Secure OpenClaw: A Guide to Not Giving Away the Keys to Your Life

What is OpenClaw?

OpenClaw is an open-source AI assistant that runs on your own hardware. Think of it as a self-hosted alternative to ChatGPT or Claude, but instead of chatting through a website, it lives on your computer(or on a Raspberry Pi in your closet) and connects with you via Signal, Telegram, Discord, Matrix, or whatever you prefer.

The appeal is obvious: you can message your assistant from your phone while you're out. It can read and write files. Execute commands. Remember things between conversations. Browse the web, manage your calendar, build apps and deploy them to Vercel. It's genuinely useful in ways that are different from copying and pasting into a chat window.

I won't be able to stop you from using OpenClaw, so at least let me teach you how to configure it so you don't give away the keys to your life — or at least, make it much harder.

By the end of this guide you'll have:

• OpenClaw on a Raspberry Pi, accessible only via Tailscale
• E2E encrypted chat via Matrix
• Prompt injection protection installed
• AI provider that claims not to keep logs
• Firewall, permissions, and habits that limit damage when something fails

Time: 30 minutes if everything goes well.

The Problem Nobody Wants to Discuss

This may sound paranoid. It's not.

When you give an AI assistant access to your files, your terminal, and your daily conversations, you're creating something unprecedented: a system that knows your work patterns, your personal relationships, your passwords (if you're not careful), your schedule, your writing style, your anxieties, your half-finished projects, and the embarrassing searches you asked for help with at 2am.

OpenClaw stores all of this. It has a MEMORY.md file that accumulates data about you over time. A credentials registry with all your secrets (API Keys, etc). Complete transcriptions of every conversation. Access to whatever tools you've enabled — which could include reading any file or executing arbitrary commands.

Three categories of risk

So why bother?

Because OpenClaw is genuinely useful in ways that ChatGPT and Claude are not. A web chatbot can't read your project files, run your scripts, send you a message every morning telling you it just built 5 demos based on the latest AI trend. OpenClaw can.

That power comes with real risk. But the answer is not to avoid it — it's to run it deliberately.

Requirements

The total cost is around $100-150 if you buy everything new, although you probably already have some of this at home.

Hardware

• •Raspberry Pi 5 (4GB+ RAM) — the 4GB model works fine because the heavy lifting is done by the AI provider
• •Quality microSD card (32GB+, reputable brand — cheap ones get corrupted)
• •USB-C power supply (official Pi recommended, 5V 3A)
• •Ethernet cable (WiFi works, but cable is more reliable for a headless server)

Required accounts

Step 1: Configure the Raspberry Pi

Flash the OS

1. Download Raspberry Pi Imager
2. Choose Device → Raspberry Pi 5
3. Choose OS → Raspberry Pi OS (64-bit)
4. Choose Storage → your microSD card

Configure before writing:

• • Hostname: openclaw
• • Username/password: pi with a strong password
• • WiFi if you're not using ethernet
• • Enable SSH: Yes
• • Public-key authentication only: Yes
• • Paste your public key from ~/.ssh/id_ed25519.pub

If you don't have an SSH key:

ssh-keygen -t ed25519 -C "your-email@example.com"
cat ~/.ssh/id_ed25519.pub  # Copy this to Pi Imager

First connection

ssh pi@openclaw.local
# Or: ssh pi@192.168.1.XXX

Update everything:

sudo apt update && sudo apt upgrade -y
sudo reboot

Enable automatic security updates:

sudo apt install unattended-upgrades -y
sudo dpkg-reconfigure -plow unattended-upgrades

Step 2: Configure Tailscale

curl -fsSL https://tailscale.com/install.sh | sh
sudo tailscale up

Open the URL it prints, authorize the device. Get your Tailscale IP:

tailscale ip -4
# Example: 100.100.100.100

sudo apt install ufw -y
sudo ufw default deny incoming
sudo ufw default allow outgoing
sudo ufw allow in on tailscale0 to any port 22
sudo ufw enable
sudo ufw status

To test:

# This should work:
ssh pi@YOUR_TAILSCALE_IP

# This should NOT work:
ssh pi@192.168.1.XXX  # Local IP - timeout

Step 3: Install OpenClaw

Install Node.js

# Install nvm
curl -o- https://raw.githubusercontent.com/nvm-sh/nvm/v0.40.3/install.sh | bash
. "$HOME/.nvm/nvm.sh"

# Install Node.js
nvm install 24
node -v  # Should show v24.x.x

Install OpenClaw

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

Onboarding

openclaw onboard

During onboarding:

• • Onboarding mode: manual
• • AI provider: Venice AI
• • Model: kimi-k2-5 (completely private)
• • Gateway bind: loopback
• • Gateway auth: token
• • Tailscale exposure: off
• • Skip messaging channel selection
• • Skip skills installation
• • Enable all hooks
• • Install gateway service
• • Skip hatching for now

Step 4: Configure Matrix (E2E)

Create Matrix accounts

You need two accounts on Element:

1. Your personal account — to chat with the bot from your phone
2. Bot account — the one OpenClaw will use

Install the plugin

openclaw plugins install @openclaw/matrix

If it fails with "npm install failed", fix it manually:

cd ~/.openclaw/extensions/matrix
sed -i 's/"workspace:\*"/"*"/g' package.json
npm install

Configure Matrix in OpenClaw

Edit ~/.openclaw/openclaw.json:

{
  "channels": {
    "matrix": {
      "enabled": true,
      "homeserver": "https://matrix-client.matrix.org",
      "userId": "@your_bot:matrix.org",
      "password": "YOUR_BOT_PASSWORD",
      "encryption": true,
      "dm": {
        "policy": "pairing"
      }
    }
  }
}

Run as a service

sudo nano /etc/systemd/system/openclaw.service

[Unit]
Description=OpenClaw AI Assistant
After=network-online.target
Wants=network-online.target

[Service]
Type=simple
User=pi
Group=pi
WorkingDirectory=/home/pi
ExecStart=/home/pi/.npm-global/bin/openclaw start
Restart=on-failure
RestartSec=10
NoNewPrivileges=true
ProtectSystem=strict
ProtectHome=read-only
ReadWritePaths=/home/pi/.openclaw

[Install]
WantedBy=multi-user.target

sudo systemctl daemon-reload
sudo systemctl enable openclaw
sudo systemctl start openclaw
sudo systemctl status openclaw

Step 5: Security Hardening

Install security skills

npx clawhub install skillguard
npx clawhub install prompt-guard

• • SkillGuard: Audits skills for security issues before installing
• • Prompt-Guard: Adds layers of resistance to prompt injection

File permissions

chmod 700 ~/.openclaw
chmod 600 ~/.openclaw/*.json
chmod 600 ~/.openclaw/credentials/*

Disable mDNS Broadcasting

echo 'export OPENCLAW_DISABLE_BONJOUR=1' >> ~/.bashrc
source ~/.bashrc
sudo systemctl restart openclaw

Run security audit

openclaw security audit --deep
# If there are issues:
openclaw security audit --fix

Operational Security

Technical hardening only goes so far. How you use the bot matters just as much.

If you get compromised

1. Stop immediately: sudo systemctl stop openclaw
2. Rotate all credentials
3. Review logs: less ~/.openclaw/logs/
4. Look for unauthorized changes: find ~/.openclaw -mtime -1 -ls
5. When in doubt: reflash the SD card — it's the only way to be sure

Limitations

Conclusion

You now have an AI assistant that:

• Runs on hardware you physically control
• Uses a provider that claims not to keep logs
• Has no public attack surface
• Uses E2E encrypted messaging
• Has prompt injection protection installed
• Only responds to your Matrix account

It's not perfectly secure. Nothing is. But it's better than pasting your life into ChatGPT.

Use your bot. Enjoy the convenience. Do it with your eyes open.