OpenClaw Scheduled Jobs & Unattended Runs on a Cloud Mac mini (LaunchAgent Playbook, 2026)

Interactive demos are easy—production automation means OpenClaw wakes up on Tuesday at 03:15 while your laptop is closed. On a rented Mac mini that job belongs to launchd, usually via a LaunchAgent plist running under the same macOS user that owns your OpenClaw state. Before you trust a plist, finish the macOS TCC, Accessibility & Screen Recording checklist in a real GUI session—otherwise headless jobs will “succeed” while doing nothing. This 2026 playbook compares cron versus LaunchAgent, shows the environment parity traps that break HTTPS proxies, and gives logging plus recovery steps aligned with install & deploy, production workflows, and troubleshooting & recovery. Security baselines remain in OpenClaw security & secrets.

TL;DR

• Use LaunchAgents in ~/Library/LaunchAgents for user-scoped OpenClaw jobs; avoid root cron unless you fully understand permission mapping.
• Mirror HTTPS_PROXY, NO_PROXY, and API-related env vars into the plist EnvironmentVariables dict—not only in ~/.zshrc.
• Send stdout/stderr to rotated files; alert on non-zero exit codes.
• After editing plists: launchctl bootout gui/$(id -u) then bootstrap to avoid duplicate daemons.

LaunchAgent vs Cron (Decision Table)

Option	Best when	Risk
LaunchAgent	Per-user OpenClaw CLI with Keychain secrets	Mis-typed plist keys silently fail
LaunchDaemon	System-wide tooling shared by multiple users	Harder TCC story; avoid unless required
cron	Calling a wrapper script that execs as the right user	No GUI session—Keychain prompts may stall

Recommended Plist Pattern

Keep arguments minimal: point ProgramArguments at the OpenClaw binary or your shell wrapper, and pass config paths explicitly. Use StartCalendarInterval for clock schedules or StartInterval for periodic polling. Pair with ThrottleInterval if vendor APIs rate-limit bursts.

Environment Parity Checklist

1. Copy interactive printenv output into a scratch file.
2. Diff against what the LaunchAgent sees via a temporary job that runs env > /tmp/openclaw-agent.env.
3. Align PATH to include Homebrew if you rely on /opt/homebrew/bin.
4. Confirm OPENCLAW_STATE_DIR points to local SSD—not a cloud-synced Desktop folder.

Logging & Rotation

Use StandardOutPath and StandardErrorPath targeting a directory your user owns, e.g. ~/Library/Logs/OpenClaw/. Rotate weekly or when files exceed ~100MB. For regulated teams, forward ERROR lines to your SIEM but strip model prompts. Cross-link noisy failures with deployment recovery before opening vendor tickets.

Failure Matrix

Symptom	Likely cause	Fix
Job never starts	Invalid plist XML or wrong label	plutil -lint + launchctl print gui/$(id -u)
401 from LLM API	Keychain unlock missing in headless mode	Pre-authorize keys via one VNC session per VNC guide
Duplicate executions	Two plists with overlapping labels	Delete stale plist, bootout, bootstrap once

Five-Step Recovery Runbook

1. Capture last 200 lines of stdout/stderr.
2. Verify network egress per proxy / WireGuard article.
3. Compare env diff between interactive shell and plist.
4. Restart job via launchctl kickstart -k gui/$(id -u)/label.here.
5. Document root cause in help ticket with timestamps for support.

Frequently Asked Questions

Should I ever use cron?

Only when a minimal wrapper elevates into the correct user context. LaunchAgent remains the macOS-native default for GUI-adjacent permissions.

Why did HTTPS_PROXY vanish?

LaunchAgents do not source shell profiles—duplicate variables in the plist or a shared config file on disk.

Where should logs live?

Use plist stdout/stderr paths under your home directory, rotate aggressively, and never log secrets.

Why ProxyMac Mac mini for Always-On OpenClaw

Laptops sleep; data centers do not. A dedicated HK / JP / KR / SG / US mini keeps LaunchAgents alive with stable power and cooling, while you keep human access via SSH per SSH vs VNC. Scale capacity from the pricing page when batch windows overlap.