OpenClaw Heartbeats: Making Your AI Agent Proactive

What Are Heartbeats?

Heartbeats are periodic check-ins where your OpenClaw agent wakes up and does useful background work — even when you haven't asked it anything.

How They Work

At configured intervals, OpenClaw sends a heartbeat prompt to your agent. The agent reads HEARTBEAT.md and decides what to do:

Something needs attention? → Send you a message
Nothing new? → Reply HEARTBEAT_OK and go back to sleep

Configuring HEARTBEAT.md

markdown
# HEARTBEAT.md

## Checks (rotate through these)
- [ ] Check email for urgent messages
- [ ] Review calendar for upcoming events (next 24h)
- [ ] Check website uptime for copypastelearn.com
- [ ] Weather forecast if rain expected

## Rules
- Only alert for genuinely important things
- Quiet hours: 23:00 - 08:00 (don't disturb)
- If nothing important, reply HEARTBEAT_OK

Heartbeats vs Cron

OpenClaw also supports cron jobs. When to use each:

Feature	Heartbeats	Cron
Timing	Approximate (every ~30min)	Exact (9:00 AM sharp)
Context	Has recent chat history	Isolated session
Batching	Multiple checks per beat	One task per job
Model	Uses main session model	Can use different model

Rule of thumb: Use heartbeats for batched periodic checks, cron for precise schedules.

Tracking State

Use memory/heartbeat-state.json to avoid redundant checks:

json
{
  "lastChecks": {
    "email": 1703275200,
    "calendar": 1703260800,
    "weather": null
  }
}

The agent checks timestamps and skips recently-completed tasks.

Practical Examples

Email Monitor

markdown
## Email Check
- Look for unread emails from VIP senders
- Summarize anything marked urgent
- Skip newsletters and marketing

Calendar Alert

markdown
## Calendar
- Alert 2 hours before any meeting
- Include video call links if available
- Mention if preparation is needed

Infrastructure Monitor

markdown
## Infra
- Ping copypastelearn.com
- Check SSL certificate expiry
- Alert if any service is down

Best Practices

Keep HEARTBEAT.md small — it's read every interval, burning tokens
Rotate checks — don't run everything every heartbeat
Respect quiet hours — nobody wants 3 AM alerts
Track state — avoid redundant API calls
Be selective — only alert on genuinely important things