Start here for macOS and Linux. They share the same native install path. Windows still uses a separate WSL2 flow.

​
Install Cue CLI

Use the standard installer on the machine where your agent will run: 
curl -fsSL https://cueos.ai/install.sh | bash
To opt into the dev channel: 
curl -fsSL https://cueos.ai/install.sh | bash -s -- --channel dev

​
Validate

After install: 
cue --version
cue -m doctor --json

​
Connect Your Account

Run the setup flow: 
cue setup
# or
cue onboard
Then review Authentication if you need multiple Cue accounts, direct provider keys, or environment-based auth.

​
Non-Interactive Onboarding

Use cue onboard --non-interactive for automation, sandbox bootstrap, or any headless setup flow. Example with a Cue API key plus a direct provider key: 
cue onboard \
  --non-interactive \
  --cue-api-key "$CUE_API_KEY" \
  --anthropic-key "$ANTHROPIC_API_KEY" \
  --provider anthropic \
  --model claude-sonnet-4-5 \
  --json
Example with Cue access and refresh tokens: 
cue onboard \
  --non-interactive \
  --cue-access-token "$CUE_ACCESS_TOKEN" \
  --cue-refresh-token "$CUE_REFRESH_TOKEN" \
  --openai-key "$OPENAI_API_KEY" \
  --provider openai \
  --model gpt-5 \
  --json
Notes:
  • cue setup is the interactive path.
  • cue onboard --non-interactive is the headless path.
  • Cue onboarding is workspace/account/provider oriented. It does not expose OpenClaw-style gateway or channel setup flags.

​
macOS and Linux Notes

  • Open a new terminal after install so your shell picks up the updated PATH.
  • If node or cue is not found, verify your shell init files loaded correctly and rerun the validation commands.
  • On Linux, prefer a user-local Node setup if your machine has environment-specific restrictions.

​
Windows

If cue --version or cue -m doctor --json fails on macOS or Linux, check the notes above before opening an issue.