Bare metal to custom production clusters in 48h

Aranya is the multicluster operating system that turns any bare metal into custom production-ready clusters in 48 hours. We automate all of the moving parts in GPU orchestration, from individual kernel drivers to federation across clusters. Powered by clusterdOS, our open-source distributed OS engine.

clusterdOS is the open-source engine that lives inside each cluster and runs the core cluster-level applications. Aranya builds on top of those applications and runs both within and across your clusters, federating them so every cluster you run operates as one system.

On the demand-side, AI inference and training companies who cannot wait weeks for production clusters. Instead of working with datacenter hardware or generic managed Kubernetes, Aranya automates GPU orchestration, from kernel drivers to federation across clusters. On the supply-side, neoclouds and datacenters at 20MW and below who have the hardware but not the ability to turn it into compute for AI workloads. Aranya brings demand to your capacity without a platform team of your own.

If you have the hardware but not the platform layer to turn it into compute for AI workloads, Aranya closes that gap. We turn your bare metal into custom production-ready clusters in 48 hours and bring the demand side to your capacity. No platform team of your own, and no losing a customer because you can't match their specs. Built for neoclouds and datacenters at 20MW and below.

Every cluster is architected around your requirements: custom networking, storage, GPU scheduling, and inference or training setup. Kubernetes, VMs, Slurm, or another orchestration layer entirely, we build for the architecture your workload actually needs. A high-throughput inference deployment looks different from a distributed training setup. That starts before deployment: Aranya specifies the hardware and infrastructure requirements your data center needs to meet, then deploys and runs the full stack on top.

Under 48 hours from bare metal to production-ready.

clusterdOS is the open source distributed OS engine built on Kubernetes and extending to VM and Slurm environments. K8s is the kernel; clusterdOS is the core cluster-level applications on top of it. GitOps-native configuration, inference and training tooling, storage, networking, and GPU management, all production-configured out of the box. Push to Git; the cluster manages itself.

Different hardware vintages, configs, and failure modes across providers, with no standard layer making that variety perform consistently. Aranya is that layer. We specify the hardware and networking requirements up front, work directly with the datacenter to meet them, then run Aranya multicluster OS to discover, diagnose, and resolve failures automatically.

EKS, GKE, and AKS give you a generic Kubernetes cluster, then the rest is on you. They're not built for the configuration demands of inference or training workloads: custom GPU scheduling, high-throughput storage, specific networking setups. You're also paying hyperscaler margins the whole time. Aranya fully managed clusters takes you from any bare metal to production in 48 hours, with the full production stack configured for your specific workload. No generic template, no platform team required to run it.

Yes, GPLv3-or-later. We believe in making the future of infrastructure transparent and community-driven. You can find us on GitLab and we welcome contributions.