Can my Mac sleep while serving?
No — inference needs the machine awake. Either disable sleep in System Settings, or pass --prevent-sleep to provider.py. The flag wraps a caffeinate subprocess on macOS.
I haven't pulled the requested model. What happens?
The orchestrator only routes you tasks for models you advertised at registration time. Your node won't be assigned tasks for models it can't serve. If demand for a model exceeds supply, the orchestrator will fall back to a model in the same tier — your node won't notice anything different.
Can I run native and browser nodes on the same wallet?
Yes. They register as separate node identities under the same wallet. Both accrue points independently, both sum into one balance.
How fast must my internet be?
Modest. Inference is local; only the prompt and the streamed response go over the wire. 5 Mbps up/down is plenty. Latency matters more — lower latency means the orchestrator picks you more often.
What happens if my node disconnects mid-task?
The task fails after the orchestrator's timeout (120 s) and is re-queued. The original caller gets a 502 only after our retry window is exhausted. Your node doesn't accrue reward for that attempt.
Do I need to expose ports / port-forward?
No. The provider initiates all connections (long-polling outbound HTTP). No inbound port is opened on your machine. NAT / firewalls / consumer routers all just work.
Is my node visible to other users?
The dashboard's "Live nodes" table shows your label (you choose it), hardware tier, and tokens/sec. It does NOT show your IP, wallet address (only first/last 4 chars), or location.
How do I stop providing?
Ctrl+C the provider.py process, or close the browser tab if you're running a browser node. The orchestrator marks you offline after 60 s of no heartbeat.
Where does my wallet sign?
Once at registration, with a one-time message Sign in to GPU: <nonce>. No on-chain transactions, no gas. The signature gives the orchestrator a session cookie tying your wallet to your node id.