> ## Documentation Index
> Fetch the complete documentation index at: https://help.teable.ai/llms.txt
> Use this file to discover all available pages before exploring further.
# Routine
> Let Cuppy run a prompt on a schedule and review the result of every run.
Available on all Cloud plans; Self-Hosted requires Business or higher.
A routine hands a prompt to Cuppy and repeats it on a schedule, which suits work that nobody needs to trigger by hand: a daily report, a periodic cleanup of expired records, a weekly roll-up. Every run is a full AI chat, so a routine can do everything AI chat can, including reading and writing tables, calling skills, and producing files.
A routine is a resource inside a base and sits in the left sidebar next to tables, apps, and automations.
An automation fires on an event such as a record change, a form submission, or a webhook, and runs the steps you configured in advance. A routine fires only on time, and runs a prompt the agent interprets itself. Use an automation when you need fixed steps and predictable output; use a routine when you want the agent to decide what to do with the current data each time.
## Create a Routine
In the left sidebar, click **+** and choose **New routine**.
Describe what each run should do in **Prompt**. The prompt is the entire instruction for a run, so state where the data comes from, how to process it, and where the result goes. For example, "Summarize records added to the Tasks table yesterday, group them by owner, and write the summary to the Daily Report table."
Choose a frequency under **Schedule**, and set **Starting** and **Ending (optional)** as needed.
Click **Activate**. The configuration must be saved first, and the schedule must have a future occurrence.
## Settings
Besides the prompt and the schedule, the form carries these settings:
| Setting | Description |
| ------------------------------------ | ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| **Model** | The model and effort level this routine runs on. Leave **Default model** to use the space's default chat model |
| **Max run time (minutes)** | A run past this limit is stopped and marked failed. 5–120 minutes; default 30 |
| **Chat** | **New chat for every run** keeps runs independent of each other; **Continue the previous run's chat** carries the context of earlier runs forward, which suits work that needs to refer to the last result |
| **Starting** / **Ending (optional)** | The bounds of the schedule. Without an end time the routine keeps running indefinitely |
The model and effort level are recorded when you save, so run history shows what each run actually used.
### Schedule
Pick a frequency from the presets: **Hourly** at a given minute, **Daily** and **Weekdays** at a given time, **Weekly** on a weekday and time, **Monthly** on a day and time.
For anything more specific, choose **Custom (RRULE)** and write an RFC 5545 rule, such as `FREQ=DAILY;BYHOUR=9;BYMINUTE=0`. A custom rule has these limits:
* The frequency must be `HOURLY`, `DAILY`, `WEEKLY`, `MONTHLY`, or `YEARLY`, and two runs must be at least 1 hour apart.
* You can use `INTERVAL`, `COUNT`, `BYDAY`, `BYMONTHDAY`, `BYMONTH`, plus one `BYMINUTE` and one `BYHOUR`. `COUNT` tops out at 1000, and `COUNT` or `INTERVAL` requires a **Starting** time.
* The timezone and the bounds come from the form, so `TZID`, `DTSTART`, `UNTIL`, and `BYSECOND` are rejected.
For a schedule that runs only once, use a custom rule with `COUNT=1`.
A schedule is evaluated in the timezone of whoever created the routine and does not follow the viewer. The **Next run** time shown in the interface is already converted to your local time.
## Drafts, Updates, and Running Now
A new routine is a draft and does not run on schedule until you activate it. When you edit an active routine, the change is also saved as a draft while the live version keeps running on the old configuration: click **Update** to apply it, or **Discard changes** to drop it.
**Run now** executes once without waiting for the schedule, which is useful for checking a prompt. A routine cannot be run by hand again until its previous run has finished.
Turn the switch off to deactivate. The schedule stops firing, and existing run history is kept.
## Run History
Open the routine and switch to **Run history**. The run list can be filtered by status and time range, which helps locate a particular failure; selecting a run shows its planned, started, and finished times, its duration, and the full conversation of that run.
Runs are reported as **Queued**, **Running**, **Completed**, **Failed**, or **Canceled**. **Canceled** appears when someone interrupted that run, or when the routine or its base has been deleted.
A **Failed** run states its reason, and each reason calls for a different response:
| Message | Meaning | What to do |
| --------------------------------------------------- | ----------------------------------------------------------------- | ----------------------------------------------------------------------------------------------------------------------------- |
| **Run failed** | The run started but hit an error | Open the conversation for that run and judge from the failure point whether the prompt or the data is at fault |
| **Run timed out** | The run passed **Max run time** and was stopped | Raise **Max run time**, or split the prompt into smaller units of work |
| **Skipped: not enough credits** | Credits ran out, so the run never started | Top up the space's credits |
| **Skipped: the previous run was still in progress** | The previous run had not finished, so this occurrence was skipped | Lower the frequency, or reduce how much one run processes |
| **Skipped: timed out waiting in the queue** | The run waited too long in the queue and was skipped | Occasional occurrences need no action; a recurring one means too much is scheduled at the same time, so stagger the schedules |
The conversation of a run is read-only. Members who can edit the routine may continue it at the end to investigate how a particular run proceeded.
Run history requires permission to edit the routine. A base's owner and creator can create, edit, and delete routines; other collaborators have read access.
## Failure Alerts and Automatic Deactivation
Teable sends a notification when a run fails or credits run out. It goes to the member who last updated the routine, named under **Notifications will be sent to** at the top of run history. Failure notifications are not sent every single time, so that a long run of failures does not flood the recipient.
After 5 consecutive failures the routine is deactivated automatically and a separate notification goes out. Turn the switch back on once the problem is fixed; the count resets after the next successful run.
Not every unsuccessful run counts toward that total: runs skipped because the previous run was still in progress or because they timed out in the queue, along with **Canceled** runs, are not failures and produce no notification. A credit shortfall does count, so leaving credits unfunded eventually deactivates the routine.
## FAQ
Yes. Every run is an AI chat and is billed against the space's credits by actual usage, listed under type **Routine** in **Credit usage summary** on the billing page. When credits run out the run is skipped and a notification is sent, and repeated skips eventually deactivate the routine.
No. The change is saved as a draft and only reaches the live version when you click **Update**. A run already under way keeps the configuration it started with.
Yes. After a template install its routines are switched on the way its workflows are. A schedule with no future occurrence stays a draft.
When the context approaches its limit, Teable compacts the conversation, so runs are not interrupted by it. Choose **New chat for every run** when each run should start from a clean context.