Configure Your Orchestrator

Set operational flags for video transcoding, AI inference, or dual mode - docker-compose templates, pricing, networking, and GPU assignment.

Choose the tab that matches the intended workload. All three modes use the same binary and differ only by flags. Keep Arbitrum connection and staking flags for the connect-and-activate step.

Build the docker-compose file that will carry the node through the rest of setup. Pick one operating mode: video transcoding, AI inference, or dual mode.

Common configuration

Networking

The -serviceAddr flag declares the public address gateways connect to. This address must be reachable from the internet:

# IP address
-serviceAddr 203.0.113.42:8935

# Domain name (preferred - survives IP changes without re-registration)
-serviceAddr orch.yourdomain.com:8935

Port 8935 must be open inbound on the firewall. Test reachability from a different machine before connecting on-chain:

curl -k https://YOUR_PUBLIC_IP:8935/status

Any response (including a JSON error) confirms the port is open. A connection timeout means the firewall is blocking it.

Port 7935 (-cliAddr) is for local CLI access and Prometheus metrics. Keep it bound to localhost unless external monitoring is configured.

GPU selection

nvidia-smi -L

This lists GPU device IDs:

GPU 0: NVIDIA GeForce RTX 4090 (UUID: GPU-...)
GPU 1: NVIDIA GeForce RTX 3090 (UUID: GPU-...)

Use the device index in -nvidia:

-nvidia 0        # Single GPU
-nvidia 0,1      # Two GPUs
-nvidia all      # All available GPUs

For mixed GPU configurations (different VRAM sizes), assign specific GPU IDs to specific pipelines in aiModels.json using the gpu field.

Pricing

-pricePerUnit sets video transcoding price in wei per pixel (not ETH). A typical starting range is 500-2,000 wei per pixel. Set it below 1,000 wei per pixel initially and adjust based on job volume.

For AI pricing, set price_per_unit per pipeline in aiModels.json. Values are in wei per output pixel (or per ms for audio, per token for LLM).

For competitive positioning guidance, see .

Persistent data

Always mount the data directory so the keystore and data survive container restarts:

volumes:
  - ~/.lpData:/root/.lpData

Without this mount, go-livepeer creates a new Ethereum account on every container start, losing the previous keystore and all bonded LPT.

For AI workloads, also mount the models directory:

volumes:
  - ~/.lpData:/root/.lpData
  - /var/run/docker.sock:/var/run/docker.sock

The aiModelsDir must be a host path. Docker mounts that path into AI runner containers.

Verify configuration before connecting on-chain

Start the node and confirm it initialises cleanly before proceeding to the Connect step:

docker compose up -d

Check for clean startup:

docker compose logs -f 2>&1 | grep -iE "gpu|transcode|ai-runner|error|FATAL" | head -20

Expected for video mode:

Transcoding on Nvidia GPU 0
Listening for RPC on :8935

Expected for AI mode (after model download):

Starting AI worker
Warm model loaded: ByteDance/SDXL-Lightning

Resolve FATAL or repeated error lines before connecting on-chain.

Next step

Connect to Arbitrum

Fund the wallet, stake LPT, register on-chain, and activate the orchestrator on the Livepeer protocol.