Prerequisites
Before you install anything, get the basics in place: a machine with Node.js and npm available in your terminal, a Discord account you control, and API credentials for the model provider you want OpenClaw to use. You do not need a perfect production stack on day one, but you do need a clean environment where the CLI can create files, store config, and stay running.
What you need
Node.js, npm, terminal access, Discord, and at least one model API key.
What to expect
A guided CLI setup, a generated workspace, and a first test conversation inside Discord.
Installation
OpenClaw starts with a global install so the CLI is available anywhere on your system. The basic flow is simple:
npm install -g openclawAfter that, run the setup wizard. The wizard handles the early questions that usually trip people up: project name, model provider, auth details, and where to create the workspace. The teaser version stops here because the exact prompt sequence and recommended answers are part of the paid guide.
The important thing is understanding the outcome: when the wizard finishes, you should have a structured OpenClaw project with the core files in place and enough configuration to begin connecting channels and defining the agent persona.
Want the full chapter?
The full guide includes complete config templates, annotated examples, and troubleshooting for every step.
Skip the guesswork and follow the exact setup that gets a new OpenClaw agent from blank machine to first useful conversation.
Get the KaiShips Guide to OpenClaw — $29Discord Setup Overview
Once the workspace exists, the next job is giving your agent a place to talk. In practice that means creating or selecting a Discord application, enabling the right bot permissions, inviting it into your server, and connecting those credentials back into OpenClaw. This part matters because most first-run failures are channel or token mistakes, not model problems.
Keep the first Discord environment small. Use a test server, one or two channels, and a narrow scope. That makes it easier to see what the agent is doing and debug it before you expand to a public community setup.
Meet the Workspace Files
Three files shape the early behavior of a serious OpenClaw agent:SOUL.md,AGENTS.md, andUSER.md. Think of them as personality, operating model, and user context.
SOUL.md
The voice, values, and behavioral constraints of the main agent.
AGENTS.md
How sub-agents are defined, when they exist, and what jobs they handle.
USER.md
Durable context about the human operator, preferences, and working relationship.
The full guide includes complete starter versions of each file, but the teaser takeaway is enough: these documents are not filler. They are part of the control surface that makes an agent feel consistent instead of random.
Your First Conversation
The first successful run is not about intelligence. It is about proving the whole loop works: the agent boots, reads its files, connects to Discord, receives a message, and replies in the tone you intended. Once that loop is stable, iteration gets much faster. You can tune prompts, memory, and tools later. First make the system reliably speak.
Final CTA
Get the complete setup chapter
If you want the full walkthrough with the recommended wizard choices, config templates, and step-by-step troubleshooting, get the guide for $29.
Get the KaiShips Guide to OpenClaw — $29