Skip to content

Scheduled AI tasks, ready for later.

Give Ghost the work, the timing, and what a good result looks like. Run a task once or on a recurring schedule, then review its output in Scheduler.

One-time or recurring · Timezone-aware · Private beta for Mac

SchedulerIllustrative workflow · Atlas

Atlas readiness review

Brief
Check current tasks and the readiness note. Summarize completed work, blockers, and decisions needing review.
Timing
Weekdays at 08:30 · America/Los_Angeles
Result
Run status and output saved in Scheduler.
Example schedule · output is separate from notification delivery

What are scheduled AI tasks?

Scheduled AI tasks run a defined instruction at a future time or on a recurring schedule. In Ghost, each task has a standalone brief, timing, and an explicit timezone for local times. Scheduler keeps the task manageable and saves run status and output so you can inspect what actually happened.

Give a future run a complete brief.

A schedule needs more than a time. Watch a recurring brief become a manageable task with its own instructions, timezone, stored result, and run history.

Tell the future run what good work looks like.

A scheduled run does not inherit the current conversation. Its instructions need the goal, sources, constraints, expected output, and what to do when information is missing.

Scheduled task · standalone brief

Atlas readiness review

Check the current Atlas tasks and readiness note. Summarize completed work, remaining blockers, and decisions needing review. Keep the source references. If a source is unavailable, say which one.

A complete task brief, not “do the same thing tomorrow.”

Set the timing in the right timezone.

Choose a one-time run, an interval, or a recurring schedule. Local wall-clock timing needs an explicit timezone so the meaning of “morning” is not left to guesswork.

Example recurring schedule
Task
Atlas readiness review
Schedule
Weekdays · 08:30
Timezone
America/Los_Angeles
State
Enabled
RecurringUser managed

Read the result in the run history.

When the task runs, Scheduler retains the status and output. Open the relevant run to inspect what happened instead of assuming that a schedule being enabled proves the work succeeded.

Scheduler · example run

Atlas readiness review

CompletedOutput stored

Ready: owner confirmations recorded.

Open: review the rollout brief and onboarding risk.

Sources: current Focus items and readiness note. No notification implied.

Keep the automation under your control.

Edit the instructions as the work changes, inspect recent runs, or pause the schedule. A scheduled task is visible and manageable, separate from an ordinary to-do.

Scheduler · task management
Instructions
Editable
Recent runs
Status and output available
Schedule
Paused in this example
Paused
Use To-dos for the work list and Heartbeat for context-aware check-ins.

Illustrative schedule and output. Generic scheduled runs save results in Scheduler; they do not guarantee a desktop, chat, or push notification.

Write the brief for a fresh run.

A future run does not inherit the conversation that created it. “Do that again tomorrow” leaves important details behind. Give the task its goal, the sources to inspect, any constraints, and the output you want to review.

For a recurring project review, name the project and its relevant records. Say whether you want a short summary, a list of changes, or a draft. Include what Ghost should do when a source is missing or the evidence is inconclusive.

An instruction can reference your working notes and current tasks. It should still be understandable on its own, even if the original chat is no longer in view.

Goal
What should this run accomplish?
Sources and scope
Which notes, projects, URLs, or connected tools should it use?
Expected output
What should the result contain, and how much detail?
Missing evidence
What should it report if a source is unavailable or no relevant change is found?

Run it once. Or make it recurring.

Schedule a single future run, repeat after an interval, or use a recurring calendar pattern. An explicit timezone keeps local requests such as “every weekday morning” from depending on an unstated assumption.

A one-time review can fit a particular milestone. A recurring task can inspect a changing set of sources and produce a fresh result each time. The brief says what to do; the schedule says when to attempt it.

This is AI task automation for defined work, not automatic time-blocking of your whole calendar. Enabling a schedule is also not proof that every future run will succeed: available services, sources, and permissions still matter.

Timing choicesIllustrative workflow · Atlas

Match the schedule to the job.

Once
Review the release note tomorrow afternoon.
Interval
Check the selected source on a recurring interval.
Calendar pattern
Produce a readiness summary every Monday morning.

Give recurring work a repeatable starting point.

Use scheduled tasks for work whose question stays useful while the underlying information changes. The output can be a daily brief, a weekly project review, a research digest, or a follow-up check with named evidence.

A research task can revisit a defined topic and report what changed. A project review can read Focus items and their linked records. For a workflow that needs a website, specify the source and the permitted browser work rather than assuming unlimited access.

The same assistant connects the run to the rest of your work. Clear instructions still matter: ask for a draft when the result needs review, identify which actions are permitted, and keep consequential decisions with you.

Timing is separate from execution structure. Task orchestration coordinates a multi-step job and its dependencies; a schedule specifies when a brief should run. Neither grants permission for unrelated actions.

Daily brief
Summarize the current state of named sources.
Weekly review
Separate completed work, open blockers, and decisions needing attention.
Research digest
Recheck a defined topic and keep the source references.
Future follow-up
Run one focused check after a milestone or change.

Inspect the run, not just the schedule.

Scheduler stores the output and run status. Open the relevant run to see what Ghost produced, whether it completed, or whether something prevented the work. A task that exists and is enabled is different from a task whose last run succeeded.

Edit the instructions, timing, or timezone as the work changes. Pause a task when it is no longer useful, resume it when needed, or remove it. Scheduled tasks are visible and user-managed.

Generic scheduled runs do not guarantee a chat, desktop, or push notification. Their output lives in Scheduler. For context-aware check-ins across current work, see Heartbeat.

Instructions
A standalone brief you can update.
Run history
Saved output and status for inspection.
Task controls
Edit, pause, resume, or remove the schedule.

Product details

Scheduled tasks, at a glance.

Supported work, useful outputs, and what to check before you start.

Schedule
Run once, at an interval, or on a cron schedule with an explicit timezone. Each run needs a standalone brief with the relevant sources and expected output.
Result
Scheduler keeps the run status and output so you can inspect what happened. A future run does not inherit the full conversation that created it.
Control
Edit, pause, resume, or delete a scheduled task. Chat, desktop, and push notifications are not guaranteed by scheduling alone; delivery needs its own setup.

One assistant. Connected work.

Ghost is a personal AI assistant for Mac. Meetings, Notes, To-dos, Browser, and Memory are connected capabilities of the same product, with records you can bring into the next task.

A recurring review can inspect the to-dos, find the note that explains each blocker, and refer back to the source meeting. Browser and research work can contribute fresh evidence when the brief and permissions allow it. The result is stored in Scheduler for review.

  • Meetings

    Capture the discussion and its follow-ups.

  • Notes

    Keep an editable record with the source attached.

  • To-dos

    Track the next action, priority, and progress.

  • Browser

    Read and act in a connected Chrome session.

  • Memory

    Retrieve relevant saved history and preferences.

Questions about scheduled tasks.

Can I schedule a task through conversation?

Yes. Tell Ghost what should run, when, and what output you expect. A future task needs a complete standalone brief, including sources and constraints, rather than relying on the current chat.

What schedules are supported?

Ghost supports a one-time future run, an interval, and a recurring cron schedule. Local wall-clock requests need an explicit timezone.

Does a scheduled task remember this conversation?

No. The future run does not inherit the current chat. Put the relevant context, names, sources, constraints, and desired output into the saved instructions.

Where do I read the results?

Open the task’s run history in Scheduler. Generic scheduled runs save their output and status there; they do not guarantee delivery to chat, desktop notifications, or push notifications.

Can I pause or change a recurring task?

Yes. You can edit its instructions, schedule, timezone, or enabled state, inspect recent runs, pause or resume it, and delete it when it is no longer needed.

Is a scheduled task the same as a to-do?

No. To-dos track work and progress. Scheduled tasks attempt defined work later. Heartbeat checks current context for something worth surfacing.

Will every task work while my Mac is unavailable?

Do not assume that. A run can depend on the agent service, connected integrations, permissions, and tools on your Mac. Inspect the run history for the actual result; the existence of a schedule is not an availability guarantee.

Give the next run a clear brief.

Set the timing once, keep the task under your control, and inspect each result in Ghost. Private beta for Mac.

Request access