Core Concepts
The account, project, workspace, machine, and agent terms used throughout XO.
XO puts an AI agent inside an isolated remote environment. A few terms describe the layers between your account and the work the agent does.
The Main Building Blocks
| Term | What it means |
|---|---|
| XO account | Your identity on XO. It owns the projects you create and controls which projects you can access. |
| XO project | The item you create from an agent template and select from the XO dashboard. It ties an agent to its remote workspace. |
| Workspace machine | The isolated remote machine provisioned for the project. It runs the agent and its supporting services and can be started, stopped, or restarted. |
| Agent | The software doing the work, such as OpenClaw, Claude Code, Hermes, or XO Cowork. Each agent offers a different interface and workflow. |
| Service | A process running inside the workspace, such as the Cowork API or OpenClaw Gateway. A service must be ready before its interface can open. |
| Model provider | The AI model account that powers the agent. Depending on the agent, you connect it with an API key or OAuth. |
| Channel | A messaging app through which people can reach an agent, such as Telegram or Slack. Channel support varies by agent. |
| Data source | An external account, such as Google Drive or GitHub, that you connect so an agent can read relevant information. |
Two Meanings of “Project”
The word project appears at two different levels:
- An XO project is the dashboard item that owns the remote workspace and agent.
- A work project or project directory is a folder inside that workspace where the agent keeps files and context for a particular piece of work.
For example, you might create an XO project named xo-cowork, then create a work project named marketing inside it. Creating the inner project does not provision another machine.
Where You Control Things
| Area | Use it for |
|---|---|
| Setup | Connect models, data sources, channels, and other credentials exposed by the selected agent. |
| Launch | Open the agent interface, browser IDE, or service controls available for the project. |
| Workspace IDE | Inspect files and run commands directly inside the remote machine. |
The exact controls vary by agent, so start by choosing the agent that matches your workflow.