Action Schedules

An action schedule is a cadence attached to an action in Treasure AI Studio, so the action runs on its own — the Monday morning report already waiting when you open Studio, the nightly data check that only needs your attention when it fails. You keep the same action you run by hand; a schedule adds a clock to it.

Each run starts a fresh chat in the action's work folder, under your account, exactly as if you had clicked Run yourself. The run happens whether or not you are signed in, and the result waits for you as a chat you can open, read, and continue.

Schedule an action once the manual version has proven itself — when you already trust what the action produces and the only thing left to remove is the act of starting it.

• Signed in to Treasure AI Studio (Getting Started)
• At least one action in the work folder (Actions) — plain skills cannot be scheduled

A scheduled run produces a new chat in the work folder, named after the action, and runs as the user who created the schedule. Treasure AI Studio has no separate "scheduled" execution environment: the run sees the same work folder files, the same instructions, and the same skills a manual run sees, and it consumes credits from the same account.

A schedule runs at the time you set, read as the local time in the schedule's own timezone. Where that zone observes daylight saving, the run follows the clock change rather than drifting an hour. A run that could not happen is not made up later — if an occurrence is skipped because the schedule was off, the next run is the next matching time, not an immediate catch-up.

In the sidebar's Recent Chats and in the folder's Chats section, a chat started by a schedule carries a clock badge on its chat icon, distinguishing it from the ⚡ badge of a chat you started with Run. Its ⋯ menu offers Go to schedule, which opens the work folder page with the originating schedule highlighted.

You can start from an action, which pre-fills the target, or from the Automations section, which asks you to pick one.

1. Open the work folder and scroll to the Actions and Skills section at the folder's root.
2. On the action's row, click Add schedule. The New schedule dialog opens with that action fixed as the target.
3. Choose a Repeat cadence: Daily, Weekdays, Weekly, Monthly, or Custom.
4. Set the Time, plus Day of week or Day of month if the cadence needs one.
5. Confirm the Timezone. It defaults to the timezone in your Studio settings; the time you entered is interpreted in this zone, not in the browser's.
6. Check Next runs — the dialog lists the next three run times, so you can confirm the cadence before saving.
7. Click Create schedule. The new schedule appears in the Automations section, switched on.

To schedule an action from the section instead, click New schedule in the Automations header and select the action from the Action list. The button is unavailable while the folder's actions are still loading, and when the folder holds no actions to schedule.

The four presets cover the common cadences and write the underlying cron expression for you. Custom exposes that expression directly, for cadences the presets cannot express — every second hour, the first weekday of each quarter, several times a day.

A custom expression has exactly five fields — minute, hour, day of month, month, day of week — where * means unconstrained. For example, 0 9 * * 1-5 runs at 09:00 Monday through Friday, and 30 */2 * * * runs at half past every second hour.

Two rules are checked as you type and again when you save:

• An expression must have exactly five fields. Six-field expressions with seconds are not accepted.
• Day of month and day of week cannot both be constrained. Constrain one and leave the other *. "The 1st or every Monday" cannot be expressed in one schedule — use two.

The ? wildcard is not supported anywhere in the expression; write * for an unconstrained field. A rule the expression breaks is reported in red under the field as you type.

An expression that fires more often than the minimum interval still previews, with a warning under the field, and is rejected when you save.

The work folder page lists your schedules for that folder in an Automations section, shown at the folder's root below Actions and Skills. The heading carries the schedule count and the New schedule button. The section is hidden entirely when the folder has neither schedules nor actions to schedule.

Each row shows the action's directory name, the cadence in plain words ("Every weekday at 9:00"), the timezone, and — on the right — the state that matters most: Next run with the time in the schedule's own timezone, or Turned off, or Turned off automatically. A second line reports the last run: its status, when it finished, and a View chat link to what it produced — or Never run on a schedule that has not fired yet.

The controls on a row are:

A row also warns you before its schedule ever fails. When the action a schedule points at has gone missing, has a broken definition, or is no longer an action, the row says so in amber under the cadence line — while the schedule is still enabled and still counting down to a run that will fail.

An action that has schedules also carries a chip on its row in Actions and Skills: the cadence itself when there is one schedule ("Weekdays 9:00"), or a count ("×2") when there are several. The chip turns amber when any of the action's schedules has been turned off automatically, so a broken automation is visible from the action row without opening the section. Clicking the chip opens the run timeline.

Run history is the timeline of one action's runs, opened from ⋯ → Run history on any schedule row or by clicking that action's automation chip. It merges every run of the action — scheduled runs across all of the action's schedules, and the manual runs you started with Run — into one list, newest first. Each row opens with an icon naming its source on hover: a blue clock for Schedule, an amber bolt for Manual. The dialog's subtitle says how many schedules it drew from.

Scheduled rows carry an outcome:

Manual runs record no outcome — open the chat to see how they went.

A failed row names the cause and offers Details, which expands the diagnostic message recorded at the time of the run. The newest failed row also offers Run now, which starts the action immediately as an ordinary manual run so you can confirm a fix without waiting for the next occurrence. The resulting chat appears in the same timeline as a manual row. Run now is unavailable while the action is missing or its definition is broken — fix the definition first.

A run that produced a chat is kept as long as that chat is, and View chat keeps working for exactly that long. A run that produced no chat — a failure, a skip, or one still pending — is deleted 60 days after the time it was due.

Once a schedule's own rows have aged out, its last-run line says history no longer retained in place of a status: the schedule still remembers when it last ran, even after the record of that run is gone. A schedule that has genuinely never fired says Never run instead, so the two are never confused.

A schedule stops on its own when the thing it points at can no longer run. Three consecutive failures with a cause you can fix turn it off. The three need not share one cause; the row then shows an amber banner naming the most recent one, what to do about it, and a Turn back on button. A run that fails because your account is unavailable turns the schedule off immediately, without waiting for a third failure.

Trouble on Treasure AI's side does not count toward this. Transient infrastructure failures are retried automatically and never reach the run history, a run that fails on an internal error leaves the counter untouched, and a single successful run resets the counter to zero.

Turn back on clears the automatic disable and gives the schedule a fresh budget of failures. It does not check whether you fixed the cause — if you have not, the schedule fails its way back to off.

Converting a scheduled action back to a plain skill produces the same outcome by a slower route: the schedules stay registered, every run fails, and each one turns itself off after three failures. Studio asks you to confirm before that happens — see Converting an Action That Has Schedules.

Action schedules in Treasure AI Studio are bounded by the following limits. The two you are most likely to meet are the per-user cap on enabled schedules and the one-hour minimum between runs.

• Actions — Create the action a schedule runs
• Work Folders — The folder a schedule and its chats live in
• Settings — Set the timezone new schedules default to
• Credit Usage — How scheduled runs consume credits