GPU Provider Guide

Who this is for

GPU providers that want to join future ZORAI AI compute pilots. The production provider network is not yet presented as an SLA-backed market.

No public IP required

The recommended mode is to run compute-agent on the internal GPU host. The Agent initiates an outbound WebSocket connection to the public compute-relay. Providers do not need to open public ports, and should not expose Docker API, SSH or Compute Node APIs to the internet.

Provider launch

./compute-agent \
  --relay wss://compute-relay.zorai.co/v1/provider/connect \
	  --relay-id hz-1 \
  --provider-key "$PROVIDER_PRIVATE_KEY" \
  --compute-url http://127.0.0.1:8645 \
  --gpu-name "NVIDIA A100" \
  --gpu-vram-mb 40960

What the Agent does

• Signs the registration message with the provider key, binding provider address, relay_id, session_id, timestamp and GPU metadata.
• Sends heartbeats to maintain online/offline state.
• Calls the local Compute Node /compute, /status and /result endpoints after task assignment.
• Builds a ComputeProof, signs it with the provider key and returns it through the Relay.

Provider checklist

• During the current two-Relay test deployment, use sticky routing or pin to one backend so task submission and provider session land on the same Relay.
• Prepare GPU inventory, region, bandwidth, runtime stack, pricing and uptime monitoring.
• Run the local Compute Node and restrict it to 127.0.0.1 or trusted private networks.
• Keep proof, task logs, result hashes and payout wallet records for settlement and disputes.
• Expect anti-cheat, dispute handling, real slashing, benchmark validation and public e2e tests before mainnet launch.

Related code

cmd/compute-agent, cmd/compute-relay, pkg/computerelay, pkg/computenode, pkg/gpu, pkg/taskmarket, pkg/rpc/compute_rpc.go.