Automations

48 launchd jobs running on a Mac Mini in Singapore. The boring-but-load-bearing crons that keep the agents on the rails. Sourced from ~/Library/LaunchAgents/com.voltade.* and com.yash.*.

Customer pipeline

Lead intake, takeover-watch, conversation QA, message archival. Mostly Ambassador and Claudia signals.

  • ambassador-extractor every 30s

    Pulls new WhatsApp leads into the queue for Ambassador to chat with.

  • ambassador-lead-alert every 1m

    Pings the team on Telegram when a new high-signal lead appears.

  • ambassador-takeover-watcher every 1m

    Detects when a human steps into an Ambassador chat, stops the agent.

  • ambassador-qa Sun 10:00

    Weekly LLM-as-judge eval over Ambassador conversations.

  • ambassador-throughput-weekly Fri 18:00

    Weekly digest: leads in, leads handed off, conversion to demo.

  • main-agent-qa Sun 10:30

    Weekly LLM-as-judge eval over Claudia (main customer agent) conversations.

  • wa-chaser every 30m

    Re-engages stale customer threads that the team owes a reply on.

  • wa-log-archiver every 5m

    Archives raw WhatsApp logs to long-term storage.

  • outbound-replay every 5m

    Replays any failed outbound WhatsApp sends from the queue.

  • customer-dri-refresh daily 07:00

    Rebuilds the customer → DRI mapping used by routing.

Voltade internal ops

Standup, morning brief, weekly ops digest, tender scanning, drift detection. The boring-but-load-bearing stuff.

  • standup-init Mon–Fri 09:00

    Starts the daily standup thread on Telegram.

  • standup-nudge Mon–Fri 10:30

    Nudges anyone who has not posted in standup yet.

  • standup-summary Mon–Fri 11:00

    Summarises everyone’s standup into a single team-readable digest.

  • morning-brief daily 08:00

    Personal morning digest: agent activity, gym, trading, calendar.

  • ops-digest daily 01:00

    Overnight roll-up of Voltade ops state.

  • drift-detector daily 09:30

    Scans agent outputs for stale-recitation drift (the gym-schedule pattern, the wrong-price pattern).

  • llm-spike daily 09:30

    Pages me if a single agent burned an outsized chunk of yesterday’s spend.

  • self-improve-review daily 08:17

    Reviews agent corrections from the day and proposes skill patches.

  • talenox-team-pulse daily 09:00

    Pulse-check from Talenox HR: who is on leave, who joined, who left.

  • ccclub-daily daily 18:00

    Posts the daily Claude Code Club update.

  • gebiz-daily-scan daily 10:00

    Scans GeBIZ (Singapore Gov tender portal) for tenders we should bid.

  • queue-worker always on

    Background worker that drains all OpenClaw task queues.

  • reminder-gspl-demo daily 09:07

    Reminder ping for an ongoing customer demo cycle.

Cost & billing

Weekly Anthropic / OpenAI bills, daily usage pacing, Xero AR. Cost is a product feature.

  • anthropic-cost-weekly Mon 09:00

    Weekly Anthropic spend digest broken down by agent.

  • openai-cost-weekly Mon 09:05

    Weekly OpenAI spend digest.

  • usage-pacer 12:00, 16:00, 20:00

    Three daily pacing checks so we catch a spend spike before midnight.

  • anthropic-news-check daily 08:05

    Watches for new Anthropic announcements / model releases.

  • claude-code-release-check daily 08:00

    Watches for Claude Code releases I should adopt.

  • xero-ar-brief Fri 09:30

    Friday A/R brief: who owes us, how overdue, suggested chase.

  • xero-auth-keepalive Mon 09:00

    Refreshes Xero OAuth token before it expires.

OpenClaw infrastructure

The harness watching itself. Health checks, error watchers, backups, patches.

  • openclaw-health-check every 15m

    Synthetic ping across every agent, alerts if anyone fails.

  • openclaw-error-watcher every 5m

    Tails error logs across all agents, alerts on new exceptions.

  • openclaw-doctor daily 09:15

    Daily doctor: checks the harness can boot, cron health, credentials valid.

  • openclaw-cron-audit every 4h

    Audits that all expected launchd jobs are actually loaded and not crashing.

  • openclaw-apply-patches every 4h

    Pulls and applies any agreed agent skill patches.

  • openclaw-backup every 2h

    Incremental backup of the OpenClaw config + memory.

  • openclaw-backup-weekly Sun 04:00

    Full weekly backup of the OpenClaw fleet.

  • openclaw-security-weekly Mon 06:00

    Weekly security audit of the OpenClaw config and credentials.

  • claude-creds-backup every 5m

    Backs up Claude credentials so a fresh machine can take over.

  • github-status-watcher every 1m

    Watches GitHub status, alerts before agents try to push during downtime.

  • dangerclaude-tmux every 1m

    Keeps a perpetual `claude --dangerously-skip-permissions` tmux session alive.

  • pakgu-pei-cleanup daily 02:00

    Nightly cleanup of Pakgu’s working memory artefacts.

  • media-cleanup Sun 03:00

    Weekly clean-up of large media files cached by the WhatsApp pipeline.

Personal

Mine, not Voltade’s. Gym, trading, local model UI.

  • gym-booker daily 08:57

    Auto-books my gym classes 2 days in advance.

  • gym-weekly-checkin Fri 09:15

    Weekly summary of which classes I actually attended.

  • iplus-booker daily 21:55

    Books an iPlus class for next-day windows.

  • donchian-live daily 08:05

    Donchian-channel crypto trading strategy, runs the morning entry/exit pass.

  • ollama-web always on

    Local Ollama web UI for off-cloud model experiments, served over Tailscale.

No alert channel column because the answer is mostly the same: failure pages me on Telegram via the Happy bot. The diagram on Projects shows where each lives.