Deploying OpenClaw (ClawdBot) on GPU clouds with dstack

Prerequisites

1. Install dstack
2. Set up a dstack gateway (already set up for you if you sign up with dstack Sky)

Deploy a model (Qwen 32B)

type: service
name: qwen32

image: lmsysorg/sglang:latest
env:
  - HF_TOKEN
  - MODEL_ID=deepseek-ai/DeepSeek-R1-Distill-Qwen-32B
commands:
  # Launch SGLang server
  - |
    python3 -m sglang.launch_server \
      --model-path $MODEL_ID \
      --host 0.0.0.0 \
      --port 8000

port: 8000
model: deepseek-ai/DeepSeek-R1-Distill-Qwen-32B

# Uncomment to use allow spot instances
#spot_policy: auto

resources:
  gpu: H100
  disk: 200GB

export HF_TOKEN=...
dstack apply -f qwen32.dstack.yml

Once the model is deployed, the OpenAI-compatible base URL will be avaiable at https://qwen32.<your gateway domain>/v1.

Deploy OpenClaw

type: service
name: openclaw

env:
  - MODEL_BASE_URL
  - DSTACK_TOKEN

commands:
  # System dependencies
  - sudo apt update
  - sudo apt install -y ca-certificates curl gnupg

  # Node.js & openclaw installation
  - curl -fsSL https://deb.nodesource.com/setup_24.x | sudo -E bash -
  - sudo apt install -y nodejs
  - curl -fsSL https://openclaw.bot/install.sh | bash -s -- --no-onboard

  # Model configuration
  - |
    openclaw config set models.providers.custom '{
      "baseUrl": "'"$MODEL_BASE_URL"'",
      "apiKey": "'"$DSTACK_TOKEN"'",
      "api": "openai-completions",
      "models": [
        {
          "id": "deepseek-ai/DeepSeek-R1-Distill-Qwen-32B",
          "name": "DeepSeek-R1-Distill-Qwen-32B",
          "reasoning": true,
          "input": ["text"],
          "cost": {"input": 0, "output": 0, "cacheRead": 0, "cacheWrite": 0},
          "contextWindow": 128000,
          "maxTokens": 512
        }
      ]
    }' --json

  # Set active model & gateway settings
  - openclaw models set custom/deepseek-ai/DeepSeek-R1-Distill-Qwen-32B
  - openclaw config set gateway.mode local
  - openclaw config set gateway.auth.mode token
  - openclaw config set gateway.auth.token "$DSTACK_TOKEN"
  - openclaw config set gateway.controlUi.allowInsecureAuth true
  - openclaw config set gateway.trustedProxies '["127.0.0.1"]'

  # Start service
  - openclaw gateway

port: 18789
auth: false

resources:
  cpu: 2

export MODEL_BASE_URL=<your model endpoint>
DSTACK_TOKEN=<your dstack token>
dstack apply -f openclaw.dstack.yml

The Moltbot UI will be available at https://openclaw.<your gateway domain>/chat.