The OpenClaw Setup Guide

To You, the Human — Read This First

This guide was written to be handed to an AI agent. The fastest way to use it: get the Claude Code desktop app, turn on Opus, give it this file, and let it set everything up while you watch and ask questions. You do not need to understand every technical detail. You need to be able to approve decisions and answer questions about your setup.

The Fastest Path

  1. 1Subscribe to Claude Pro (includes Opus 4 + Claude Code)
  2. 2Download the guide file →
  3. 3Open Claude Code, paste the handoff prompt, and attach this file
  4. 4Watch it work. Ask questions. Approve decisions.

Guide-Datei herunterladen

E-Mail eingeben, um herunterzuladen openclaw-guide.md. Übergib diese Datei an Claude Code — die Einrichtung erfolgt automatisch.

Kein Spam, niemals. Deine E-Mail wird sicher gespeichert, damit wir dir Updates zu neuen Anwendungsfällen und Workflows senden können.

Step-by-Step: How to Use the Guide

1

Subscribe to Claude Pro

Go to claude.ai/upgrade and subscribe to the Pro plan. This gives you access to Opus 4 for complex reasoning, Haiku for fast lightweight tasks, and Claude Code for the setup process itself.

2

Open Claude Code

Open Claude Code and paste the handoff prompt below along with the full guide file. Claude Code will handle the terminal steps. You handle anything that requires a browser.

3

Paste the Handoff Prompt

I want to set up OpenClaw, an open-source AI agent platform. I have attached a setup guide that was written specifically to be used by an AI agent to configure everything correctly. Please read the full guide first. Then walk me through the installation step by step, checking in with me at each major decision point before proceeding. Over-explain what you are doing and why in plain language. Assume I have never used a terminal before. If something in the guide is unclear, ask me before assuming anything. Do not skip steps. If a step requires me to do something in a web browser, such as creating a Discord bot, pause and describe what I should click and where. Once I confirm I have completed that step, continue. Start by asking me 5 questions about my current setup and goals before we begin.

Reading the guide below? The fastest path is to download the file and give it to Claude Code.

Download ↑

What You Are Building

A team of AI agents running 24 hours a day on a small computer in your home or office. Each agent has a name, a personality, and a specific job. They talk to you through Discord. They remember what they have learned, update their own knowledge over time, and handle tasks while you sleep.

This is not a chatbot that you have to prompt every time. It is a team that runs continuously.

“Pulled last 24hrs of emails and calls. Ran the follow-up list, 3 prospects touched. Pushed enterprise pricing to Samantha, SMB deck to Fred. Brochure out to James, Sandeep, and Karl. Pipeline updated across the board. Everything's moving.”

Ray Kroc — AI Sales Agent, running overnight on a Mac Mini

The Hardware

Recommended

Mac Mini

M-series chip. $600-800. Draws 10 watts. Size of a hardback book.

Windows

OpenClaw runs on Windows. Terminal commands need adaptation.

Virtual Machine

Ubuntu or Debian. 4GB RAM, 2 CPU cores. Great for experimenting.

Do not set this up on your main laptop. Start on a dedicated machine until you understand the system and trust your configuration.

What You Need

A Mac Mini (or Windows machine or VM)
A Claude account on the Pro plan at claude.ai
A Discord account and a server you control (free)
At least one AI API key (Anthropic recommended)
An internet connection
Tailscale (free) — secure remote access(Optional)
A GitHub account (free) — for automated nightly backups(Optional)

How Agents Work

An agent is defined by a set of text files in its workspace folder. These files are loaded automatically at the start of every session.

SOUL.mdWho the agent is. Personality, voice, values.
AGENTS.mdHow the agent operates. Startup sequence, delegation.
AUTONOMY.mdWhen to act, when to ask, which model to use.
USER.mdWho you are. Name, timezone, preferences.
TOOLS.mdTechnical setup. Channel IDs, file paths.
IDENTITY.mdThe agent's name, model, and Discord identity.
MEMORY.mdLong-term curated memory. Updated over time.

The Team

AgentRoleWhat they do
Don DraperCreative DirectorContent creation, social publishing, engagement, analytics
Ray KrocBusiness DevelopmentSales pipeline, proposals, CRM updates, follow-ups
GilfoyleSystems EngineerSecurity, infrastructure, monitoring, skeptical review
DineshGeneralistFast tasks, lightweight work, quick lookups
RichardChief of StaffOrchestration, research, routing work between agents
ClawdxpertPlatform ExpertOpenClaw configuration, agent setup, maintenance
Ted MosbyDesign EngineerUI/UX, frontend, visual craft
Jian-YangReasoning SpecialistLogic, structured problem solving

Use Cases

Each use case below is a real workflow running in production. For full details on each, visit the Use Cases page.

Build Your Audience

  • X, Threads & LinkedIn Content Publishing Engine
  • Automated Social Feed Engagement
  • Social Post Analytics That Feed Back Into Content

Grow Your Business

  • Automated Sales Pipeline & Proposal Generation
  • Inbox & Scheduling That Handles Itself
  • Social Watchlist Monitoring & Alerts

Build Knowledge

  • Searchable Second Brain (4,000+ Notes)

Build Software Faster

  • Overnight Coding Team (Architect & Coders)
  • Cross-Functional Agent Collaboration
  • Design & Frontend Engineering

Keep It Running

  • Daily Security Auditing
  • Self-Improving Agents
  • Automated Nightly Backup to GitHub
  • 22 Automated Cron Jobs
  • Self-Healing Error Resolution
  • Platform Updates

What Makes This Different

This is not a demo or a weekend project. It is a production system that runs every day.

Agents improve themselves

Every week, agents review their own performance and propose changes to their prompts and workflows. You approve what ships.

Complete config, ready to copy

Every workspace file, cron job, and prompt is included. You are not starting from scratch — you are customizing a working system.

22 automated jobs

Content publishing, security audits, backups, analytics, error recovery, and more run on schedule without you touching anything.

Agents talk to each other

When one agent discovers something relevant to another, they route it directly. You do not have to relay context between them.

Security

Your agents have real tool access. They can browse the web, run commands, read files, send emails, and post to social media. This is why the security setup matters.

Gateway binding — loopback only
API keys in wrapper script, not config
File permissions — chmod 600
Discord access control per bot
Elevated access off by default
Tool restrictions per agent
Log redaction for sensitive patterns
Automated daily security audit

Ready to Start?

Download the guide, hand it to Claude Code, and your agent team will be running in under an hour.