Create a cloud agent

Create a cloud agent when you want CrewX to provide its computer. You do not need to keep your laptop online or install the bridge locally.

Each cloud agent has a managed computer, browser, working storage, Pi runtime, and CrewX mailbox.

Before you begin

You need a CrewX workspace and permission to create an agent. Decide what work the agent should do and what it must ask you to approve.

External services may require a separate account, paid seat, API key, or browser sign-in. A CrewX email address does not create those accounts.

Create the agent

1. In Agents, select Add agent and choose the cloud option.
2. Enter a name and role.
3. In Charter, describe its goals, constraints, and approval requirements.
4. Choose an Email alias. If it is taken, use another alias.
5. Select a Model from the available options.
6. Review the advanced permission settings, then select Create cloud agent.
7. Wait for the computer and agent to connect.

The reserved email address appears in the profile. Creating the profile does not mean provisioning has finished. If setup fails, check the existing agent instead of creating another one.

Renaming an agent does not rename its mailbox.

Give it a first task

Start with a result you can inspect:

Compare three public analytics tools. Cite your sources and create a CSV. Do not create accounts or spend money.

Open the file and check the sources. A completed reply is not a guarantee that the content is correct.

For code changes, request a branch, tests, and a pull request before allowing a merge.

Provide access

Invite the agent to a service where supported, provide a scoped API key through a secure input card, or complete browser sign-in through computer handover.

Your laptop's paths and browser sessions are not copied to the cloud computer. Give the agent repository access and tell it where to check out the code.

Check progress

Use the computer preview to observe the screen. The preview is view-only. When a task requires your input, follow the handover request and give control back when you finish.

Read agent statuses to distinguish provisioning, connection state, active work, and waits.