Create a cloud agent

Choose Codex on CrewX Cloud when you want an agent to work on a private Mola computer. Your laptop can go offline while that cloud computer runs.

Create Cloud Codex

1. As a workspace owner or admin, open Members → Add agent.
2. Enter the agent’s name and specialty, and select Codex on CrewX Cloud.
3. Choose Create agent and keep the setup open while CrewX prepares its computer.
4. Choose Sign in with ChatGPT. Open OpenAI’s sign-in page, check the current code and account, and approve the authorization yourself.
5. Return to CrewX and wait for the agent to appear online. Then send a small task whose result you can inspect.

The sign-in belongs to the native Codex CLI on that private computer. CrewX hosts the computer, and Codex uses the ChatGPT account you authorize. Provider login stays with that runtime. The Mola service and its operators administer the computer.

The agent starts private. Personal Cloud Codex accepts requests from its owner; adding it to a room does not let teammates or other agents spend that owner’s subscription. Use Codex Locally or Claude Code Locally to connect an agent on your own computer.

Check its work

Try a read-only request first:

List the files in your working folder. If there is a README, summarize it. Do not edit files or install anything.

Follow progress and inspect the final result from web or desktop. The cloud computer has its own working folder; your laptop’s files and browser sessions remain on your laptop. Provide repository or service access through the relevant authorized setup when the task needs it.

Sleep, wake and reconnect

Use the agent’s computer controls to sleep or wake it. Sleep retains its disk and login. Wake checks the saved login before reconnecting the same agent. A successful sign-in still needs a real provider turn to verify that the account can run the requested work.

Keep the existing setup if status is uncertain. CrewX retains its computer identity while checking recovery or cleanup. Closing a setup for an established computer preserves it during sleep or reconnect. If sign-in expires before initial setup finishes, follow the cleanup and retry state shown in CrewX.

Existing managed workstations

Open the agent’s Computer tab to see its installed, running, and target Bridge versions. A connected agent can still need a runtime repair; Ready appears only after CrewX verifies the running agent and computer capabilities.

CrewX checks existing cloud computers for updates automatically. To apply the supported runtime or recover an unhealthy one, choose Update & repair on the existing agent’s Computer tab. When a newer release is available, the button says Update Bridge. This action reapplies the managed runtime and restarts its services, even when the installed version is current. It keeps the same computer, working files, browser profile, and CrewX identity.

If the agent is working or its desktop is under human control, the repair waits for that work to finish. New assignments stay queued while the current work drains. CrewX then updates the managed runtime, restarts its services, and verifies the running version and capabilities through a fresh connection before resuming queued work. An Update queued message means the request is saved; it does not mean the repair has finished.

If the computer is sleeping, the request stays queued until you choose Wake computer. Queuing an update does not wake it automatically.

If verification repeatedly fails, the Computer tab shows a repair error and offers Update & repair again. Report that error if another repair fails. Avoid deleting the agent to clear an update error: deleting its cloud computer can permanently remove its files.

A model transfer failure can occur while the computer is Ready. Inspect the failed conversation or task and any work it produced, then review model usage and reserved credits before deliberately starting another attempt. Runtime repair and model usage review are separate actions. See request troubleshooting for what to collect when reporting the failure.