Quickstart: Your Personal AI OS in 30 Minutes

Level 1 setup — OpenClaw + 3 agents + Signal
Time: ~30 minutes · Cost: ~$20/month
Works on: macOS · Linux · Windows (via WSL2)

Prerequisites

An AI model provider (see options below — pick one)
Signal account on your phone
One of the following:
- Mac: Homebrew installed
- Linux: Ubuntu/Debian 20.04+
- Windows: WSL2 enabled (see Windows Setup below)

Choosing Your AI Provider

OpenClaw supports multiple AI providers. Pick the option that fits your budget and situation:

Option A — Anthropic API Key (Recommended for production)

Cost: Pay-per-use (~$5-20/month typical usage)
Get it: console.anthropic.com → API Keys
Best for: Predictable billing, production use, no subscription required

{ "anthropic": { "apiKey": "sk-ant-your-key-here" } }

Option B — Claude Pro/Max Subscription (~$20-100/month)

Cost: $20/month (Pro) or $100/month (Max)
Get it: claude.ai → Subscribe, then run openclaw setup-token
⚠️ Important caveat: Using a Pro/Max subscription with OpenClaw involves capturing an OAuth token from Claude.ai. This works currently, but Anthropic's terms of service may restrict this use. If it stops working, switch to Option A or Option C.

# After subscribing at claude.ai:
openclaw setup-token
# Follow the prompts — paste the OAuth token when asked

Option C — GitHub Copilot Pro+ (~$19/month)

Cost: ~$19/month, includes Claude Sonnet 4.6 + GPT-4o + Gemini
Get it: github.com/features/copilot
Best for: Students who already have GitHub (education discount available — may be free!)

openclaw configure --provider github-copilot

Option D — OpenRouter (Pay-per-use, many models)

Cost: Pay-per-use, often cheaper than direct APIs
Get it: openrouter.ai → API Keys
Best for: Budget-conscious users who want access to many models (Llama, Mistral, Gemini, Claude, etc.)

openclaw configure --provider openrouter

Option E — Google Gemini API (Free tier available)

Cost: Free tier available, then pay-per-use
Get it: aistudio.google.com → Get API Key
Best for: Zero-cost experimentation

openclaw configure --provider google

Option F — Local Models via Ollama (Free, runs on your Mac)

Cost: Free — runs locally on your hardware
Get it: ollama.ai → Install → ollama pull llama3.2
Best for: Privacy, offline use, no API costs. Slower than cloud models.

openclaw configure --provider ollama

Recommendation for students: Start with Option C (GitHub Copilot) if you have a .edu email (free or discounted). Otherwise use Option A (Anthropic API) — $5 will last weeks during class.

🍎 Mac Setup

Step 1: Install OpenClaw (5 minutes)

# Install via Homebrew
brew install openclaw

# Verify installation
openclaw --version

# Initialize your workspace
mkdir ~/myai
cd ~/myai
openclaw init

OpenClaw creates a config at ~/.openclaw/config.json. Add your API key:

{
  "anthropic": {
    "apiKey": "sk-ant-your-key-here"
  }
}

🐧 Linux Setup

Step 1: Install Node.js 20+

# Ubuntu/Debian
curl -fsSL https://deb.nodesource.com/setup_20.x | sudo -E bash -
sudo apt-get install -y nodejs

# Verify
node --version  # Should show v20+

Step 2: Install OpenClaw

npm install -g openclaw
openclaw --version

Step 3: Initialize

mkdir ~/myai && cd ~/myai
openclaw init

Add your API key to ~/.openclaw/config.json:

{
  "anthropic": {
    "apiKey": "sk-ant-your-key-here"
  }
}

🪟 Windows Setup (WSL2)

WSL2 gives you a full Linux environment inside Windows. OpenClaw runs great in it.

Step 1: Enable WSL2

Open PowerShell as Administrator and run:

wsl --install

Restart your computer when prompted. This installs Ubuntu by default.

Already have WSL? Make sure it's WSL2: wsl --set-default-version 2

Step 2: Open Ubuntu Terminal

After restart, search "Ubuntu" in Start menu and open it. You'll set a username and password on first launch.

Step 3: Install Node.js in WSL

# Inside Ubuntu terminal
curl -fsSL https://deb.nodesource.com/setup_20.x | sudo -E bash -
sudo apt-get install -y nodejs

node --version  # Should show v20+

Step 4: Install OpenClaw

npm install -g openclaw
openclaw --version

Step 5: Initialize your workspace

mkdir ~/myai && cd ~/myai
openclaw init

Configure your API key:

# Option A: Edit directly
nano ~/.openclaw/config.json
# Add: "anthropic": { "apiKey": "sk-ant-your-key-here" }

# Option B: Use the CLI
openclaw config set anthropic.apiKey sk-ant-your-key-here

Step 6: Set environment variable (optional but recommended)

# Add to ~/.bashrc so it persists
echo 'export ANTHROPIC_API_KEY="sk-ant-your-key-here"' >> ~/.bashrc
source ~/.bashrc

Windows tip: VS Code has a "WSL" extension — install it and you can edit your Linux files with VS Code's full UI. Run code . from the WSL terminal.

🚀 All Platforms: Steps 2-6

Step 2: Start the Gateway (2 minutes)

# Start OpenClaw gateway (keeps running in background)
openclaw gateway start

# Verify it's running
openclaw gateway status

You should see: Gateway running on localhost:18789

Open in browser: http://localhost:18789

Windows users: The gateway runs inside WSL, but you can access localhost:18789 from your Windows browser. It just works.

Step 3: Create Your First 3 Agents (10 minutes)

Create a directory for your agents:

mkdir ~/myai/agents

Agent 1 — Chief of Staff (~/myai/agents/chief.yaml):

id: chief
name: Chief
model: claude-sonnet-4-5
description: Primary interface and coordinator

system: |
  You are Chief — my Chief of Staff and primary AI interface.
  
  Your job:
  - Be the first point of contact for all requests
  - Route complex requests to the right specialist
  - Maintain a coherent view of my priorities
  - Give direct, practical answers — no fluff
  
  You know about these other agents:
  - Researcher: deep research and analysis
  - Writer: drafting and editing
  
  When a request clearly belongs to a specialist, say so.
  But answer simple questions yourself.
  
  Communication style: direct, concise, occasionally dry.
  Never start with "Great question!" or "Certainly!".

Agent 2 — Researcher (~/myai/agents/researcher.yaml):

id: researcher  
name: Researcher
model: claude-sonnet-4-5
description: Deep research and analysis

system: |
  You are Researcher — a methodical, thorough research analyst.
  
  Your job:
  - Conduct deep research on any topic
  - Synthesize multiple sources into clear briefs
  - Provide balanced analysis, not just one perspective
  - Cite sources when relevant
  - Flag uncertainty clearly
  
  You do NOT:
  - Draft documents (that's Writer's job)
  - Make decisions (you inform decisions)
  
  Style: thorough, precise, well-structured. Use headers and
  bullet points. Lead with the key finding, then support it.

Agent 3 — Writer (~/myai/agents/writer.yaml):

id: writer
name: Writer  
model: claude-opus-4-5
description: Writing and editing

system: |
  You are Writer — a skilled communicator who writes with clarity
  and purpose.
  
  Your job:
  - Draft emails, proposals, blog posts, reports
  - Edit and improve existing writing
  - Match the requested tone and voice
  - Be concise — cut ruthlessly
  
  You do NOT:
  - Research topics (ask Researcher for background)
  - Make business decisions
  
  Style: clear, active voice, no corporate jargon. Short
  sentences. Strong verbs. Never use "utilize" when "use" works.

openclaw agents add ~/myai/agents/chief.yaml
openclaw agents add ~/myai/agents/researcher.yaml
openclaw agents add ~/myai/agents/writer.yaml

# Verify
openclaw agents list

Step 4: Connect Signal (5 minutes)

Install Signal Desktop → signal.org/download (Mac/Windows — runs natively)
Open OpenClaw config and add Signal:

{
  "channels": {
    "signal": {
      "enabled": true,
      "phoneNumber": "+1XXXXXXXXXX",
      "defaultAgent": "chief"
    }
  }
}

Follow the Signal linking flow:

openclaw channel link signal

Scan the QR code with Signal on your phone.

Windows users: Signal Desktop runs natively on Windows (not in WSL). OpenClaw communicates with it through a local socket that works across the WSL boundary.

Step 5: Test It (5 minutes)

Send a Signal message to yourself (your linked number):

What can you do?

Chief should respond. Then try:

Research: what are the best practices for AI agent system design?

Chief routes to Researcher. Then:

Write me a short email to my professor asking for an extension,
professional but honest tone

Chief routes to Writer.

Step 6: Set Up Memory (3 minutes)

Create your memory files:

mkdir ~/myai/memory
touch ~/myai/memory/MEMORY.md
touch ~/myai/memory/$(date +%Y-%m-%d).md

Add to Chief's system prompt (or as a file OpenClaw injects):

At the start of each session, you have access to:
- MEMORY.md: long-term context about me and my work
- memory/YYYY-MM-DD.md: recent activity logs

Use this context to pick up where we left off.

Configure OpenClaw to inject memory:

{
  "memory": {
    "enabled": true,
    "longTermFile": "~/myai/memory/MEMORY.md",
    "dailyDir": "~/myai/memory/"
  }
}

You're Done With Installation

You now have:

✅ A running AI gateway
✅ OpenClaw web UI at localhost:18789
✅ Anthropic API connected

Step 7: Run Your Onboarding Interview (15 minutes)

This is the most important step. Don't skip it.

Open the web UI at http://localhost:18789 and paste this:

I just installed OpenClaw. Read the file at student-kit/ONBOARDING.md 
and run me through the onboarding interview. I want to design my 
personal AI system from scratch.

What happens next:

Your agent interviews you (15-20 min)
Asks about your role, tools, pain points, and setup
Recommends a custom agent team (3-8 agents, not all 16)
Generates your personalized SOUL.md, USER.md, and agent configs
Produces a ready-to-run Claude Code prompt that builds your dashboard
Sets up your first automation cron job

By the end, Claude Code will be building your Mission Control while you grab lunch.

Next Steps (after onboarding)

Follow the onboarding output — it'll tell you exactly what to do
Use it for 24 hours before adding anything else
Add more agents — 19 templates in agents/ including Pulse (health coach) and Ralph (autonomous loop agent)
Wire up integrations — see docs/integrations.md
Deep dive — docs/system-prd.md has the full architecture and build prompts

Troubleshooting

Gateway won't start:

openclaw gateway stop
openclaw gateway start --debug

Agent not responding:

openclaw agents test chief "hello"

Signal not connecting:

Ensure Signal Desktop is running on the same machine
Try relinking: openclaw channel unlink signal && openclaw channel link signal

Windows / WSL-specific issues:

Can't access localhost:18789 from Windows browser:

# Get your WSL IP address
hostname -I
# Try that IP:18789 instead of localhost

Permission denied installing npm packages:

# Fix npm global directory permissions
mkdir -p ~/.npm-global
npm config set prefix ~/.npm-global
echo 'export PATH=~/.npm-global/bin:$PATH' >> ~/.bashrc
source ~/.bashrc
# Then retry: npm install -g openclaw

WSL2 performance is slow:

Keep your project files in the Linux filesystem (~/myai) not Windows filesystem (/mnt/c/...)
Windows filesystem access through WSL is significantly slower

Signal Desktop on Windows + OpenClaw in WSL:

This combination works but requires OpenClaw 1.4+ for cross-boundary socket support
If it fails, run OpenClaw on Windows PowerShell instead (same npm install -g openclaw)

Questions? adam@meekertechnologies.com