Guides

Compute

Compute instances are GPU/CPU machines provisioned in your cloud for notebooks and jobs. The agent runs your workloads in containers and streams metrics and logs back to the console.

Create an instance

polnor compute create training-a100 \
  --provider prod --flavor gpu.a100 --auto-stop 2h

Or via API:

POST /api/v1/compute
{ "name": "training-a100", "flavor": "gpu.a100", "auto_stop": "2h" }

Start & stop

polnor compute start training-a100
polnor compute stop  training-a100

The image used by an instance is persisted on first start, so restarts are reproducible. Failed starts record the error against the instance instead of silently disappearing.

Auto-stop

Set an idle timeout (30m, 2h, or never) so forgotten instances don't accrue cost. The reaper stops anything past its window.

Live metrics

The agent collects CPU, memory, GPU and disk utilisation and streams them to the console, the same numbers you see on the dashboard. Metrics are authenticated with the agent's bearer token end to end.

Quotas

A workspace can cap max_active_computes. Hitting the cap returns 429 with the exact ceiling, e.g. max_active_computes=10, current=10.