Observing several machines' engines from one place.
Each machine runs the daemon — it takes no Spinloop of its own, just its flags:
# on studio.local and gpu-box, with a token since they are network-reachable
SPINLOOP_API_TOKEN=… spinloop daemon
# on this machine, loopback-only needs no token
spinloop daemon --api-addr 127.0.0.1:4242What a node runs is decided by whoever starts it — fleet.yaml names the
Spinloop that describes it (see studio's entry: studio/Spinloop, found
automatically since the subdirectory's name matches the node's), so
fleet start knows what to push. Then from anywhere that can reach them:
cp .env.example .env # fill in each node's token
spinloop fleet status
spinloop fleet metrics -w
spinloop fleet start gpu-boxA node that is down shows as unreachable and the rest of the fleet still
renders. See docs/commands/fleet.md.
examples/fleet-docker/ brings up a real multi-node fleet
in containers — real daemons with a fake engine — so you can try spinloop fleet
on a laptop before setting any of this up for real.