Skip to main content
This guide is for operators and scripts. For a first Agent, use the dashboard quickstart instead.

​
Before you begin

  • Install a released Clawdi CLI and check the version-specific command help.
  • Use browser authorization. A legacy manually entered API key does not authorize Cloud Agent deployment.
  • Decide whether this is one logical deployment attempt or an intentionally different deployment before assigning its request ID.

​
Run the interactive CLI wizard

Sign in with the browser authorization flow, confirm the credential source, then start the wizard:
In a headless environment, use the non-opening browser flow shown by clawdi auth login --help. Do not use clawdi auth login --manual for Cloud Agent deployment. The CLI loads the Agent software, AI access, model, compute, and payment choices currently available to the signed-in account. Before automating, read the help shipped with that CLI version:
Supply the choices required by the current help output. Do not copy a fixed set of optional flags from an older guide, and never put provider secrets on the command line.

​
Make non-interactive retries idempotent

Every non-interactive deployment requires a caller-supplied stable UUID in --request-id before the CLI sends a create or checkout mutation. Treat that UUID as the identity of one logical deployment attempt.
  • Reuse the same request ID, selections, and payment intent when retrying an attempt whose outcome is unclear.
  • Use a new request ID when you intentionally change the Agent software, AI access, compute, payment, or another deployment choice.
  • If the CLI reports a timeout or network error, the request may already have been accepted. Check Agents in the dashboard before retrying the same logical attempt with the same request ID.
  • If checkout is pending, complete or recover that checkout instead of starting a second payment.
Use --json for automation. Its top-level result status distinguishes: These result values are not the Agent’s lifecycle status. After acceptance, use the dashboard as the status authority and follow Cloud Agent states and recovery.

​
Recover an unclear outcome

  1. Keep the original request ID and deployment choices.
  2. Open Agents and check whether the deployment already exists.
  3. Recover any pending checkout rather than creating another payment intent.
  4. Re-authenticate if needed, then retry only the same logical attempt.
  5. After acceptance, wait for the Agent’s lifecycle status instead of using a new request ID to work around a transition.
For customer-visible failures after acceptance, use Cloud Agent troubleshooting.
Last modified on August 2, 2026