overview
The Agent Builder walks you through 9 sequential steps to create a complete, deployable AI agent configuration.
How It Works
Each step focuses on one aspect of your agent's design. Steps 1 and 2 are required — everything else has sensible defaults and can be skipped or customized later.
The builder generates two outputs:
System Prompt — A comprehensive instruction set compiled from your goal, persona, policies, tools, and memory settings. This is what the LLM actually reads at runtime.
OpenClaw JSON — A structured configuration file containing the system prompt, tool definitions, connector configs, trigger rules, memory settings, and MCP pipeline graph. This is what you deploy.
Navigation
Progress through steps sequentially using the Next Step button
Jump back to any completed step by clicking its icon in the progress bar
You cannot skip ahead to uncompleted steps
The current step glows in the progress bar; completed steps show a checkmark
Saving vs. Exporting
When you click Pay & Export on Step 9, two things happen:
The agent is saved to your ClawEngine account (visible on the Agents page, editable anytime)
The OpenClaw JSON file is downloaded to your computer for deployment
The on-chain payment of 100 $CWE is required for the export/download. Saving alone (via the API) costs 100 $CWE from your in-platform balance.
Starting Over
The wizard resets each time you open the Create Agent page. If you leave mid-wizard, progress is not saved. Once exported, agents are stored permanently in your account.
What's Next
Start with Step 1 — Goal & Identity, or browse Templates to fork a pre-built agent.
Was this helpful?