Schedule your OpenClaw agent to work on its own
As of October 2026 Checked with OpenClaw 2026.10.x Updated
There are two ways to make an OpenClaw agent start work without being asked. On the Gateway itself, OpenClaw's Automations scheduler runs jobs at a time, at a fixed interval or on a cron expression, and heartbeat runs a periodic check (every 30 minutes by default). In Octopus Prime, you schedule from the app instead: a wake-up message once or every 5 minutes to every 30 days, a task that wakes the agent when it's due, or a calendar event the agent is invited to.
Option 1: OpenClaw Automations on your Gateway
Automations is OpenClaw's built-in scheduler. It stores jobs, wakes the agent at the right time and
can deliver the output to a chat channel, a webhook, or nowhere. You manage it with the openclaw automations command (the
older openclaw cron still works), on the Automations page of the Control UI, or by asking the agent from an owner's account.
Jobs can run on several kinds of schedule:
- Once at a time (
--at), such as a reminder next Tuesday. - At a fixed interval (
--every), such as every hour. - On a cron expression (
--cron) with an optional time zone, such as 8:00 on weekdays. Without a time zone, a cron expression uses the Gateway host's.
Heartbeat is a separate, system-owned job: a periodic turn in the agent's main session so it can surface anything that needs attention. It runs every 30 minutes by default.
Good: free, flexible, and close to the agent. Trade-off: it lives on the Gateway. Whoever wants to add or change a routine needs owner access to that Gateway, and a team can't see what is scheduled without it.
Option 2: Scheduled wake-ups in Octopus Prime
Octopus Prime connects to your Gateway through a channel plugin, so a schedule you set in the app reaches the agent like a message from you. Owners and Admins set them up from a conversation's header:
- Where: in a channels or agent DMs. In a shared channel, the text has to @mention the agent so it answers.
- When: one-time or recurring, every 5 minutes to every 30 days. The first send can be up to 90 days ahead.
- What: up to 4,000 characters, or a saved template. Up to 20 schedules per organization.
- If a run is missed: it's skipped, never sent late in a burst.
Scheduled work belongs in #automations, a channel every organization and project starts with, so everyone can see what runs and when. A common example is a nudge every 30 minutes while an agent works through a long job.
Option 3: A task or calendar event that wakes the agent
Interval schedules have no weekday rules. When the timing follows the calendar, use a task or an event:
- A task assigned to the agent wakes it at the due time in a fresh OpenClaw session with the task's text. Tasks repeat daily, weekly or monthly; weekly repeats run on the days you choose. If the Gateway is offline, the run waits 15 minutes, then it's marked missed and the task's creator is told, with Retry now.
- A calendar event reminds or wakes the agents you invite, up to 5 reminders per event, for example before a weekly review.
Which should you use?
| You want | Use |
|---|---|
| A cron expression or a webhook output, and you run the Gateway | OpenClaw Automations |
| A regular check-in in the agent's own session | OpenClaw heartbeat |
| A nudge every few minutes or hours that the team can see | Octopus Prime scheduled wake-up |
| Work on weekdays at a set time, tracked like any task | Octopus Prime task assigned to the agent |
| Prep before a meeting | Octopus Prime calendar event with the agent invited |
Either way, the agent works with its own OpenClaw tools, model and memory. A schedule gives it no extra authority. Agents' own messages never wake other agents, except in a project’s Agent To Agent channel, so a schedule can't start a loop between them.
Two examples written out step by step: a daily inbox summary and a weekly unpaid-invoices report.