The company is also shipping the clusterdOS v0.4 series, with per-engine inference stacks, opt-in observability and alerting, and a networking and security refresh for production GPU clusters.
SALT LAKE CITY — November 10, 2026 — Aranya, the AI-native multicluster operating system bridging the gap between what data centers offer and what AI workloads really need, today announced it will give the first public demo of Aranya OS at KubeCon + CloudNativeCon North America 2026. Aranya OS is built on clusterdOS, Aranya’s open-source engine built on Kubernetes. It gives teams plain-language control and visibility across every cluster they run. Requests resolve to the actions the requesting user is already permitted to take, so a request in Slack can do nothing an engineer couldn’t already do.
In a live demo at booth #369, attendees will watch Aranya OS operate a GPU fleet in response to plain-language instructions to:
- Diagnose and resolve a hardware failure before it cascades into downtime
- Get a live view of every cluster in the fleet from one place
- Run a request from Slack that stays within the permissions the engineer already has
“Getting a Kubernetes cluster to production is the easy part. Keeping it current, observable, and recoverable across a fleet of heterogeneous GPU hardware is where platform teams lose their time,” said Christian Bhatia Ondaatje, co-founder and CEO of Aranya. “clusterdOS handles that lifecycle inside each cluster, and Aranya OS lets a team direct the whole fleet at once. KubeCon is the first time we’re showing it publicly.”
clusterdOS v0.4: what’s new
clusterdOS delivers each component as a GitOps-managed “gitapp” through Argo CD, enabled with high-level feature flags. The v0.4 series focuses on inference, observability, and networking:
- One gitapp per inference stack. Inference gitapps are now named for the stack they deploy: llmd for llm-d, sglang for SGLang, and a new engine-only vllm for vLLM. The rename is a breaking change, and the upgrade notes cover the migration path for existing deployments.
- Alerts that reach a human. Previously, the bundled Alertmanager fired alerts, but by default they reached no one. A new opt-in notifications block routes alerts to Slack or any webhook. The webhook URL is kept in a Kubernetes Secret the user owns, not in values files or git.
- Observability, GPU health, and backups. New opt-in components include:
- A self-hosted metrics backend
- NVSentinel and node-problem-detector, for detection-only GPU and node health monitoring
- Velero, for provider-agnostic backups
- MinIO, as a disposable S3-compatible cache
- Networking and security refresh. Fourteen external charts have been brought current with upstream, including:
- the Tailscale operator, moving to 1.102.3 and picking up eight upstream security bulletins
- cert-manager v1.21.1
- Traefik 40.1.0, deliberately held below 41.x so that Argo CD pruning can’t delete a cluster’s live Gateway API CRDs
- Safe by default. Every new component ships disabled, so enabling one is a deliberate per-cluster decision. A misordered upgrade fails loudly at render time instead of silently deploying nothing. Major chart versions advance one per release, so a bad upgrade can’t strand two at once.
Aranya already manages more than $500M in GPUs for leading inference providers. In its deployment with Hydra Host, an NVIDIA Cloud Partner, Aranya cut cluster setup from the industry-standard two to six weeks to under 48 hours and reduced cluster downtime by 90%.
clusterdOS v0.4.2 is available now. To see Aranya OS, visit booth #369 at the Salt Palace Convention Center, November 10–12. To join the waitlist for Aranya OS, visit aranya.tech/#waitlist.
###
About Aranya
Founded by Christian Bhatia Ondaatje, Sasivarnan Kanaghasalam Sathyapriya, and Aryamika Bhatia Ondaatje, Aranya converts bare metal into custom, production-ready clusters in under 48 hours for fast-growing inference providers, AI startups, and data centers. Powered by clusterdOS, Aranya’s open-source distributed OS engine built on top of Kubernetes, which turns any quantity of raw bare metal servers into a self-healing, enterprise-grade, cohesive whole, reducing average downtime by 90%. Its proprietary technology, the AI-native multicluster operating system, gives teams direct, plain-language control over how inference or training is designed, scaled, and operated. The company is building toward a future where no AI company is limited by the gap between what they need and what their infrastructure can deliver. For more information, visit aranya.tech.
Resources for Aranya Users
- Release tag tree: https://gitlab.com/aranya-tech/public/clusterdos/-/tree/v0.4.0
- Upgrade notes: https://gitlab.com/aranya-tech/public/clusterdos/-/blob/v0.4.0/README.md
- Velero docs: https://gitlab.com/aranya-tech/public/clusterdos/-/blob/v0.4.0/docs/velero.md
- Bootstrap chart: helm show chart oci://registry.gitlab.com/aranya-tech/public/clusterdos/charts/clusterdos –version 0.4.0
- v0.4.2 release tag tree, promotion MR and observability chart README: [links needed from client]
- Upstream: https://velero.io/docs/ · https://tailscale.com/kb/1236/kubernetes-operator · https://github.com/kubernetes-sigs/gateway-api-inference-extension · https://docs.victoriametrics.com/operator/resources/vmalertmanagerconfig/ · https://github.com/traefik/traefik-helm-chart/issues/1669
Media Contact
Tess Pawlisch
608-333-9788
aranya@needlepr.com
