VNClagoon Infrastructure

Sovereignty needs ground to stand on.

The complete sovereign digital workplace — and the AI factory beneath it — deployed on your metal, in your data center, at any provider, in any country. Declared in Git. Deployed by ArgoCD. Mirrored anywhere on earth.

On-premisePrivate cloudAny providerAir-gap capable
100%

Data sovereignty — every byte stays on infrastructure you control

Horizontal scale-out — from pilot cluster to national platform

0

Vendor lock-in — open source, open standards, portable by design

<1s

Public IP failover when a node goes down

Deploy Anywhere

Your infrastructure. Your jurisdiction. Your rules.

VNClagoon is not tied to a cloud, a provider, or a country. The entire stack — workplace, data services, and AI — installs identically wherever you decide it should live.

Who builds sovereign platforms on it

MSPs & Hosting Providers

Sovereign workplace as a service

One Git repository per tenant stack — identical, isolated, fully auditable. Onboard your first customer and your hundredth on the same automation, and operate them all from a single GitOps control plane.

Telcos & Carriers

Turn data centers into sovereign cloud regions

Offer national-scale collaboration and AI to enterprise and public-sector customers — on facilities and network you already own, under the jurisdiction your customers require.

Corporates

The workplace inside your perimeter

Bring communication, collaboration, and the AI factory behind your own firewall. Compliance is met by architecture, not by contract clauses — no data processor outside your control.

Governments & Public Sector

From ministry cluster to national platform

Air-gap capable, GDPR- and BSI-aligned by design. Every byte, every model, every log under national jurisdiction — with an open-source stack your own teams can inspect.

GitOps Automation

The entire factory, declared in Git.

Every component of a VNClagoon stack — cluster add-ons, workloads, PostgreSQL clusters, model configurations, prompt templates, tool definitions — is YAML in a Git repository. ArgoCD reconciles the running platform against it, continuously.

01

Reproducible

Point the same repository at new hardware and get an identical stack — in hours, not months. That is how a stack is born, and how it is mirrored.

02

Auditable

Every change to the platform is a commit: who, what, when, why. Your auditors read history instead of interviewing admins.

03

Recoverable

The repository is the disaster-recovery plan. Rebuild the platform from Git and restore data from replicated backups — a rehearsed procedure, not a hope.

 platform/argocd/app-of-apps.yaml
apiVersion: argoproj.io/v1alpha1
kind: Application
metadata:
  name: vnclagoon-platform
spec:
  source:
    repoURL: https://git.yourdomain.tld/platform
    path: stacks/production
  destination:
    server: https://kubernetes.default.svc
  syncPolicy:
    automated:
      selfHeal: true      # drift auto-corrected
      prune: true         # removed = removed
---
# workplace · data services · AI engine ·
# observability — one repo, one source of truth.

workplace · data services · AI engine · observability — one repo, one source of truth.

No Technology Lock-In

Kubernetes your way. Bare metal where it counts.

The stack is plain Kubernetes — standard manifests and Helm charts. It runs on the distribution you operate or your provider mandates, and steps down to bare metal where physics demands it: real-time video media relay and GPU inference run directly on the hardware.

MicroK8sK3s / RKE2SUSE RancherRed Hat OpenShiftOpenStackAny CNCF-conformant distributionBare metal for media & GPU workloads

Distribution-agnostic

Switch distributions without switching platforms. The manifests port; your investment survives every vendor decision — including ours.

Bare-metal video

Video conferencing media relay is CPU- and network-bound. It runs directly on dedicated metal — full cores, direct UDP, no container overhead — while staying wired into the same encrypted fabric.

GPU-native AI

Inference nodes join the cluster with direct GPU access. Kubernetes autoscaling spins model servers up and down with demand.

Federation & Scale

One fabric. Any number of locations.

A VNClagoon stack is never an island. Stacks mirror to any location, join into one encrypted fabric, and let workloads move across every connected destination.

Mirror

Stand up an identical stack anywhere on earth from the same repository — new region, new provider, new country.

Combine

Join stacks into one fabric over an encrypted WireGuard mesh — private connectivity even across untrusted networks.

Move

Workloads relocate across connected destinations. Capacity follows demand; data follows policy.

Grow

Scale-out is horizontal and unbounded: add a node, a rack, a region. Storage and compute scale independently.

WIREGUARD · ENCRYPTED WG WG SITE A · PRIMARY Your data center full stack · workplace + data + AI SITE B · MIRROR Regional provider same Git repo · warm standby or active-active SITE C · EXPANSION Private cloud region joined for capacity · pods follow demand
// one encrypted fabric — workloads relocate across every connected destination
Architecture

Inside a site: anatomy of a sovereign stack.

Every site is fully self-contained. Users enter through redundant virtual IPs, workloads run on interchangeable nodes, storage is replicated underneath — and AI and real-time media get the hardware they deserve. All of it declared in Git.

Entry

Redundant virtual IPs (MetalLB) with Traefik ingress and sub-second failover.

Compute

Interchangeable Kubernetes nodes; workloads reschedule automatically.

Storage

Ceph replicates every volume three times, self-healing underneath.

Accelerated

GPU inference nodes and bare-metal media relay wired into the same fabric.

Users · any network PUBLIC ENTRY MetalLB virtual IPs · sub-second failover Traefik ingress · automated TLS via cert-manager NODE 1 talk mail project files PostgreSQL · HA Ceph OSD NODE 2 calendar task directory contacts pgvector · Solr Ceph OSD NODE 3 registry git metrics argocd object storage · S3 Ceph OSD GPU NODE vLLM 70B+ RAG embed autoscaled inference Ceph OSD MEDIA · BARE METAL SFU SRTP direct UDP · full cores video relay without container overhead DISTRIBUTED STORAGE Ceph — every volume replicated ×3 across nodes · self-healing ×3 replicas WIREGUARD ENCRYPTED MESH — ALL NODE-TO-NODE AND SITE-TO-SITE TRAFFIC
// one site, fully self-contained — compute, storage, AI, and media on hardware you control
Embedded Data Services

Cloud services, without the cloud.

Everything a hyperscaler sells you as a managed service ships inside the stack — operated by Kubernetes operators, declared in Git, running on your hardware.

PostgreSQL, operated

CloudNativePG runs highly available Postgres clusters with automated failover and point-in-time recovery — the platform's transactional backbone.

pgvector

Vector embeddings live inside PostgreSQL. Conversation history, documents, and skill context become the AI factory's long-term memory — sovereign by construction.

Apache Solr

Battle-hardened full-text and faceted search for large-scale deployments with massive document repositories.

Distributed block storage

Ceph replicates every volume three times across nodes. A disk or node failure is a non-event — the cluster self-heals.

S3-compatible object storage

Backup targets, artifacts, and large objects on your own hardware — the same API as the public cloud, none of the exposure.

Sovereign registry & Git

Containers are built, signed, and served from your own registry. The software supply chain never leaves the perimeter.

Ingress & load balancing

Traefik ingress with MetalLB address management: public services stay reachable with sub-second failover between nodes.

Automated certificates

cert-manager issues and rotates TLS everywhere — public ACME or your internal CA.

Full observability

Prometheus and Grafana watch everything from disk health to GPU utilization and LLM token throughput.

Security

Secure by architecture, not by policy.

Security guarantees that depend on a provider's promise are not guarantees. VNClagoon's are structural — they hold because of how the stack is built.

Encrypted fabric, end to end

A WireGuard full mesh encrypts all control-plane and pod-to-pod traffic — confidential even inside a provider's own network.

Encryption at rest

LUKS disk encryption and storage-level encryption protect data on every drive, in every location.

Air-gap capable

The full platform — including on-device AI — operates with zero external connectivity for classified and high-security environments.

Sovereign supply chain

No runtime dependency on external registries or repositories. Every image is built and served in-house.

End-to-end encrypted communication

All channels across the workplace suite remain fully encrypted, with zero third-party data sharing.

Compliance built in

GDPR- and BSI-aligned architecture. Every AI action is logged, traceable, and reversible — audit evidence is generated by the platform itself.

Backup & Disaster Recovery

Engineered for the day something fails.

Resilience is layered by failure domain. Each layer has a defined mechanism and a rehearsed recovery path — because a backup that was never restored is a rumor, not a strategy.

METALLB RE-ANNOUNCES · <1S VIP VIP · public IP NODE 1 healthy · serving NODE 2 failed · workloads rescheduled NODE 3 healthy · takes the VIP
// public IPs survive node loss — clients barely notice
What failsWhat catches itRecovery
A diskTriple-replicated distributed storage self-heals and rebalances automatically.zero interruption
A nodeKubernetes reschedules workloads; MetalLB re-announces public IPs from surviving nodes.seconds
A volumeScheduled backups of every persistent volume to object storage; automated restore verifies and repopulates empty volumes on startup.minutes
A siteEncrypted off-site backup sync plus a mirrored stack deployed from the same Git repository — warm standby or active-active, your RPO.minutes–hours
A providerThe stack is provider-agnostic by design. Redeploy or fail over to any mirrored location — sovereignty includes the freedom to leave.your choice

Per-workload policy

Backup schedule, retention, and encryption defined per workload — in Git, like everything else.

Tested restore paths

Restores are automated and exercised through documented runbooks, not improvised during an incident.

Off-site by default

Encrypted backups sync to a second location as standard — independent of the primary site's fate.

Rolling everything

Upgrades roll one node at a time with health gates — the platform stays up while it evolves.

The Workload on Top

The leading sovereign workplace, running on top.

This infrastructure exists to run VNClagoon — the full-stack sovereign digital workplace trusted by government agencies, healthcare, and financial institutions. Every product deploys as cloud-native microservices on the same fabric.

VNCtalkVNCmailVNCprojectVNCtaskVNCcalendarVNCfilesVNCcontactsVNCdirectory…and the full VNClagoon suite

Cloud-native mail

The mail backend is built on Stalwart — Rust-native, microservices-ready, deployed on Kubernetes with PostgreSQL storage. Mail finally scales like the rest of the platform.

Video where physics lives

Conferencing media relay runs on dedicated bare metal for uncompromised call quality — orchestrated with, and secured by, the same platform.

One directory, one fabric

Identity, access, and policy span every product and every site. Users see one workplace, wherever the pods happen to run.

The AI Strategy

An AI factory you own — technically, legally, operationally.

On this infrastructure, VNClagoon becomes a sovereign AI factory: the Agentic Suite assists every user interaction, and the Agent Factory orchestrates specialist agents that plan, develop, test, and deploy software end to end — with sovereign inference, RAG memory, an MCP tool registry, and a human gate at every stage. No cloud dependency. No data sharing.

The Economics

Host wherever you like. The cost will excite you.

A complete sovereign stack for thousands of users runs on a handful of commodity servers — in your own data center, at any regional provider, in any private cloud. Strip away hyperscaler rent, managed-service markups, and egress fees, and the same workload lands at a factor of 5–10 below hyperscaler cost.

Sovereign stack

Commodity hardware · full stack embedded

baseline

Hyperscaler equivalent

5–10×

Managed Kubernetes + billed-per-service add-ons

plus egress · plus lock-in

No managed markups

PostgreSQL, vector search, object storage, load balancing, registry, monitoring — embedded in the stack, not billed per service, per gigabyte, per request.

No egress fees

Your data moves freely between your own sites over your own encrypted fabric. Mirroring, backups, and failover cost bandwidth — not a toll.

Predictable by design

Capacity is hardware you chose, at prices you negotiated. Scale by adding a node — not by climbing a pricing tier.

Run the numbers on your own footprint. Our team sizes a reference build for your data center, your provider of choice, or your national cloud — and prices it against your current cloud bill. The comparison is the sales pitch.

Ownership & Liability

One vendor. Total accountability.

VNClagoon is not a loose bundle of community projects. VNC AG is the single point of entry — and the final point of responsibility — for the entire stack: contractual certainty, zero fragmentation, and vendor-backed compliance assurance.

1st Level

Certified experts

Delivered by certified VNClagoon partners close to your operation.

2nd Level

VNC AG support team

Directly through VNC AG or certified partners — structured escalation for regulated environments.

3rd Level

Product engineering

The engineers who build the platform stand behind it — legally and operationally.

Take the Next Step

Build your sovereign AI factory.

Today. On your terms. On your infrastructure. Our team designs the reference architecture for your data center, your provider, or your national cloud.