OpenClaw Session Cleanup

Use this skill when the runtime becomes unstable after long uptime, especially on small hosts such as 1 vCPU / 2 GB RAM.

Trigger Pattern

Treat these as the main signals:

• Sessions: 10+ active
• Agents: 5+
• browser control service timeout
• gateway 1006 abnormal closure
• repeated gateway disconnects or slow recovery after idle periods

What To Do

1. Inspect the current runtime state.
2. Prune stale sessions first.
3. If the runtime stays unhealthy, clear sessions.
4. Reduce runtime ceilings for sessions, agents, and browsers.
5. Add recurring cleanup and a watchdog when the host is expected to run for days.

Immediate Commands

Run:

openclaw sessions
openclaw sessions prune
openclaw status

If the runtime is still unhealthy, escalate to:

openclaw sessions clear
openclaw status

Healthy target after cleanup:

• Sessions: 1-3
• Agents: 3
• Gateway reachable
• browser running

Safe Runtime Defaults

For 1 vCPU / 2 GB RAM, prefer:

• maxSessions = 5
• sessionTTL = 30m
• maxAgents = 3
• maxBrowsers = 1
• swap = 2G

Use the starter config at {baseDir}/templates/openclaw.json.

Automation

Install periodic pruning:

bash "{baseDir}/scripts/install-cron-prune.sh"

Install watchdog unit files:

bash "{baseDir}/scripts/install-watchdog.sh"

Render a tuned runtime config:

bash "{baseDir}/scripts/render-openclaw-config.sh"

Browser Constraint

Treat browser control as the heaviest resource consumer on a small VPS.

Recommended mode:

openclaw browser start --single

Swap

If the host has no swap, add 2G:

fallocate -l 2G /swapfile
chmod 600 /swapfile
mkswap /swapfile
swapon /swapfile
free -h

References

• Detailed runbook: {baseDir}/docs/openclaw.session_cleanup_v1.md
• Runtime template: {baseDir}/templates/openclaw.json
• Watchdog service: {baseDir}/templates/openclaw-watchdog.service
• Watchdog timer: {baseDir}/templates/openclaw-watchdog.timer