Installation and Setup

This lesson is adapted from Claire Vo's guide on Lenny's Newsletter.

Getting OpenClaw running involves choosing your hardware, running the installer, and walking through the onboarding wizard. Claire recommends a Mac Mini M4 (~$600) for its simplicity and power, but there are several paths.

Hardware Options

Choose your deployment method

Three pathways exist:

Hosted services — the easiest option if you don't want to manage hardware:

StartClaw, MyClaw, SimpleClaw, UniClaw, Every's Plus One

Virtual private servers — a good middle ground:

Railway, Hostinger, DigitalOcean, Google Cloud, Render

Personal devices — full control, Claire's recommendation:

Mac Mini M4 (the "meme" option, ~$600)
MacBook Air or comparable hardware
Any machine with 8GB+ RAM

Claire's pick: Mac Mini M4. Compact, powerful, and simple. It sits on a shelf and runs 24/7 without fuss.

Pre-installation checklist

Before installing, prepare:

Create a dedicated admin account on your deployment machine — do NOT use your personal account
Set up a Gmail address for your agent(s) to use
Install Chrome browser (some skills require it for web automation)

Critical: Do not install OpenClaw on a work or personal computer that's actively in use. Agents have full system access and can edit files, delete content, install software, and send communications. Use an isolated, dedicated machine.

Run the installer

The automated install script handles all dependencies:

curl -fsSL https://openclaw.ai/install.sh | bash

This downloads OpenClaw, installs required dependencies, and kicks off the guided onboarding process.

Walk through onboarding

The onboarding wizard will guide you through several key decisions:

Model selection: Claire recommends Claude Opus 4.6 or Codex 5.4 (or the latest equivalents). Premium models produce significantly better results for autonomous work.

Authentication:

Use an API key from Anthropic or OpenAI developer accounts (recommended)
Alternatively, connect existing Claude or ChatGPT subscriptions — though Claire notes account-sharing risks with this approach

Channel setup: Start with Telegram — it's the most beginner-friendly channel. You can add WhatsApp, Slack, and others later.

Web search integration: Optional but valuable. Gives your agent access to real-time information. Brave API is the default; alternatives include Exa, Perplexity, or Firecrawl.

Skills installation: Install these two skills to start:

"gog" — Gmail, Google Calendar, and Google Docs access
"summarize" — Content summarization capabilities

Hooks activation: Enable session memory — this is essential for agents to maintain context across conversations. Debugging and optimization hooks are optional but useful as you learn.

Verify Your Setup

After onboarding completes, your agent is "hatched" and ready. Send a message via your connected channel (e.g., Telegram) to verify it responds.

Your agent's workspace files are stored in the .openclaw/ directory on your machine. These files define everything about your agent — we'll explore them in the next lesson.

Troubleshooting: If your agent doesn't respond, check that the Gateway is running with openclaw status. You can restart it with openclaw restart. For deeper issues, Claire recommends leveraging Claude Code for complex workspace troubleshooting.