Virtual laboratory MVP
One operator on site. Remote engineers on our VDI.
At House of BioHealth, ARGUS runs a small physical room: 4–6 desks with monitors and only one chair — for the office operator. Contract engineers work remotely on our virtual desktops. Agents orchestrate project tasks. Heavy compute uses LuxProvide / MeluXina as batch — not as each person’s interactive desktop.
How it works
- One on-site operator watches the monitors: each screen = one remote engineer session
- New person = new monitor + isolated Linux VM
- Access via browser (Apache Guacamole) + SSO with MFA
- Agents create tasks, accept artifacts, revoke access; they do not sit in the remote desktop
- Human decides always — access, money, merge/deploy need a person
MVP stack
- Guacamole
- Proxmox / EU host
- Keycloak MFA
- Linux golden image
- Vault
- Agent | human task queue
Compute split · batch path
- Interactive VDI
Engineer desktops stay on the EU host / Proxmox. They do not run on MeluXina.
- Heavy training & simulations
MeluXina batch only (LuxProvide). Guidance from the Oct 2026 tour: ~€8 per GPU node-hour. Public / deck capacity: 200 GPU nodes, 800× NVIDIA A100 40GB.
- Integration path
Task queue → hpc-gw → Slurm sbatch and/or Flinky Models-as-a-Service; artifacts return to the queue. LuxProvide · docs.lxp.lu
No SSH allocation is claimed today. Batch access is requested through LuxProvide; the queue path above is the integration design.
Physical room
4–6 stations without chairs; 1 operator chair; live view of remote work; cost-efficient remote talent.