Skip to main content
OpenClaw is a self-hosted personal AI agent gateway. It connects chat apps and tools to the model provider you choose, and it keeps its state in one volume. This guide runs the gateway as a single Levelrail app. To let an agent operate Levelrail itself, see control Levelrail from OpenClaw.

Deploy it

Then add a domain so the gateway is served over HTTPS. See domains and ingress.

Security and requirements

Level: Standard. The template has no privileged-container finding. What this template needs:
  • Publishes ports directly on the host (openclaw-gateway): 18789:18789.
  • Passwords and keys are generated at deploy time (openclaw-gateway): OPENCLAW_GATEWAY_TOKEN.
  • Stores data in named volumes (openclaw-gateway): openclaw_data.
Before you expose the gateway, review how the template starts it. The container command binds the gateway to the LAN interface (--bind lan) and passes --allow-unconfigured, so it starts without a completed setup. Use the generated gateway token for any client that connects to it.

What Levelrail adds

What a raw docker compose misses

  • 18789:18789 is published on the host. The gateway is reachable directly unless you filter it.
  • The gateway token is a $SERVICE_HEX_32_GATEWAYTOKEN placeholder. Plain Docker Compose cannot resolve it, so you would generate and store the token yourself.
  • Compose does not back up openclaw_data, so the agent state and workspace exist on one disk.
  • Compose has no rollback to an earlier revision.

Next steps

Last modified on October 20, 2026