← Back to Blog

The Heartbeat System: How Your AI Stays Awake All Night

The Heartbeat System: How Your AI Stays Awake All Night

Most AI tools work during the day. You prompt them. They respond. You go home.

At night, the system is asleep.

A real AI employee works at night. It's awake. It's executing tasks. It's generating value while you sleep.

This requires one thing: a heartbeat.

What Is a Heartbeat?

A heartbeat is a cron job that fires every 10-15 minutes. It wakes your AI up and says: "What should you do next?"

Your AI then: 1. Reads MEMORY.md (remembers who you are) 2. Reads today's task list (knows what to work on) 3. Picks the highest-priority unblocked task 4. Executes it 5. Logs what happened

Then goes to sleep for 10 minutes until the next heartbeat.

Simple. Powerful. Non-stop.

Without a Heartbeat

Without a heartbeat, your AI is reactive: - You ask → it answers → you leave → it sleeps - Next time you ask, it starts from zero (no memory) - It never does any work on its own

This is ChatGPT.

With a Heartbeat

With a heartbeat, your AI is proactive: - Heartbeat fires → AI wakes up - Reads task list: "Write blog post, research leads, check server health" - Picks the highest-priority task - Executes it - Logs the work - Sleeps until next heartbeat

All without you asking. Repeat 144 times per day (every 10 minutes).

144 "wake-ups" × 15 minutes of work per session = 2,160 minutes = 36 hours of work per day.

Your AI is doing 36 hours of work for you every 24-hour period.

The Task List

The heartbeat works because there's a task list. You give it direction:

  • **Daily notes** (`memory/YYYY-MM-DD.md`):
  • [ ] Write 1 blog post about AI automation
  • [ ] Research 15 leads in SaaS space
  • [ ] Check Stripe for new customers
  • [ ] Monitor server health
  • [ ] Respond to customer emails (templates OK)

Your AI reads this every heartbeat. It picks the next task. It executes.

Priority Order

Your AI doesn't just pick random tasks. It uses priority:

  • **Tier 1:** Revenue tasks (processing orders, handling customers)
  • **Tier 2:** Growth tasks (writing content, researching leads)
  • **Tier 3:** Operations tasks (monitoring, maintenance)

If a customer email arrives, the AI handles it first (tier 1).

If no tier-1 tasks, it writes a blog post (tier 2).

If no tier-1 or tier-2, it checks server health (tier 3).

Always working. Always prioritizing.

Autonomy Levels

You configure how much freedom the AI has:

  • **Full auto:** "Execute everything within SOUL.md rules. Report after."
  • AI writes and publishes blog posts
  • AI sends templated customer emails
  • AI deploys code if tests pass
  • AI escalates only real emergencies
  • **Confirm big decisions:** "Execute small tasks, ask about big ones."
  • AI writes content (small)
  • AI asks before spending $100 (big)
  • AI asks before changing production systems (big)
  • **Always ask:** "Run nothing without approval."
  • Slow. But safe for ultra-sensitive operations.

The Logging

Every heartbeat, your AI logs what it did:

## 3:00 AM — Heartbeat 144

Tasks completed: - [x] Blog post: "Why AI Fails at Business" (3,200 words) - [x] Research: 15 SaaS leads (Slack, Notion, Zapier...) - [x] Monitoring: All systems green

Blockers: - Customer email from John — needs human response (sales complexity)

Next task queued: - Check competitor pricing

Time spent: 28 minutes ```

You wake up to a complete summary. You didn't do anything. The AI did 4 hours of work.

The Compounding Effect

  • **Day 1:** AI writes 1 blog post. 80% of the value is wasted (nobody reads it yet).
  • **Week 1:** AI writes 7 blog posts. 30% of the value is realized (some random traffic).
  • **Month 1:** AI wrote 30 blog posts. 60% of the value is realized (search engines are indexing, traffic flowing).
  • **Month 6:** AI has written 180 blog posts. 90% of the value is realized (Google ranks them, you get $2-5K/month organic traffic).
  • **Year 1:** AI has written 365 blog posts. 100% of the value is realized (you're #1 for 50+ keywords, $10-20K/month organic).

A heartbeat creates compound execution.

How to Set It Up

1. **Write HEARTBEAT.md** ```markdown 1. Read MEMORY.md (remember who I am) 2. Read memory/YYYY-MM-DD.md (know today's tasks) 3. Pick highest-priority task 4. Execute it 5. Update daily notes with result ```

2. **Create a cron job** ```bash openclaw cron add --interval 600000 \ --prompt "Run HEARTBEAT.md" ```

3. **Write your task list** ```markdown ## Tasks for Today - [ ] Task 1 (priority 1) - [ ] Task 2 (priority 2) ... ```

4. **Let it run** Every 10 minutes, your AI wakes up and executes.

The Reality Check

Your heartbeat only works if: - Tasks are clear (not vague) - Your AI has the right tools (file access, web search, etc.) - SOUL.md is specific (not just "be careful") - You don't interrupt it constantly

If you interrupt every heartbeat with new tasks, it never finishes anything.

Give it 24 hours. Let it execute a full day/night cycle without interruption. You'll see what's possible.

The Future of Work

Heartbeats are the infrastructure that enables actual AI employees.

Without a heartbeat, you have a chatbot.

With a heartbeat, you have an employee that works while you sleep, never gets tired, and costs $50/month.

[Get AldenAI — $49 →](/#pricing)

Get the Kit — $49 →

14-day money-back guarantee · Instant download