Your machine

On the machine:

pip install surfra
surfra login
surfra runner register   # name it desk-4090
surfra runner start

Then deploy with the runner as the first row:

deploy.py

import surfra

app = surfra.App("desk-hello")
app.deploy(
    image="ghcr.io/you/hello:1",
    entrypoint="app.py",
    runtime={"engine": "python"},
    autoscale={
        "max_replicas": 3,
        "per_provider": True,
        "providers": [
            {"provider": "runner", "runner": "desk-4090", "priority": 1, "min_replicas": 1, "max_replicas": 1},
            {"provider": "runpod", "gpu": "rtx4090", "priority": 2, "max_replicas": 2},
        ],
    },
)

Your machine stays on as the base. When requests queue up, up to two RunPod replicas join and stop again after idle_timeout_sec. If your machine is off, the RunPod row takes over. See Your machines.