Scheduled Tasks

Let an agent run on a timer. Set up a recurring brief, report, or check, choose whether the result arrives by email or waits in botts.ai, and see what each run cost.

A scheduled task is a personal recurring agent run. You write the instruction once ("check our suppliers' status pages and flag anything that changed"), pick a rhythm, and the agent does it on schedule. Every result is saved to a chat conversation, and optionally emailed to you.

Scheduled tasks live under Scheduled in the left sidebar.

Schedule firesdaily · weekly · monthly
The agent runs

same knowledge and tools as chatting by hand

Result

saved to a chat, emailed or left unread

No credits? The run is skipped, not queued. It is not made up later.

Three failures in a row? The task pauses itself and emails you.

Personal, not organizational

A task belongs to the account that created it, not to the organization. The task list is yours alone: Admins and Owners get no shared view of everyone's schedules, and no one can edit, run, or delete someone else's task.

The results follow the internal-chat rules rather than the task's ownership, because a run posts into an ordinary internal-chat conversation. Under the default Private policy, Admins and Owners see only its metadata. Under Full, they can read the content, the same as any other internal chat. Each conversation snapshots the policy in force when it was created. See Internal-chat privacy and oversight.

Who can create one

Anyone who can use the chat can create a scheduled task, which includes the chat-only Member role. What a task is allowed to run is bounded by what you could do by hand:

  • A deployed agent — only agents you are allowed to chat with. An invite-only or role-restricted internal chat you have no access to is not schedulable, and the agent must be active.
  • A model directly — the same general-purpose models the regular chat offers you. If your organization has switched Regular Chat off, a plain-model task is refused too, and the model must satisfy your organization's server location setting, because a schedule runs unattended.

Creating a task needs credits, because naming it calls a model.

A task reaches exactly what the agent reaches

A scheduled run is the same conversation you would have typed by hand, so it can use the agent's knowledge bases, its connections, and web search if that agent has it. It has no special view of the platform: it cannot read the Conversations hub, other members' chats, or your usage figures. Ask for something its tools can actually fetch, or it will produce a confident summary of nothing.

Creating a task

Describing it in one sentence

The Scheduled page has a box where you can type the whole thing at once: "summarise the news every Monday at 8". botts.ai reads it and opens the wizard with the fields filled in for you to confirm.

It deliberately stops at the wizard rather than creating the task outright. Phrases like "every other Monday" or a timezone-shifted "tonight" are exactly what a model gets wrong, and a schedule nobody confirmed would run unattended and spend credits. If the sentence can't be read, the wizard just opens blank.

The wizard

1

What

Choose the agent (or the model) and write the instruction. Write it like a message to a colleague, and point it at something the agent can reach.

2

When

Pick the frequency and time. Times are shown in your own timezone, taken from your browser.

3

Where

Choose how you're told: Email or Just in botts.ai. The result is saved either way.

4

Try it

Optional. Run it once now to see what it sends and what it costs.

Frequency

OptionWhat you pick
DailyJust the time
WeeklyWhich weekdays, plus the time
MonthlyWhich days of the month (1–31), plus the time

Months that are missing a selected day run on their last day instead, so a task set to the 31st still runs in February. It runs once in such a month, however many of the selected days are missing.

A task can also carry an end date, after which it stops. A schedule whose end date is today is how you express "just once, tonight".

Delivery

ChannelBehavior
EmailEach result lands in your inbox.
Just in botts.aiThe result waits for you, marked unread.

Either way the run is written into a chat conversation belonging to the task, so the full result is always readable in Chat and the history builds up in one place.

Trying it before you turn it on

The last step offers Run it once now. It takes about a minute, costs credits like a real run, and shows you the actual output and its exact cost before you commit to a rhythm. This is the cheapest way to find out that an instruction is ambiguous.

Managing tasks

The Scheduled page lists your tasks with the task name, its schedule, whether it is working, the last result, and the cost this month.

ActionWhat it does
Run nowRuns the task immediately, outside its schedule. A task already running is not started twice.
Pause / ResumeStops and restarts the schedule without losing the task or its history.
EditChanges the instruction, the schedule, or the delivery channel.
Open resultsJumps to the conversation holding every past run.
DeleteStops the task. The conversation with its past results is kept.

Run status

StatusMeaning
DeliveredThe run finished and the result was saved (and emailed, if that's the channel).
Running…The run is in progress.
FailedThe run errored. The reason is shown in translated form.
Skipped (no credits)The organization had no credits at the scheduled moment. Nothing was charged and nothing was sent.

Automatic pausing

After three consecutive failures a task pauses itself and emails you. The list shows it as "Paused after repeated failures". This stops a broken instruction, or an integration that has gone away, from quietly burning credits every morning. Fix the cause and press Resume.

Asking an agent to schedule something

You do not have to open the Scheduled page at all. In chat you can ask the agent to set something up ("send me this summary every Monday"), and it creates the task for you. The transcript shows a Scheduled task created marker, and the task then appears on the Scheduled page like any other, where you can edit or pause it.

Cost

A scheduled run is billed exactly like the same conversation typed by hand: the model's usage in credits, plus whatever tools it uses. Two things follow from that:

  • The cost this month column is the honest number to watch. A daily task is thirty runs.
  • Runs are skipped, not queued, when the organization is out of credits. A skipped run is not made up later.

Tip

Ask for the output you actually want to read. "Summarise the top five themes in under 150 words" costs a fraction of "give me a full report" and is far more likely to be read on a Monday morning.

Last updated on September 1, 2026