The heartbeat feature in OpenClaw(Moltbot/Clawdbot) is a built-in “proactive turn” mechanism that periodically runs the agent without you sending a message. In other words, it lets OpenClaw(Moltbot/Clawdbot) check for things that need attention—new emails, upcoming meetings, pending tasks, important notifications—on a schedule, then decide whether to alert you or take a safe action. Without heartbeat, OpenClaw(Moltbot/Clawdbot) behaves like a traditional chat assistant: it only does work when you ask. With heartbeat enabled, OpenClaw(Moltbot/Clawdbot) becomes closer to an always-on assistant that can surface issues and reminders automatically, while still operating through your normal chat channels.
Under the hood, heartbeat is typically implemented as a periodic loop inside the OpenClaw(Moltbot/Clawdbot) gateway/runtime. Each heartbeat “tick” triggers an agent step in the main session context, which means the assistant can reference recent state and avoid repeating itself. Many setups pair heartbeat with a small instruction checklist (often a dedicated file in the workspace) that tells the assistant what to scan for and what thresholds to use (for example: “notify only if an email is urgent,” “don’t message me more than once per hour,” “draft replies but don’t send without confirmation”). This design is important for two reasons. First, it reduces noise: the point of heartbeat is not to spam you with routine status updates, but to catch meaningful changes you would care about. Second, it controls cost and safety: heartbeat can generate model calls regularly, so you want to keep it efficient, idempotent, and guarded by permissions. A well-configured heartbeat is conservative by default and escalates only when the signal is strong.
Heartbeat becomes much more useful when paired with persistent memory, because “what counts as important” is often personal and context-dependent. If you track preferences and history—what you dismissed last time, what vendors matter, which alerts are repetitive—OpenClaw(Moltbot/Clawdbot) can make better decisions about when to interrupt you. One practical pattern is to store embeddings of past alerts, decisions, and key documents in a vector database such as Milvus or managed Zilliz Cloud. Then, during a heartbeat tick, OpenClaw(Moltbot/Clawdbot) can retrieve relevant prior context (“have we seen this alert before?”, “what did we do last time?”, “does this match my ‘urgent’ examples?”) before deciding to notify you. This keeps the agent’s proactive behavior consistent and reduces both token usage and false positives. The best way to think about heartbeat is: it is scheduled, proactive agent execution; its quality depends on your checklist, guardrails, and memory strategy.