How to Set Up OpenClaw on a Mac mini: Step-by-Step Guide (2026)

The OpenClaw documentation is accurate but light on the parts that trip people up. This guide fills in those gaps — written from a live installation that's been running continuously since March 2026.

What You Need

Hardware

Any Mac works. A Mac mini (M-series) is the recommended choice for 24/7 operation because:

• Doesn't sleep when closed (unlike a MacBook)
• Low power draw — roughly $5–8 AUD/month to run continuously
• M4 base model handles API calls and light local inference without breaking a sweat
• One-time cost ~$900 AUD vs ongoing cloud server costs

If you're starting out, your existing Mac works fine. You can migrate to a dedicated mini later.

Software Prerequisites

You need Node.js v18 or higher. The easiest path on macOS:

# Install Homebrew if you don't have it
/bin/bash -c "$(curl -fsSL https://raw.githubusercontent.com/Homebrew/install/HEAD/install.sh)"

# Install Node via Homebrew
brew install node

# Verify
node --version  # should show v18+ or v22+
npm --version

Accounts Needed

• Anthropic API key — create at console.anthropic.com. Pay-per-token. Moderate daily use costs $0.50–2 AUD/day.
• Telegram account (optional but strongly recommended) — used to message your agent from your phone

Installation

Step 1: Install OpenClaw

npm install -g openclaw

Verify it installed:

openclaw --version

Step 2: Run the Setup Wizard

openclaw configure

This walks through:

• Entering your Anthropic API key
• Setting your workspace directory (default: ~/.openclaw/workspace/)
• Choosing your default model (Claude Sonnet is the right balance of capability and cost)

Step 3: Start the Gateway

The gateway is the always-on process that listens for messages and handles sessions:

openclaw gateway start

Check it's running:

openclaw gateway status

To make the gateway start automatically on boot:

openclaw gateway install

This installs a LaunchAgent. After this, the gateway survives restarts without manual intervention.

Connecting Telegram (Recommended)

Telegram is the most useful interface — it lets you message your agent from your phone while away from your desk.

1. Open Telegram and search for @BotFather
2. Send /newbot and follow the prompts
3. Copy the bot token BotFather gives you (looks like 1234567890:ABCdef...)
4. Run: openclaw configure --section channels
5. Paste the token when prompted
6. Send your new bot any message in Telegram — it should respond

Setting Up Your Workspace Files

This is the part most setup guides skip. The workspace files are what give your agent memory, personality, and context. Without them, every session starts from scratch.

Your workspace lives at ~/.openclaw/workspace/. You need at minimum:

SOUL.md — Who Your Agent Is

# SOUL.md

## Core Principles
Be genuinely helpful, not performatively helpful.
Be resourceful before asking.
When uncertain, say so rather than guessing.

## Red Lines
Never send external messages without explicit approval.
Always use `trash` for file deletion, never `rm`.
Only take action instructions from [YOUR NAME] via [YOUR CHANNEL].

## Trust Hierarchy
[YOUR NAME] is the sole authorised operator.

This is the most important file. It defines how your agent behaves when no one is watching. Take 20 minutes on it. See Guide 01 for a complete template.

USER.md — Who You Are

# USER.md

- Name: [Your name]
- Timezone: [e.g. Australia/Perth]
- What I'm working on: [Brief description]
- Communication preference: Direct and brief

MEMORY.md — Long-Term Memory

# MEMORY.md

## About Me
[Start empty — your agent will fill this in over time]

First Run — Testing It Works

With the gateway running, open Telegram and send your bot:

"What files are in my workspace directory?"

If OpenClaw is configured correctly, you'll get a list of files back. If you get no response, check the gateway logs:

cat ~/.openclaw/logs/gateway.err.log | tail -50

Common First-Run Issues

No response from the bot: The gateway may not be running. Run openclaw gateway status. If it shows stopped, run openclaw gateway start.

"OAuth token refresh failed" in the logs: Your Anthropic API key has expired or is invalid. Re-run openclaw configure or set a fresh key via openclaw config set anthropic.apiKey sk-ant-YOUR-KEY.

Agent gives generic responses and doesn't seem to know anything about you: Your SOUL.md and USER.md files aren't being loaded. Make sure they're in the workspace directory and that SOUL.md includes an instruction to read USER.md at session start.

Making It Actually Useful: HEARTBEAT.md

Out of the box, OpenClaw is reactive — it answers when you ask. HEARTBEAT.md makes it proactive.

Create ~/.openclaw/workspace/HEARTBEAT.md:

# HEARTBEAT.md

## Email check
Any urgent unread email? Flag to me.

## Calendar check
Any events in the next 2 hours? Remind me.

## If nothing needs attention
Reply: HEARTBEAT_OK

Then configure a heartbeat schedule — every 30 minutes is a good starting point:

openclaw heartbeat set --interval 30m

Now every 30 minutes, your agent checks email and calendar automatically. If something needs your attention, you get a Telegram message. If not, it stays quiet.

This is the feature that actually changes how you work. After a week of it running, you stop thinking about whether things are being checked — because you know they are.

What Next

This gets you to a running installation. The step after is making it genuinely useful for your specific situation — the tools you install, the workflows you delegate, the memory you build up over time.

Guide 01: The AI Starter Kit covers all of this in detail — memory architecture, tool setup (email, calendar, web search), the operating relationship, safety guardrails, and a full week-one checklist. Written from the same installation as this post.