How do I schedule Agent to work automatically?

Some work repeats: summarize yesterday's email each morning, check for overdue tickets before the end of each workday, or draft a weekly report on Monday. You should not need to open a conversation and remind Agent every time.

Scheduled tasks Scheduled tasks are built for this. Define what to do, who will do it, and when. Agent then runs on schedule and leaves the result for you. Think of it asa standing work order for Agent.

Where to find it

Choose Automationin the left rail beside New task, Skill library, and Import data. It opens a full-screen Automations page subtitled “Use scheduled tasks to let Agent run automatically.”

The task list includes task, status, type, instruction summary, and next or latest run. Above it are search, a status filter (All / Active / Paused / Ended), and a workspace scope selector. On the right is New button on the Plan card.

Two additional shortcuts:

• The conversation sidebar has a Scheduled tasks section for the current workspace, including expandable run history
• Cmd + K accepts commands such as tracker new "Daily report" --schedule daily --at 09:00and opens the New dialog with those values filled in

Two ways to create a task

Let Agent create it (recommended)

At the top of the New dialog, switch to Create with Agent and describe the task in plain language:

At 9:00 every morning, find GitHub PRs untouched for more than 24 hours, make a list, and send it to Inbox

Click Send to conversation. Agent in the conversation on the right creates and enables the task, then sends you a link to it.

You can also say “Every morning at 9:00, please…” in any conversation. Agent will create the schedule when it understands the request.

Fill in the form yourself

Switch to Manual setuprequires only a few fields:

• Task name — shown in the list so the task is easy to identify
• Instructions — sent to Agent when the schedule fires. This is the core of the task, so be specific
• Schedule — when it runs, explained below
• Workspace and Agent — two selectors beside Schedule: a folder for workspace and a robot for Agent. New tasks default to the current workspace and first Agent

More options contains List note and What to do after a missed run, described separately below.

Note: there isdoes not contain Skill selector here. By default, Agent decides which Skill to use when the task runs; you only need to describe the outcome clearly.

Available schedules

There are seven:

Level	When it runs	What to enter
Manual	Never runs automatically; starts only when you click Run now	—
Hourly	At the start of every hour	—
Daily	At a fixed time every day (default)	Time
Weekdays	At a fixed time Monday through Friday	Time
Weekly	At a fixed time every Monday	Time
Interval	Every N minutes	Minutes
One time	Run once at a specific date and time, then end	Date & time

Times use this computer's current time zone ; no additional setting is required.

Current limitations:

• Weekly is fixed to Monday; other weekdays cannot be selected
• There is no Monthly option
• and no custom cron expression

For a schedule such as every Friday, use a workaround: choose Daily and instruct Agent to check whether today is Friday and skip otherwise, or use Interval with 10,080 minutes to approximate seven days. Neither workaround is ideal.

New tasks are enabled immediately

After successful creation, the task is enabled automatically, and the app says “Created and enabled.” No extra switch is required.

One exception: if the message says only “Created” plus “Task saved, but enabling failed. Click Enable on the detail page to retry,” the task remains in drafts. Draft tasks never trigger. Open details and click Enable .

One more easy-to-miss detail:a Manual task does not run automatically even when Active; click Run now.

After enabling, the detail page summarizes the schedule, such as “Runs automatically every day at 09:00.” Intervals under 30 minutes trigger an extra warning because they are often accidental.

Where to see results

At the scheduled time, Agent starts a new conversation. There are three ways to inspect it:

1. Run history on the task detail page. Each run shows status, start time, duration, and tokens used. Successful runs expand to Run result; failed runs expand to Failure details, written in plain language rather than raw stack traces.

2. Open the run's conversation. Every history row is clickable and says “Open this run's conversation.” It opens the complete Agent session: what it inspected, which tools it called, and why. TabTin shows the work, not only the result.

3. Notifications. A desktop notification says “Scheduled task completed” or “Scheduled task failed.” Click it to open the result.

Run statuses:

Status	Meaning
Queued	Waiting in the queue
Executing	Agent is working
Succeeded	Completed
Failed	Did not complete; expand for details
Cancelled	Stopped manually
Waiting for device	The device is offline, so the run is suspended (see next section)

To stop a running instance, click Cancel this run. This stops only the current run and leaves future schedules intact. To stop the whole task, use Pause.

Requirement: an available computer

This is the most important concept:Scheduling happens in the cloud, but execution is dispatched to the bound device. A scheduled task is not a server-side script. It lets your Agent work on your computer.

Therefore:

• A task must have a Workspace and an executing Agent. A workspace maps to a directory on a device: which computer and folder will host the work
• If the Agent has no bound device, the task fails with “No available device is bound; unattended task cannot run”
• If the device is only temporarily offlinebecause it is off, asleep, signed out, or disconnected, the run does not fail immediately. It becomes Waiting for deviceand sends a notification. Execution resumes automatically when the device returns
• A suspended run waits up to 6 hours. After that, the run fails, notifies you, and recommends an always-on device

Recommendation:Bind daily tasks to a computer that stays on. A laptop closed overnight will probably miss a 9:00 AM task. The run waits until the device wakes or gives up after six hours.

What happens when the computer misses the schedule

If execution cannot start because the computer is off, the app is closed, or a service is restarting, delays under ten minutes are treated as normal and run late. Longer delays are handled by the selected missed-run policy under More options :

• Run once after recovery(default) — run automatically after recovery. Best for daily reports or summaries where skipping loses data
• Skip missed run — do not run this occurrence; wait for the next cycle. Best for reminders that are no longer useful after their time

With Skip, TabTin notifies you and the detail page says “The latest missed run was skipped according to settings; N skipped in total.”

One-time and Manual tasks do not have this setting.

What to do after a failure

First open Run historyand expand Failure details.

The system handles some failures automatically:

• Transient failuressuch as model rate limits, mid-run disconnects, or Agent timeouts are retried up to twice at two-minute intervals
• After three consecutive failures, the task is paused automatically. The detail page says “Several recent runs failed. Review the history below.” After fixing the issue, click Restore to continue
• A run lasts at most 30 minutes. A watchdog recovers genuinely stuck tasks, so status may take roughly half an hour to change to Failed. Do not immediately recreate a task that remains Running

Common causes:

• The device is offline or the bound device changedafter a reinstall or device replacement → edit the task and select an available workspace
• The executing Agent's model is misconfiguredor no longer available → check the Agent's model under Model settings
• Instructions are too vagueand Agent does not know where to deliver output or what completion means → state what to produce and where to send it

Windows currently has some known execution reliability issues. If the same task repeatedly fails there, test it once on macOS to distinguish a task problem from a platform problem.

Three ready-to-use examples

Summarize yesterday every morning

Schedule: Daily, time: 08:30. Instructions:

Summarize unread emails I received yesterday under “Needs my reply” and “For awareness only.” For each, state in one sentence who sent it, what they need, and the deadline. Send the result to Inbox. If there were no unread emails yesterday, reply “No new email yesterday.”

Missed-run policy: Run once after recovery. A late summary is better than none.

Run an end-of-day check on weekdays

Schedule: Weekdays, time: 18:00. Instructions:

Find PRs in the project repository that have had no activity for more than 24 hours and P0 issues that remain open. List each with its owner and blocked step. If nothing is overdue, reply “No backlog today.”

The explicit “No backlog” response matters; otherwise you cannot tell whether nothing was wrong or the task never ran.

Draft a weekly report every Monday

Schedule: Weekly(Weekly currently means Monday), time: 10:00. Instructions:

Summarize last week's document changes, completed scheduled tasks, and key conversation findings in this workspace. Draft a weekly report with “Completed / In progress / Next week” sections, save it as a TabDoc, and include last week's date range in the title.

Let Agent draft, then edit and send it yourself. This is much faster than starting from a blank page.

Tips for writing instructions

• State where the output should gosuch as Inbox, a document, or a table. Otherwise, the result remains only in the run conversation
• Define an empty-result response. Ask for a message even when there is nothing to report, so you can confirm the task actually ran
• Prefer a lower frequency. A task that runs every minute looks good in a demo but wastes money and floods real workflows
• Run it manually first. Test the instructions in a normal conversation, confirm Agent's interpretation and output, then schedule them

What to know about permissions

Scheduled tasks run unattended . Nobody is present to approve actions at run time, so they always use the highest permission level . This cannot be changed during creation. The locked note says: “Yolo · Highest · Unattended automatic execution · Cannot be changed.”

Take this seriously:Before scheduling work that changes real data or sends real messages, run it manually in a normal conversation. Schedule it only after confirming Agent behaves as expected.

FAQ

The task was created, but nothing happened on schedule

Check in this order:

1. Is the schedule Manual? Manual tasks start only when you click Run now
2. Is the status drafts? Automatic enabling failed during creation; click Enable
3. Is the status Paused? Repeated failures may have paused it; inspect Run history, fix the cause, then click Restore
4. Does Run history show Waiting for device? The bound device was offline

What is the difference between Pause and Cancel this run?

Pause stops the entire task and all future triggers until you click Restore.

Cancel this run stops only the current run. The schedule remains active and runs next cycle.

The task paused itself

Automatic pause happens after three consecutive failures or too many triggers in a short period, which activates storm protection.

Use Run history to find the cause, fix it, then click Restore. The detail page does not yet identify which cause triggered the pause; we are improving this.

Can I schedule the first of every month or every Friday?

Not currently. Only the seven schedules above are available, and Weekly is fixed to Monday. See Available schedules for workarounds.

Does deleting a task remove its history?

No. Deletion archives the task: it stops triggering and disappears from the list, but the full run history remains for reference. The dialog also explains this.

Where does a completed one-time task go?

Its status becomes Endedand it no longer triggers. To run it again, click Run nowor create another task.

Will tasks run when the TabTin app is closed?

The cloud schedule continues, but execution must return to your device. If the app is closed or the computer is off, the run waits up to six hours for the device rather than failing immediately. If the computer stays off, the task effectively does not run.

Can a table change trigger a task automatically?

The backend supports this, but the New dialog does not expose it yet. Webhook triggers and upstream-task completion triggers are also not yet exposed. Contact us if you need one; we may be able to configure it manually.

I cannot find Draft tasks in the list

The status filter intentionally has no Draft option. Tasks created through the dialog normally enable automatically; drafts appear only after an enable failure or command-line creation. Choose All statuses to see them.

Still stuck?

Take screenshots of the task detail page and failed run history, and send them to contact@larchiveai.com. We will help you diagnose it. Beta users can also ask their contact in the group, which is usually faster.