Skip to main content

Day 7: Lock It Down

You have a personalized, multi-channel, automated AI assistant. Today you make sure it stays running, stays safe, and doesn't surprise you with a bill.

Time: ~45 minutes

Security in 5 Minutes​

The full Security Overview covers threat models and incident history. Here's the condensed version — do all five steps right now.

  1. Update to latest: npm update -g openclaw
  2. Verify localhost binding — never expose the gateway to the network
  3. Verify auth is enabled — token mode at minimum
  4. Set channel allowlists — only accept messages from known senders
  5. Run openclaw security audit --deep — fix whatever it flags

Here's every security-relevant setting in one config block:

~/.openclaw/openclaw.json
{
  "gateway": {
    "host": "127.0.0.1",       // NEVER use 0.0.0.0
    "port": 18789,
    "auth": {
      "mode": "token"          // Requires auth for all gateway connections
    }
  },
  "channels": {
    "telegram": {
      "allowed_chat_ids": [123456789]   // Your Telegram user ID only
    },
    "discord": {
      "allowed_server_ids": ["987654321"]
    }
  },
  "security": {
    "confirm_dangerous_actions": true,   // Prompt before shell commands, file deletes
    "browser_domain_allowlist": [        // Restrict browser automation targets
      "github.com",
      "docs.google.com"
    ],
    "quiet_hours": {
      "enabled": true,
      "start": "23:00",
      "end": "07:00",
      "timezone": "UTC"                  // Matches your USER.md timezone
    }
  }
}

Run the audit to confirm everything is set correctly:

openclaw security audit --deep
openclaw security audit --fix    # Auto-fix what it can
tip

Set quiet_hours to match when you're asleep. During quiet hours, the bot holds non-urgent actions until you're back.

Cost Check​

AI APIs charge per token. A chatty bot with a powerful model can rack up real costs. Here's how to see what you're spending.

Check token usage​

# Token usage summary for the current billing period
openclaw stats tokens

# Cost breakdown by model, channel, and task type
openclaw gateway usage-cost

Check your provider dashboard​

If you're using OpenRouter, visit openrouter.ai/activity for a detailed breakdown by model and time period. Other providers have similar dashboards — check your API account settings.

Quick cost-reduction tips​

  • Use a cheap heartbeat model. Your heartbeat runs every few minutes. It doesn't need GPT-4 or Claude Opus — a small, fast model handles "check if anything needs attention" just fine. See Model Selection Guide for recommended pairings.
  • Increase heartbeat intervals. Every 5 minutes is aggressive. Every 15 or 30 minutes is fine for most setups.
  • Set quiet hours. No heartbeat, no token burn while you sleep.
  • Keep sessions short. Long conversations accumulate context. If a conversation has been going for 50+ messages, start a new one — the bot has your memory files for continuity.

Monitoring​

A production setup means knowing when something goes wrong before it becomes a problem.

Quick health check​

openclaw status

This shows gateway status, connected channels, active tasks, memory usage, and any warnings.

Watch real-time logs​

openclaw logs --follow

Useful for debugging channel issues, watching skill execution, or monitoring what the heartbeat is doing. Press Ctrl+C to stop.

Watch memory growth​

Long-running OpenClaw processes can grow in memory over time, especially with browser automation or large context windows.

# Check RSS (resident set size) — the actual RAM usage
ps aux | grep openclaw

If RSS climbs past 500-600 MB consistently, restart the process. A healthy idle instance sits around 150-250 MB.

Self-monitoring heartbeat​

Set up a heartbeat task that watches OpenClaw itself. In your heartbeat config or via chat:

Check your own health — CPU usage, memory, disk space, and whether
all channels are connected. If anything looks wrong, send me a
Telegram message with the details.

This gives you a bot that monitors itself and alerts you when it needs attention.

Backups​

OpenClaw stores everything in ~/.openclaw/. Back it up before upgrades, experiments, or just on a regular schedule.

# Back up everything that matters
tar czf ~/openclaw-backup-$(date +%Y%m%d).tar.gz \
  ~/.openclaw/openclaw.json \
  ~/.openclaw/workspace/ \
  ~/.openclaw/memory/ \
  ~/.openclaw/skills/

What each directory contains:

PathContents
~/.openclaw/openclaw.jsonAll configuration — models, channels, security, heartbeat
~/.openclaw/workspace/SOUL.md, IDENTITY.md, USER.md, and any workspace files
~/.openclaw/memory/Persistent memory — what the bot remembers about you and past sessions
~/.openclaw/skills/Installed skills from ClawHub and any custom skills you've built
info

Credentials (API keys, tokens) are in openclaw.json. Treat your backup file with the same care you'd give a password database.

Upgrades​

OpenClaw moves fast. Upgrades usually go smoothly, but the community has been burned before. Follow this procedure.

1. Back up first (use the command above)

2. Check release notes:

gh release view --repo openclaw/openclaw

Read the notes. Look for breaking changes, migration steps, or known issues.

3. Update:

npm update -g openclaw

4. Run doctor:

openclaw doctor

Doctor checks that your config, dependencies, and environment are compatible with the new version.

5. Verify:

openclaw status

Confirm all channels reconnect and the heartbeat is running.

6. If something breaks: restore from backup.

# Stop OpenClaw first
openclaw stop

# Restore config and data
tar xzf ~/openclaw-backup-YYYYMMDD.tar.gz -C /

# Downgrade if needed
npm install -g openclaw@<previous-version>

# Restart
openclaw start
warning

Some past upgrades (v2026.5.x) have wiped configs. Always back up before upgrading. This is not theoretical — it has happened to real users.

What's Next?​

You're past the basics. Here's where to go from here:

Congratulations​

You've gone from "hello" to a production-ready, multi-channel, automated AI assistant in 7 days. Your bot has a personality, handles daily tasks, runs on a schedule, speaks across platforms, and is locked down for safe long-term operation. The rest is up to you and your imagination.