Scheduled Tasks

What are scheduled tasks?

Your agent can run tasks automatically on a schedule. These aren’t reminders that ping you — each task runs with full agent capabilities. Tools, connected apps, filesystem access, everything. The agent actually does the work, whether you’re watching or not.

Some examples:

• Triage your inbox every morning before you wake up
• Compile a weekly summary of your GitHub activity
• Post a standup summary to a Slack channel every weekday at 9am
• Check a list of websites for changes daily

Creating a task

Just tell your agent what you want and when. Something like:

“Every weekday at 9am, check my Gmail for new emails, summarize anything important, and post the summary here.”

“Every Friday at 5pm, look at what I committed this week and write a summary.”

“In 30 minutes, check if the deploy finished and let me know.”

The agent figures out the schedule, writes the task instructions, and creates it. It’ll link you to the task in the UI so you can see it. You can do recurring schedules (every day at 9am, every hour, every Monday), one-time tasks (in 30 minutes, tomorrow at noon), or a combination (starting tomorrow, repeat every 6 hours).

You can also create tasks manually from the scheduled tasks panel, but asking the agent is easier.

The scheduled tasks panel

Open it from the sidebar. You’ll see all your tasks with:

• Status — active, paused, or completed
• Schedule — shown in plain language (“Every weekday at 9am”, not cron syntax)
• Last run — when it last fired and whether it succeeded. Click it to jump straight into the conversation the task ran in
• Next run — when it’ll fire next

Click a task to see its full prompt — the exact instructions the agent follows each run. You can edit the prompt directly, change the schedule, pause a task to temporarily stop it from firing, or resume and delete it.

How task runs work

Each time a task fires, it starts a fresh conversation. The agent doesn’t remember previous runs by default — so the task instructions need to be self-contained. If you ask the agent to create the task, it handles this for you.

If you want the agent to build on previous runs (daily logs, running summaries, follow-ups), you can pin a task to a conversation. Every run lands in the same conversation, so context accumulates over time. Ask the agent to set this up, or toggle it in the task details.

Tips

• Let the agent write the task prompt. It knows how to make instructions self-contained. If you create a task manually, remember the agent won’t have context from your current conversation when it runs.
• Pin tasks that build on themselves. A daily journal, a running project tracker, anything where today’s run should know about yesterday’s.
• Check the history. If a task isn’t doing what you expect, look at the conversation it created — you can see exactly what the agent did and course-correct the instructions.