Open-source distributed operating system that bootstraps, maintains, and upgrades production GPU Kubernetes clusters.
clusterdOS is an open-source engine that runs inside a single cluster and assembles the components a production Kubernetes deployment needs, including Cilium eBPF networking, Rook Ceph storage, cert-manager TLS automation, Traefik ingress, and both NVIDIA and AMD GPU operators. Aranya publishes it under GPLv3-or-later in a public GitLab repository, and operators install it either from a Helm chart in the project OCI registry or from a generated install.yaml manifest.
Configuration is GitOps-native: ArgoCD watches a Git repository and reconciles the cluster to the declared state, while individual components, which the project calls gitapps, are switched on through Helm values and pinned to a tagged release or tracked against the main branch. Coverage extends past containers to virtual machines through KubeVirt, to batch and HPC scheduling through a Slurm operator, and to model serving through vLLM and llm-d.