Launch Your First Agent
Create an XO project from an agent template and reach a successful first interaction.
This guide is the common path through your first launch. The agent-specific quickstarts contain the exact fields, screenshots, and authentication steps for each template.
What You'll Need
- An XO account
- An agent selected using Choose an Agent
- The model credentials and any channel credentials listed in that agent's setup guide
Pick the Detailed Quickstart
OpenClaw
Create the project, connect a model and channel, open the gateway, and test chat.
Claude Code
Create the project, authenticate Claude, open code-server, and start a terminal session.
Hermes
Create the project, configure a provider and channel, and open the Hermes dashboard.
XO Cowork
Create the project, complete in-workspace onboarding, and start your first work project.
The Common Launch Path
Create the XO Project
Open beta.xo.builders, click New Project, and select the agent template you chose. Give the XO project a unique name and complete the template's creation form.
Wait for the Workspace
XO provisions a remote machine and starts the services required by that agent. Stay on the project page and use the readiness signal named in the detailed quickstart before continuing.
Do not start authentication or open an agent interface until the required services report that they are ready.
Finish Agent Setup
Complete any remaining model authentication, channel connection, or in-workspace onboarding in the detailed quickstart. Optional connections can be added later from the agent's setup controls.
Open the Agent
Use the Launch area or the interface opened by the setup flow:
- OpenClaw: open the OpenClaw Gateway and connect with the exact project name as the gateway token.
- Claude Code: open code-server, open a terminal, and run
claude. - Hermes: open the Hermes Dashboard and use the connected messaging app.
- XO Cowork: finish onboarding and open the project chat view.
Complete a First Test
Send a short message or start a simple coding session. A successful response confirms that the workspace, model authentication, and chosen interface are working together.