The Heartbeat & Cron System

Right now, your agent sits there doing absolutely nothing until you type something. That's like hiring an employee and telling them "only work when I'm watching you." Let's fix that. This chapter turns your reactive chatbot into a proactive operator that works 24/7.

Two Ways to Make Your Agent Proactive

When to Use Which

This is the question everyone asks, so let's make it crystal clear:

Setting Up Heartbeat

The heartbeat system is controlled by a simple file called HEARTBEAT.md. Your agent reads this file every ~30 minutes and performs whatever checks are listed:

# Heartbeat Checklist

## Every Check (~30 min)
- [ ] Any urgent unread emails?
- [ ] Any mentions/DMs on Twitter?
- [ ] Any calendar events in next 2 hours?

## 2-4x Per Day
- [ ] Check project build status (CI/CD)
- [ ] Review social media engagement on recent posts

## Once Daily (morning)
- [ ] Weather forecast (human goes running)
- [ ] Market summary (if trading cron hasn't run yet)

## Rules
- Late night (11 PM - 8 AM): HEARTBEAT_OK unless truly urgent
- If nothing actionable: reply HEARTBEAT_OK (saves tokens)
- Batch updates into ONE message, don't spam
- Track last check times in memory/heartbeat-state.json

{
  "lastChecks": {
    "email": "2026-02-22T14:30:00Z",
    "twitter": "2026-02-22T14:30:00Z",
    "calendar": "2026-02-22T13:00:00Z",
    "weather": "2026-02-22T08:00:00Z",
    "builds": "2026-02-22T12:00:00Z"
  }
}

Your First Cron Job: The Night Shift

The single most important automation is nightly memory consolidation. Every night at 2 AM, your agent reviews the day, extracts key learnings, and updates the knowledge base. This is the glue that holds the entire three-layer brain together.

openclaw cron add \
  --name "Nightly Memory Consolidation" \
  --cron "0 2 * * *" \
  --tz "America/Chicago" \
  --session isolated \
  --message "Review today's daily note (memory/YYYY-MM-DD.md). \
    Extract key decisions, learnings, and status changes. \
    Update relevant knowledge base files. \
    Update tacit.md if any new preferences were discovered. \
    If a project changed status, update its file. \
    Log what you consolidated in today's daily note." \
  --model "sonnet"

The Essential Cron Setup (4 Jobs)

Here's the minimum viable cron configuration. These four jobs cover 90% of what most people need:

openclaw cron add \
  --name "Morning Brief" \
  --cron "0 8 * * *" \
  --tz "America/Chicago" \
  --session isolated \
  --message "Morning briefing: Check emails, review \
    calendar for next 24h, check project statuses, \
    overnight social mentions, weather. Read yesterday's \
    daily note for context. Compile into a concise \
    brief with action items. Keep under 20 bullets." \
  --model "sonnet" --announce \
  --channel discord --to "channel:YOUR_ID"

openclaw cron add \
  --name "Social Monitor" \
  --cron "0 */4 * * *" \
  --tz "America/Chicago" \
  --session isolated \
  --message "Check social mentions and replies since \
    last check. Log engagement metrics. Handle simple \
    interactions (likes, thank-yous). Flag anything \
    needing human response. If nothing actionable, \
    say HEARTBEAT_OK." \
  --model "sonnet" --delivery none

openclaw cron add \
  --name "Weekly Review" \
  --cron "0 9 * * 1" \
  --tz "America/Chicago" \
  --session isolated \
  --message "Weekly review: Read daily notes from the \
    past 7 days. Summarize what was accomplished, \
    what's still pending, any recurring blockers. \
    Suggest priorities for this week. Check revenue \
    metrics if available." \
  --model "sonnet" --announce \
  --channel discord --to "channel:YOUR_ID"

After one week: 7 days of consolidated learnings and 7 morning briefings.
After one month: a rich, searchable history of everything that happened.
After three months: a genuine second brain that actually works.

Common Mistakes

Advanced: Chaining Crons

As your system matures, you can create cron chains — where one job's output feeds another:

Each cron reads from shared files (the knowledge base and daily notes), so they naturally coordinate without needing to "talk" to each other. It's like a relay race where each runner picks up where the last one left off.

Your agent now works while you sleep. But before we give it real tools and real access, we need to talk about the thing that keeps you safe: security.