that N new seats are available. Bootstrap completes in under 12 minutes.

Five operational capabilities run continuously without human intervention: A/B image distribution with auto-rollback, self-healing node replacement, certificate rotation, etcd snapshot/restore, and continuous profile drift detection. Every cluster lifecycle event — provision, upgrade, cert rotation, image push — anchors on Chrysalis as an immutable record.

When to use it

  • Multi-silicon Kubernetes fleets where drift between sites is a recurring incident source.
  • Operators who want progressive rollouts of cluster changes (not just workloads) with built-in rollback.

What it isn’t

A workload controller. Maestro manages clusters; your manifests, charts, and operators deploy through standard K8s APIs as usual.

More from the fabric

Related products

Part of the Parinita AI Edge

Bring Parinita Maestro into your stack.

Every Parinita product runs on the same 9-plane fabric across 101 edge POPs. Talk to us about a pilot, or see how the pieces fit together.

Talk to us See the infrastructure