The OpenClaw Zero-Impact Playbook

From the founder

If you've been following the OpenClaw and Anthropic story — or someone sent you this because you were freaking out in the comments — I put this together for you.

I'm Kelvin Craver. I run Crave Digital, an AI automation agency in Los Angeles. I've been deep in Claude Code infrastructure, building autonomous agent pipelines, and running AI-powered systems across multiple businesses. When the April 4 announcement dropped, my inbox started moving.

Honest truth: most people impacted by this change set it up wrong to begin with. They were using a flat $20 or $200 subscription to power unlimited agentic workflows — workflows that consume 15–20x the compute of normal use. Anthropic was quietly subsidizing that gap. Now they're not.

But if you're already running OpenClaw on GPT-4.1 — or running it completely free on Gemma 4 locally? Nothing changed for you. Not one dollar. Not one workflow.

This playbook walks you through exactly what happened, who's actually affected, and the smartest architecture to run OpenClaw on any model you want — including completely free options — without putting yourself at one provider's mercy again.

Use it. Share it. And if you want help setting this up for your business, you know where to find me.

What actually happened

Let's cut through the noise. Here's the timeline:

Are you impacted?

Most people reading this aren't affected as badly as the headlines suggest. Quick matrix:

The smart multi-lane stack

Here's what I recommend — and what I personally run. The key idea: don't depend on one company. Use the right tool for each job. Keep the billing completely separate. You don't need to be technical to do this.

Other free and low-cost options

• Gemini Flash (Google): Free cloud tier — 15 req/min, no credit card. Great for light usage.
• Qwen3 32B: Free local model. Needs 24GB+ RAM. Strong at coding.
• DeepSeek R1 14B: Free local model with strong reasoning. Runs on 16GB RAM.
• Llama 3.3 70B: Meta's open model. Needs 64GB+ RAM but very capable.

Get OpenClaw running

mkdir my-agent && cd my-agent
openclaw init    # follow the setup wizard

{
  "env": {
    "OPENAI_API_KEY": "sk-YOUR_KEY_HERE"
  },
  "agents": {
    "defaults": {
      "model": { "primary": "openai/gpt-4.1" }
    }
  }
}

{
  "agents": {
    "defaults": {
      "model": { "primary": "ollama/gemma4:26b" }
    }
  },
  "api": "ollama"
}

"messaging": {
  "telegram": { "token": "YOUR_BOT_TOKEN" }
}

openclaw start            # foreground
openclaw start --daemon   # background
openclaw logs             # view logs

Choose your deployment

OpenClaw runs anywhere Node.js runs. Pick the option that fits your situation:

VPS quick setup (DigitalOcean)

ssh root@YOUR_IP_ADDRESS
apt update && apt install -y nodejs npm
npm install -g openclaw
mkdir my-agent && cd my-agent
openclaw init

cat > /etc/systemd/system/openclaw.service << 'EOF'
[Unit]
Description=OpenClaw Agent
After=network.target
[Service]
Type=simple
WorkingDirectory=/root/my-agent
ExecStart=/usr/bin/openclaw start
Restart=always
[Install]
WantedBy=multi-user.target
EOF
systemctl enable openclaw
systemctl start openclaw
systemctl status openclaw

Lock down your setup

Your AI agent has access to API keys, messaging accounts, and potentially sensitive data. Treat security seriously from day one.

1. Never expose API keys

Store keys in environment variables or a .env file. Never hardcode. Never commit to GitHub. Add .env to .gitignore immediately.

# Create .env (never commit)
OPENAI_API_KEY=sk-your-key-here
TELEGRAM_TOKEN=your-bot-token

echo ".env" >> .gitignore

2. Set spending limits

Both OpenAI and Anthropic let you cap monthly spending. Do this before running your agent. Start at $20–50/mo and increase as you learn your usage.

• OpenAI: platform.openai.com → Billing → Usage Limits
• Anthropic: console.anthropic.com → Billing → Spending Limits

3. Rotate keys every 90 days

Generate new keys quarterly. Delete old keys immediately. If you suspect a key was exposed, revoke it and create a new one right away.

4. Secure your VPS

# SSH keys instead of passwords
ssh-keygen -t ed25519
ssh-copy-id root@YOUR_IP

# Disable password login
sudo sed -i 's/PasswordAuthentication yes/PasswordAuthentication no/' /etc/ssh/sshd_config
sudo systemctl restart sshd

# Enable firewall
sudo ufw allow OpenSSH
sudo ufw enable

5. Monitor usage

Check your API dashboard weekly. Unusual spikes could indicate a leaked key or runaway agent. Set billing alerts in both OpenAI and Anthropic consoles.

What this actually costs

Two-lane setup means two separate bills. Full picture:

Everything you need to do