Skip to main content

Open GPU orchestration
for Kubernetes.

Schedule, share and observe GPU workloads with operators and CRDs you already know how to run.

Get startedView on GitHub

GPU scheduling

Queues, priorities, gang scheduling and topology-aware placement for training and inference jobs.

Quotas & cost

Per-team GPU quotas, budgets and cost tracking, enforced by a Kubernetes operator.

Network intelligence

eBPF flow telemetry and GPU-communication visibility, exported to Prometheus and Grafana.

ML workflows

Multi-step pipelines, model registry, checkpoint guards and inference services as custom resources.

CLI, SDKs & dashboard

A Rust CLI, Go and Python SDKs, and a web dashboard over the same API.

Kubernetes-native

Everything is a CRD and an operator. Install with Helm; no separate control plane.

Early, and honest about it

Gryvia is in active development. The API is gryvia.io/v1alpha1, and it may change before a stable release. Performance figures are targets until they are backed by published benchmark results.

Read the roadmap →