Securing Your OpenClaw Agent

Security Model

OpenClaw runs with significant system access — file operations, shell commands, web browsing, and messaging. This power requires careful security practices.

API Key Management

Never Hardcode Keys

# Good: Environment variables
export OPENAI_API_KEY="sk-..."
export STRIPE_SECRET_KEY="sk_live_..."

# Bad: In config files committed to git
# apiKey: "sk-..." ← NEVER do this

Use .env Files

# .env (add to .gitignore!)
OPENAI_API_KEY=sk-...
ANTHROPIC_API_KEY=sk-ant-...

Rotate Regularly

Set a reminder to rotate API keys quarterly. If a key is ever exposed (e.g., in a chat log), revoke immediately.

Network Security

Firewall Rules

If your agent is internet-facing:

# Only allow SSH and HTTPS
sudo ufw default deny incoming
sudo ufw allow ssh
sudo ufw allow 443/tcp
sudo ufw enable

Reverse Proxy

Use Caddy or nginx as a reverse proxy with TLS:

# Caddyfile
agent.yourdomain.com {
    reverse_proxy localhost:3000
}

VPN Access

For maximum security, access your agent only via VPN (WireGuard, Tailscale):

# Tailscale — zero-config VPN
curl -fsSL https://tailscale.com/install.sh | sh
sudo tailscale up

Permission Boundaries

Agent Safety Rules

OpenClaw's AGENTS.md includes safety guidelines:

## Safety
- Don't exfiltrate private data
- Don't run destructive commands without asking
- trash > rm (recoverable beats gone forever)
- When in doubt, ask

External vs Internal Actions

Define clear boundaries: - Free to do: Read files, search web, organize workspace - Ask first: Send emails, post publicly, delete files

Data Protection

Memory Privacy

MEMORY.md contains personal context. OpenClaw's architecture ensures: - Memory files stay in the workspace - They're only loaded in main sessions (not group chats) - No cloud sync unless you configure it

Backup Your Workspace

# Regular backups
tar -czf agent-backup-$(date +%Y%m%d).tar.gz ~/agent/

Audit Logs

Review your agent's daily memory files periodically. They serve as an activity log of what the agent did.

Common Mistakes

1. Sharing tokens in chat — your agent might include them in memory
2. Running as root — always use a dedicated user account
3. Open ports — keep your agent behind a firewall
4. No backups — disk failures happen
5. Ignoring updates — keep OpenClaw and Node.js updated

---

Ready to go deeper? Check out our hands-on course: OpenClaw Agent — practical exercises you can follow along on your own machine.

Related guide

Related reading: the Ansible intelligent assistant and MCP server covers this in real-world detail.