Scheduled tasks
Put a prompt on a timer — daily, weekly, monthly or once — and read the result when it lands.
Scheduled is where prompts run without you. The page describes itself plainly — "Prompts that run automatically on a schedule — each run starts a fresh chat with the result." — and that last clause is the whole model: a schedule isn't a background job with a mysterious output, it's a conversation that starts itself.
Scheduled tasks is one of the surfaces a workspace can switch off. Where it's off, Scheduled isn't in the sidebar and the URL sends you home. Whoever manages the workspace's feature availability can turn it back on.
Making one
You need permission to create content, which members and admins have by default and viewers don't. There are three doors in.
New schedule
The button in the header of Scheduled. Opens an empty form.
A starter template
An empty page offers four: "Daily digest" — "A morning summary, every weekday", "Weekly report" — "A recap every Monday", "Monthly summary" — "A wrap-up on the 1st", and "Start from scratch". The first three land you in the form with the instructions and cadence already filled in.
From a message you already sent
In a chat, the menu on one of your own messages carries "Create schedule" — "Run this prompt on a repeating schedule". It opens the same form with that message as the instructions.
The form is three steps — Prompt, Delivery, Review — and you can't skip forward past an empty required field; it will show you what's missing instead of silently refusing.
Prompt
A Name (the example given is "e.g. Weekly analytics digest") and the Instructions — "What should run on this schedule?"
The instructions box is the chat composer, so everything you can do in a chat message you can
do here: @ to pull in a project, / to call a skill by name. What you can't do is attach a
file — a schedule has no attachments, only text.
Two more controls sit under the box: the model this schedule runs on, and Tools.
Delivery
Repeats — Daily, Weekdays, Weekly, Monthly or Once — plus the day pickers each one needs, and At, the time of day.
Then Visibility, and an "Email delivery" switch.
Review
A read-back of everything, and a live preview headed "Next 3 runs" with the timezone those times are in. This is the moment to check that "Monthly on the 31st" means what you think it does.
Long instructions are fine. The limit is around thirty-two thousand characters — a few pages of rules, sources and output format — and a counter only appears once you're near it. Go over and the form says how far over, so you can trim rather than guess. If a save is refused for any other reason, the message names the field that broke.
How cadence and timezone work
The time you type is read in your browser's timezone at the moment you write the schedule, and the review step names that zone out loud ("Buenos Aires time", for instance) so there's no ambiguity. From then on the schedule keeps firing at that wall-clock time in that zone — it doesn't follow you if you travel.
Weekly can mean several days
"Every Monday and Friday" is one schedule, not two — pick as many weekdays as you like. You can't deselect the last one.
Monthly clamps to short months
Day 31 fires on the 30th, and on the 28th or 29th in February. It never skips a month.
Once needs a future date
Pick a date that's already gone and the form tells you: "That date has already passed — pick a future date." After it fires, it stops.
Due schedules are picked up in sweeps a few minutes apart, so a 09:00 run starts at or shortly after 09:00 — not on the second. Don't build anything that depends on exact-minute timing.
What a scheduled run can do
A run is a real turn of chat, with the same abilities and the same limits.
Skills
Every skill that's on for you is on for your schedules. The Tools control on the Prompt step lets you switch individual ones off for this schedule only — the summary reads "All tools" until you do.
Connectors
Only connectors you have actually connected are offered. One that's enabled but not signed in would resolve to nothing at run time, so it isn't listed.
Projects
@-mention a project in the instructions and every run loads its knowledge
and skills, exactly as a chat would.
Generation
A run can generate images and video, write documents, and build web pages — all of which land as assets.
A tool you enable in the workspace later joins your existing schedules automatically. You don't have to reopen them.
What it can't do
Nobody is there to answer it. A scheduled run can't ask you a follow-up question, can't wait for a decision, and — this is the important one — can't pause to confirm a cost before spending credits. Creating the schedule is the authorization. If you don't want a schedule generating video every morning, don't ask it to.
Spending is still bounded by what the workspace can actually pay: a generation the remaining balance can't cover is refused, and the run reports that as a failure rather than overdrawing.
Where the output lands
The Activity tab
The second tab on Scheduled. Empty, it says "When your schedules run, their results land here — open one to read it or continue the conversation." Each entry carries its status — Success, Failed or Running — and opening one shows the run's output, including anything it produced as a Document, Web page or PDF.
You can filter the feed to one schedule, and a schedule's own menu has "View activity", which deep-links to exactly that.
The run's chat
Every run starts its own conversation, titled with the schedule's name and the time it ran. "View chat" opens it, and you can carry on talking from there.
These conversations deliberately stay out of Chat history — a daily schedule would otherwise bury everything else. Activity is the way in.
Assets
Anything a run generates becomes an asset like any other, searchable and filterable alongside the rest.
Your inbox, if you asked for it
"Email delivery" — "Email each run's result to workspace members or any external address. Results still appear in the app." Recipients can be people in the workspace or any address you type; the subject line is the schedule's name, and the run card afterwards notes who it went to.
Who can see a schedule
Visibility is set on the Delivery step and defaults to Private.
Private
"Only you can see this schedule and the chats its runs produce."
Workspace
"Any workspace member can see this schedule and its run history."
The setting covers the run conversations too, not just the schedule row — whoever can read the schedule can read what it produced.
Unlike assets, schedules do have an oversight override: "Private" means private from your colleagues, not from workspace oversight. Anyone holding the permission to read anyone's content — admins, by default — can open your schedules, edit them and delete them. See roles and permissions.
Pausing, editing, deleting
Each row has a switch and a menu — "Edit", "View activity", "Delete". Pausing stops future runs and keeps everything else; the row sinks to the bottom of the list marked "Paused".
Deleting is the permanent one: "This permanently deletes the schedule and stops future runs. Past run chats are kept. This can't be undone."
When runs fail
A failed run is marked in the feed and the schedule row shows "Last run failed".
After three consecutive failures a schedule pauses itself, and says so: "Paused automatically after repeated failures. Fix the prompt or tools, then resume." A banner at the top of the list counts how many are in that state. Fix whatever broke — a disconnected connector is the usual culprit — and switch it back on.
A run that stops reporting back entirely is marked failed after a while rather than left spinning, which is why an occasional failure can appear with no obvious cause.
What ran, workspace-wide
Individual runs live in Activity on this page. The workspace-wide record of everything that happened — including credit movements — is in Settings → Governance → Activity, which needs the governance read permission to open.
What a run costs
Runs spend credits like any other work. Chat is cheap; generation is not.
Narrow what a schedule can reach
Connectors, and how connecting differs from enabling.
Teach it a workflow first
A schedule that calls a skill stays short and stays consistent.