OpenClaw Heartbeat System: How Your AI Agent Stays Proactive 24/7

What the Heartbeat Actually Is

The OpenClaw heartbeat is a scheduled pulse — a recurring signal sent to your local AI agent at a configurable interval (typically every 15–30 minutes). When the pulse arrives, the agent wakes up, reads a file called HEARTBEAT.md from your workspace, executes any tasks defined there, and then either acts on something or quietly acknowledges there's nothing to do.

Think of it like a shift worker doing a walkthrough. Every half hour, they check the floor, scan for anything out of place, take care of small tasks, and log what they did. They don't need a manager to tell them to do it — it's built into the routine.

That's exactly what the heartbeat creates: a rhythm. Your agent isn't waiting for you to ask. It's already checking.

How to Configure Your HEARTBEAT.md

The heartbeat task list lives in a single Markdown file at the root of your OpenClaw workspace. It's deliberately simple — no YAML, no complex schemas, just a plain checklist your agent reads and acts on each cycle.

Here's a real-world example of what a small business owner might put in their HEARTBEAT.md:

# HEARTBEAT.md

## Every Cycle
- Check for unread emails from new leads — flag anything that needs a same-day response
- Scan the shared calendar for events in the next 4 hours
- Check if any invoices are past due (reference /workspace/data/invoices.json)

## Morning Only (08:00–10:00)
- Pull today's top 3 priorities from /workspace/daily-plan.md
- Send me a Telegram summary of overnight activity

## Weekly (Mondays)
- Generate a client activity digest and save to /workspace/reports/

Your agent reads that file on every heartbeat, determines which tasks apply to the current time window, executes them, and reports back only if there's something worth your attention. If everything is quiet, it replies HEARTBEAT_OK and stays out of your way.

The "Smart Filter" — When to Alert vs. Stay Silent

One of the biggest problems with notification-heavy systems is alert fatigue. If your agent pings you every 30 minutes with "nothing new," you'll start ignoring it within a week. OpenClaw is designed to avoid this. The agent uses its own judgment (guided by its SOUL.md persona) to decide whether a finding is worth surfacing.

• A new inbound lead email at 9 AM? → Alert immediately.
• A calendar event in 90 minutes? → Alert with context and prep notes.
• No new emails, no events, nothing overdue? → HEARTBEAT_OK, silent acknowledgment.
• It's 2 AM and nothing is urgent? → Stays quiet, no notification sent.

This behavior isn't hardcoded — it's guided by the agent's SOUL.md, which defines its judgment, priorities, and communication style. You shape the agent's personality once, and it applies that judgment across every heartbeat cycle.

Under the Hood: How the Heartbeat Runs on Mac Mini M4

The reason this works so reliably is that it's running locally on your Mac Mini M4, not in a cloud container that might spin down after inactivity. The M4 chip idles at around 6 watts, which means leaving it on 24/7 costs roughly $6–8/month in electricity. Your agent is always running. Always watching. Always ready.

When you install OpenClaw, the heartbeat is set up as a background daemon via a simple config. Here's how you'd set the interval using the OpenClaw CLI:

# Set heartbeat interval to every 20 minutes
openclaw heartbeat set --interval 20m

# Check current heartbeat status
openclaw heartbeat status

# View recent heartbeat logs
openclaw heartbeat logs --last 10

The daemon runs independently of any active chat session. Even if your phone is off and you haven't opened Telegram in 12 hours, the agent is still running its heartbeat cycles, processing tasks, and building up logs of what it found.

What the Agent Can Actually Do During a Heartbeat

The heartbeat isn't just a check-in — it's a full execution environment. During a heartbeat cycle, your agent can:

• Read and write files in the workspace
• Query local databases or JSON files
• Send Telegram or WhatsApp messages (with your permission)
• Run shell commands (within defined safety limits)
• Access the web for research or data retrieval
• Update memory files to capture what it learned
• Trigger other skills — weather, email, calendar, and more

The combination of a persistent local runtime (Mac Mini M4) and a structured heartbeat loop is what separates a real autonomous agent from a glorified chatbot. One waits for you. The other works while you sleep.

Building a Heartbeat-Driven Business Workflow

Let's make this concrete. Here's an example heartbeat workflow designed for a service business:

The Morning Brief (8 AM)

Every morning at 8, the agent compiles a summary of overnight emails, flags any new inquiries, checks the day's calendar, and sends a clean Telegram briefing before the owner even opens their laptop. No manual checking. No missed leads. Just a brief waiting for them when they wake up.

The Lead Watcher (Every 15 Minutes)

Every 15 minutes during business hours, the agent checks a connected inbox for new inquiry emails. When one arrives, it extracts the key info, qualifies the lead against a set of criteria, drafts a proposed response, and sends an alert. The owner reviews it in 30 seconds and hits approve — or lets the agent send it automatically if they've opted into that.

The Invoice Reminder (Weekly, Fridays)

Every Friday afternoon, the agent scans a local invoice tracker, identifies anything overdue, drafts polite follow-up messages, and either sends them or queues them for review. The kind of task that's easy to skip manually but costs real money when it slips.

Heartbeat vs. Cron Jobs: Why the AI Layer Matters

You might be thinking: "Can't I just use cron for scheduled tasks?" Technically, yes. But there's a crucial difference. A cron job runs a script — it executes a fixed set of instructions and exits. It doesn't think. It doesn't adjust based on context. It doesn't decide whether something is worth your attention. It just fires.

The heartbeat wraps your scheduled execution in an AI reasoning layer. Instead of running check-email.sh, it runs a full agent cycle that can read the email, understand its content, decide if it's urgent based on your preferences, draft a response, and then choose whether to alert you — all within the same pulse.

That's the leap from automation to intelligence. Cron does the same thing on a schedule. The heartbeat does the right thing on a schedule.

Getting Started with Your First Heartbeat Setup

If you're already running OpenClaw on a Mac Mini M4 local agent setup, adding a heartbeat takes about five minutes. Create your HEARTBEAT.md, define two or three tasks, set your interval, and let it run overnight. Check your logs in the morning. You'll see exactly what your agent was doing while you slept.

If you're not set up yet — that's what we do at BuildAClaw. We configure the whole stack: Mac Mini M4, OpenClaw install, SOUL.md personality, HEARTBEAT.md task list, and Telegram integration. You get a working, proactive AI agent in 48 hours, tuned to your business workflows from day one.