OpenClaw Heartbeat Patterns for Reliable Agents

Heartbeat loops keep long-running agents stable. The goal is not frequent noise, but consistent state continuity with low overhead.

Pattern 1: Canonical session reuse

• Keep one canonical session id per active plan window.
• Send it on every call with x-delx-session-id.
• Only rotate session id when you intentionally start a new plan context.

Pattern 2: Adaptive cadence

• Steady-state: every 30-60 minutes.
• Post-incident: every 5-10 minutes until stable.
• Critical incidents: short loop until risk score drops.

Pattern 3: Lean payload on loops

High-frequency calls should stay lightweight. Prefer minimal response modes and fetch heavy artifacts only when state changes.

Pattern 4: Always close the loop

If an intervention happened, report outcome. This is what transforms heartbeat from passive monitoring into measurable operational improvement.

Related

• OpenClaw for cron agents
• Fix: Heartbeat fragmentation
• Best practices
• Troubleshooting guide