YARS.ai
PSU & Enterprise Private Cloud

Your Infrastructure. Sovereign. Resilient. Yours.

Private cloud built entirely inside your building, on your hardware — delivering public cloud agility with government-grade security and zero vendor dependency.

Active deployments · PSU & Enterprise clients · 2026
0
Hypervisor licence fees — ever
RPO=0
Active-active stretched DR
<2min
Automated site failover RTO
Cluster Operations — Live3 nodes healthy
Control Plane ×3etcd quorumHealthy
VM Provisioning Agentpsu-client-db-02Running
Ceph Storage ×3 replicasHEALTH_OKHealthy
Migration — PSU VMware estate14 VMs completeHealthy
Compliance Report AgentScheduled 06:00Queued
16:04:31 [AGENT] psu-client-db-02 scheduled on worker-node-216:04:29 [CEPH] OSD rebalance complete · 0 degraded16:04:18 [MIGRATE] psu-oracle-vm-14 import succeeded
Compliance-ready for
FIPS 140-2STIGCERT-InDPDP ActRBI
ISO 27001CMMIMSMEDIPP
Why YARS

Six Outcomes Few Other Managed Clouds Deliver Together

Each outcome required a deliberate architectural decision. Most managed offerings give you three or four. YARS delivers all six — on hardware you own, at a cost structure that scales with revenue.

💰

Zero Hypervisor Licensing — Ever

The entire virtualisation and orchestration stack runs on open-source software. No per-socket, per-VM, or per-node fees. Infrastructure cost is hardware cost — nothing more, permanently.

₹0
annual licence fees
100%
open source
🔁

Active-Active DR With RPO = Zero

Synchronous replication across two physical sites. Every write confirmed at both locations before the application sees success. Complete site failure triggers automatic workload migration in under two minutes — zero data loss.

RPO=0
sync writes
<2min
auto failover
🛡️

Security-Grade Isolation at Full Hardware Utilisation

Kubernetes control plane and client workloads are isolated at the hypervisor layer — hardware-enforced, not just software-scheduled. No idle hardware wasted on separation. No shared kernel surface.

HW-enforced
isolation boundary
90%+
utilisation
📋

Government Audit-Ready on Day One

The infrastructure OS ships with FIPS 140-2 cryptography and official government hardening profiles built in. PSU IT auditors and CERT-In compliance teams accept it without exception requests or extended review cycles.

FIPS 140-2
built-in
STIG
hardened profiles

Legacy VMs and Modern Apps — One Platform

A decade-old database VM and a freshly built microservice run on the same cluster, deployed through the same pipeline, monitored by the same dashboard, backed up by the same policy. No separate tooling per workload class.

One
pipeline for all
One
monitoring stack
🤖

AI-Managed Operations — Not Ticket Queues

Provisioning, incident response, compliance reporting, and migration orchestration handled by AI agents connected to live infrastructure. Clients get outcomes in seconds. Engineers govern policy. Agents execute everything else.

<90s
request to live VM
70%↓
incident MTTR
Use Cases

Problems We Solve End-to-End

Real institutional challenges delivered on hardware the client owns.

PSU Sovereign Infrastructure

Multi-Tenant Sovereign Cloud — Multiple PSU Clients on Shared Infrastructure

Multiple government entities operate in cryptographically isolated environments on the same physical cluster. Each PSU client has a completely separate identity realm, dedicated VLAN, separate storage pool, and independent audit trail — invisible to every other tenant on the same hardware.

🔒
Network isolation enforced at silicon level

Each tenant's VMs communicate only within their assigned VLAN. Policy enforced at the kernel level — no software bypass possible.

🪪
Separate identity realm per client

Each PSU client logs in via their own identity realm connected to their own Active Directory. Users from one organisation cannot see resources belonging to another tenant.

📊
Per-tenant dashboards and resource quotas

Each client has a read-only view showing only their own VMs and resource consumption. Quotas prevent any single tenant from starving others of compute or storage.

Outcome

Multiple PSU clients sharing the same physical hardware — each experiencing their own dedicated private cloud with zero cross-visibility.

Key Parameters
Tenant isolationHardware + Software
IdentitySeparate AD-connected realms
NetworkDedicated VLAN per tenant
Audit trailPer-tenant log stream
Stack Used
RHEL 9RKE2KeycloakNetworkPolicyCeph RBDGrafana
Agentic AI Operations

Infrastructure That Operates Itself

AI agents connected directly to the cluster API handle provisioning, incident triage, compliance, and migration — reducing toil and response time from hours to seconds. Engineers set policy. Agents execute.

What the Agents Do

Every agent operates with a defined scope, a human approval gate for irreversible actions, and a full audit trail. No black-box automation — every decision is logged, attributable, and reviewable.

🖥️
Provisioning Agent

Plain-language VM and namespace requests translated into live Kubernetes operations. Portal request to running VM in under 90 seconds.

🚨
Incident Triage Agent

Alerts arrive with root cause analysis and recommended action — not raw metrics requiring manual interpretation.

📋
Compliance Agent

STIG scans, certificate checks, and policy deviations compiled into audit-ready reports — triggered automatically before audit dates.

🔄
Migration Agent

VM export, disk conversion, Kubernetes import, and health verification — orchestrated end-to-end with one engineer approval gate.

Human-in-the-Loop by Design

Agents handle reversible actions autonomously. Irreversible actions always pause at an approval gate — the engineer acts on full context, not a blind prompt.

Agent Decision Model
Autonomous

VM start/stop, metric queries, draft generation, log analysis, scheduling

🚦
Approval Gate

DNS cutover, production deployment, node drain, storage operations, tenant creation

🔒
Engineer Only

etcd operations, cluster upgrades, stretched DR configuration, security policy changes

VM Provisioning Agent
📥
Request received
Plain-language request parsed from client portal
🤖
LLM spec extraction
Parameters mapped to Kubernetes resource model
📋
Quota validation
Checking namespace ResourceQuota allows allocation
📝
VirtualMachine YAML generated
KubeVirt manifest with cloud-init, network, and storage
⚙️
kubectl apply — VM scheduled
Kubernetes selects optimal worker node automatically
🔑
Identity provisioned
User account created, SSH key injected via cloud-init
VM live — credentials delivered
Total time: 87 seconds from request to running VM
Audit LogIdle

Run the agent to see live audit entries

Context

Client request via portal: 'Provision Ubuntu 22.04, 8 vCPU, 32GB RAM for the analytics team.' Agent extracts spec, validates namespace quota, generates the VirtualMachine manifest, applies it to Kubernetes, provisions identity credentials, and delivers access — no engineer interaction required.

Resilience Architecture

DC + DR — One Cluster. Two Sites.

Not two systems kept manually in sync. One Kubernetes cluster stretched across both locations with synchronous storage replication and automatic failover built into the storage layer itself.

Single-Site High Availability

Baseline for every YARS deployment. Any single hardware component — disk, node, or network card — can fail and workloads continue without interruption or data loss.

3
Control Planes
Data Copies
40s
Failure Detection
Control plane node failure — quorum maintained on 2 remaining nodes
Disk failure — data intact on two other nodes, zero storage interruption
Worker node failure — VMs rescheduled automatically within 40 seconds
Planned maintenance — VMs live-migrated off node before shutdown, zero downtime

Stretched Cluster — Active-Active DR

One cluster spanning both datacenters. Storage writes synchronously to both sites before confirming success. DR hardware runs live workloads every day — not idle hardware waiting for a disaster.

RPO=0
Data Loss
<2min
RTO
Auto
Failover
Every write confirmed at both sites before application sees success — zero data loss on site failure
Site failure triggers automatic workload rescheduling to DR in under 2 minutes
DR hardware runs productive workloads daily — not idle standby
Single dashboard, single kubectl — not two systems to manage
Technology Stack

100% Open Source. 100% Production Grade.

Every component is community-backed, battle-tested, and carries zero licensing cost. Your infrastructure — permanently owned.

OS
RHEL 9

FIPS 140-2 cryptography and STIG hardening profiles on every physical server. The OS accepted without question by PSU IT auditors and CERT-In compliance teams.

FIPS 140-2STIG
Hypervisor
KVM + libvirt

Hardware-accelerated virtualisation built into the Linux kernel. 95–98% bare-metal VM performance. Zero additional licensing — the same engine underneath every major hypervisor product.

Zero CostKVM
Kubernetes
RKE2

Hardened, FIPS-compliant Kubernetes distribution. Unifies all nodes into one cluster — one API, one resource pool, one deployment model across VMs and containers.

FIPSRKE2 v1.29+
VM Layer
KubeVirt + CDI

VMs as native Kubernetes objects. CDI imports existing VMware, Oracle VM, and Hyper-V disks directly into the cluster without application changes.

KubeVirtCDI
Storage
Ceph via Rook

Distributed block and object storage with replication factor 3. Synchronous stretched replication across two datacenters for RPO = 0. Managed declaratively via the Rook operator.

RPO=0Ceph Reef
Platform
Rancher · Keycloak · ArgoCD · Harbor

Management dashboard, identity and tenant isolation, GitOps deployment, and private container registry — wired as one coherent platform.

RancherKeycloakArgoCD
AI Ops
Agentic Workflows

AI agents connected to the Kubernetes API for automated provisioning, triage, compliance, and migration — with human approval gates for all irreversible actions.

AI Agentskubectl API
Who We Serve

Built for the Organisations That Cannot Compromise

Every vertical has its own compliance regime, threat model, and operational constraint. YARS Private Cloud adapts to each — same platform, different configuration profile.

🏛️

PSU & Government Bodies

State-owned enterprises and government departments requiring sovereign infrastructure — FIPS-compliant, STIG-hardened, auditor-accepted on day one.

FIPS 140-2STIGCERT-In
🗄️

Government Data Centres

NIC and state DCs consolidating legacy hypervisor estates onto open-source KVM + RKE2. Full DC + DR stretched cluster across two sites with RPO = 0.

NIC-readyDC + DRRPO=0
🤖

Private AI-as-a-Service

Cloud providers and enterprises offering private LLM inference to regulated clients. GPU slices allocated per tenant — prompts never touch an external API.

LLM InferenceMulti-tenantAir-gapped

GPU & AI Infrastructure

A100 / H100 / L40S GPU clusters with MIG partitioning, Kubernetes-native autoscaling, and OpenAI-compatible inference API — inside the client's building.

A100 / H100MIGvLLM
🏦

BFSI & Regulated Finance

Banks, NBFCs, and insurance companies with RBI data-localisation obligations. Workloads stay on owned hardware with cryptographic audit trails and per-tenant isolation.

RBI CompliantData LocalisationAudit Trail
🏥

Healthcare & ABDM

Hospitals and health-tech platforms storing patient records under ABDM and DPDP Act requirements. Sovereign compute with end-to-end encryption and role-based access.

ABDMDPDP ActPHI Isolation
⚙️

Critical Infrastructure

Power, water, telecom, and defence adjacent organisations requiring air-gapped deployments with zero external network dependency in production operations.

Air-gapZero InternetSCADA-safe
🎓

Higher Education & Research

Universities and research institutions running HPC alongside student services — multi-tenant compute with per-department quotas and self-service VM provisioning.

HPCMulti-tenantSelf-service
🔄

VMware & Oracle VM Exit

Enterprises exiting VMware post-Broadcom acquisition or rationalising Oracle VM licensing. Full estate migrated disk-level — zero application changes, two-week rollback guarantee.

VMware ExitOracle VMZero Downtime
Get Started

Ready to Build Your Private Cloud?

Tell us about your infrastructure requirements, your clients, and your timeline. We'll respond within one business day with an architecture outline and engagement proposal.