OpenClaw setup

The simplest way to run OpenClaw on Cue OS needs no terminal: choose OpenClaw as the agent engine in the agent's settings, and Cue sets it up on the agent's Cue Computer. See OpenClaw.

Use this page to set OpenClaw up by hand, or to check and repair it. Run the commands in the computer's terminal. On a Cue Computer, open it from the web with Files, then More options, then Terminal.

Cue CLI and OpenClaw keep separate settings. Cue uses ~/.cue/. OpenClaw uses ~/.openclaw/.

1. Check that OpenClaw is installed

OpenClaw is already installed on Cue computers.

command -v openclaw
openclaw --version

The first command prints a path, and the second prints the OpenClaw version.

If it is missing on your own computer, install it with npm. This was not tested for this page:

npm install -g openclaw@latest

2. Run first-time setup

On a Cue Computer where you turned on OpenClaw from the web, Cue has already done this step, and ~/.openclaw/openclaw.json exists. Skip to step 3.

Otherwise, run setup without prompts:

openclaw onboard \
  --non-interactive \
  --accept-risk \
  --mode local \
  --gateway-port 18789 \
  --gateway-bind loopback \
  --skip-skills \
  --skip-channels \
  --skip-daemon \
  --skip-health \
  --json

What the flags do:

--accept-risk is required with --non-interactive.
--skip-daemon skips installing a background service, which a computer without a service manager cannot run.
--skip-health skips a health check that would fail before the gateway is started.
Credentials are set in a separate step, so you can choose them there.

When it finishes, ~/.openclaw/openclaw.json exists.

If setup stops early, do not continue to the chat-platform pages. Run it again until it finishes.

3. Set up model access

Continue to Model auth to choose credentials and a default model, and to check that OpenClaw answers.

1. Check that OpenClaw is installed

2. Run first-time setup

3. Set up model access

Next

On this page