Lõhmus Tech
Menu

Infrastructure that survives isolation.

Kubernetes, sovereign environments, and AI that runs inside your perimeter.

Combed network cabling, one line carrying signal blue.

Most Kubernetes consultancies assume a working internet connection. A registry to pull from, a chart to fetch, a control plane somebody else operates. I build the clusters where none of that exists, and run the virtualization and the hardened Linux underneath them.

The same problem is now arriving for AI. Regulated organisations across Europe want agentic systems and cannot send their data to an inference API in another jurisdiction. Solving that means running models on your own hardware, behind your own gateway, inside your own perimeter.

It is the same discipline, applied twice.

See the capabilities

The founder

Hi, I'm Yaro.

I founded Lõhmus Tech, and I'm the person you actually work with. No account managers, no handoffs, no "let me check with the team." You get the engineer who does the work.

What I do

Sovereign infrastructure

Clusters and platforms that keep running when the outside world is not available.

  • Air-gapped and sovereign environments

    Offline registry mirroring, artifact provenance, and a patch and CVE process that works with no egress at all.

  • Kubernetes, cloud to bare metal

    On-premise, managed, and hybrid clusters. GitOps delivery. Multi-cluster topologies that fail independently.

  • Virtualization platforms

    Proxmox and VMware as the substrate under the cluster, designed and operated as a first-class layer rather than an afterthought.

  • Hardened Linux

    Minimal images, controlled supply chain, a baseline you can defend in an audit.

  • Cross-cloud and hybrid

    Depth in GCP and Azure, and the connective work that joins them to what you already run.

  • Resilience and disaster recovery

    Redundancy design, recovery planning, and systems that heal without waking anyone up.

Sovereign AI

Agentic systems that run on your hardware, under your policy, inside your network.

  • Self-hosted inference

    Running open models on your own GPUs. Scheduling, serving and lifecycle on Kubernetes, including fully disconnected.

  • AI gateways

    One controlled ingress for model traffic. Routing, quotas, cost attribution, audit logging, key management.

  • Guardrails

    Input and output policy at the model boundary. Prompt-injection defence and data-loss prevention that a security team can actually reason about.

  • Agent fleets and harnesses

    Orchestrating and operating fleets of agents, and the harness around them: tool boundaries, permissions, isolation, observability.

  • Retrieval over private corpora

    RAG pipelines where the documents, the embeddings and the index never leave the perimeter.

  • Agentic development

    Putting agentic workflows into a real delivery process, with review gates that hold.

Beneath both, the software itself: full systems from design to delivery, legacy modernised, built with agentic coding under close human steering. More in the capabilities.

Eight years of building this.

As an employee, as a consultant, and on ventures I founded or co-founded. Almost always alongside other people.

  • Atmos Group FinlandAir-gapped medical system
  • BookkeriCo-founder, high-availability SaaS
  • St1Cloud architect and team lead
  • Suomen RahapajaBackend and coin-recognition ML
Read the full track record

Tell me what cannot leave your network.

Start a conversation