Deploy Etai
Start a control plane
Install Docker with Compose. Extract the Etai deployment bundle and run its startup script.
./start.sh
Open the printed address and sign in with the generated admin token. If a GPU host runs on another machine, give it a private address it can use to reach the control plane.
Enroll a GPU worker
Install Etai with its worker extra and install the Orme CLI on your GPU host. Create an enrollment token in the Workers page. The worker pulls the matching CUDA or ROCm runtime from the release.
etai worker doctor
etai worker join --server https://pool.example.net --name gpu-01
etai worker run
Run a saved workflow
Use your own control plane address. Import an API-format ComfyUI graph and a spec containing its runtime profile and model paths.
etai login https://pool.example.net
etai workflows add demo graph.json --spec spec.toml
etai run --workflow demo --wait --output ./out