> Documentation index: [/llms.txt](/llms.txt). Every page as one file: [/llms-full.txt](/llms-full.txt).

# How scheduled runs work

How scheduled runs work [#how-scheduled-runs-work]

A scheduled run is an ordinary agent run that starts for you at the set time, with nobody there to answer it.

Each run is a chat [#each-run-is-a-chat]

* Every run, **Run now** included, opens a new chat of yours, marked unread.
* The run has your access: the same agents, connectors and models you can use in a chat.
* The run uses the agent's published version. Changes saved only to its draft reach scheduled runs once they are published.
* When you reply in a run's chat, it behaves like any chat: approval cards and questions come back.

While nobody is watching [#while-nobody-is-watching]

| In a chat, the agent would…                                     | In a scheduled run, it…                                            |
| --------------------------------------------------------------- | ------------------------------------------------------------------ |
| Show an approval card before an action that needs your approval | Takes the action without asking.                                   |
| Ask you a question                                              | Answers the question itself and carries on.                        |
| Ask you to sign in to a connector                               | Runs without that connector and lists it under the run as skipped. |

Because nobody approves its actions, write the instruction so it says exactly what the agent may do.

Before a run starts [#before-a-run-starts]

* If you have no credits left, the run does not start and shows **Stopped by quota**.
* If your account is blocked, the run shows **Failed**.

Limits [#limits]

| Limit                   | Default                        | What happens                                                                                                                                                  |
| ----------------------- | ------------------------------ | ------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| Active tasks per person | 10, across all your workspaces | A task counts while it is not paused and has a run left. Creating or resuming another one fails until you pause or delete one.                                |
| Failed runs in a row    | 5                              | The task pauses and shows **Paused after failures**. **Failed** and **Interrupted** runs count; a run that finishes resets the count, and so does **Resume**. |
| Late start              | 15 minutes                     | A run that cannot start within 15 minutes of its time is **Missed**. Missed runs are never made up later.                                                     |

Your deployment may set different values.

Expiration [#expiration]

| Frequency   | Default expiration                    |
| ----------- | ------------------------------------- |
| **Daily**   | 7 days after you create the task      |
| **Weekly**  | 1 month after you create the task     |
| **Monthly** | 3 months after you create the task    |
| **Once**    | None: the task finishes after its run |

Change the date, or check **Never expires**, in the task's form. A task with no runs left shows **Finished**.

Times and calendars [#times-and-calendars]

* The time is the clock time in the task's timezone, and it stays the same when daylight saving time starts or ends.
* When clocks go back, a time in the repeated hour runs once.
* A **Monthly** task set to a day a month does not have, such as the 31st, skips that month.

To schedule agents with an API key, see [Schedule an agent](/docs/api/configurable-agents/scheduled-tasks) in the API documentation.