Schedules
A schedule is work that runs on its own. You set it up once, and C5 does it again and again without you asking.
Good things to schedule #
- A summary of your week, every Friday afternoon
- A check on your inbox each morning
- A report pulled together on the first of the month
- A nightly tidy-up of a folder
If you find yourself asking for the same thing over and over, make it a schedule.
Making one #
- Go to Schedules.
- Click New Schedule.
- Choose what it does — Use Recipe, Create Task or Send Request — and write what you want done, the same way you would in Chat.
- Pick when it should run.
- Save.
That is it. C5 takes it from there.
What each action does #
- Create Task — C5 makes a new task from your instructions and works on it. It shows up on the Kanban board like any other task, at Medium priority, the same as a task you create yourself, so a schedule never jumps ahead of your own work. Change the card's priority if you want it sooner.
- Use Recipe — the same, but the task follows a recipe you pick.
- Send Request — C5 sends your request to the model and delivers its answer as a notification.
When you choose Use Recipe, a recipe picker appears beside the three buttons. It says No Recipe until you pick one, and Save Schedule stays greyed out until you do. If C5 has not made any recipes yet, no picker appears — use Create Task instead.
Each task a Use Recipe schedule makes uses the newest version of its recipe at the time that task starts. If the recipe has been deleted, the task is planned from scratch.
If you edit a Use Recipe schedule whose recipe has since been deleted, the picker says No Recipe, but Save Schedule still works. Its tasks go on being planned from scratch until you pick another recipe.
Up to v341, a Use Recipe schedule made nothing when it ran. A repeating one made in one of those versions needs nothing from you: it makes its first task at its next run (once you resume it, if it is paused). A one-off that one of those versions marked Completed made nothing either. Editing it will not make it run, so if you still want it, make a new one with New Schedule.
Running recipes after each task #
A Create Task or Use Recipe schedule can line up follow-up recipes, just like a task can. Each task the schedule makes gets the list. Once that task's work is approved, its recipes start, using what the task just made.
- In Add Schedule or Edit Schedule, find Post-Completion Recipes, below the task options. It is the same editor you use on a task. While the list is empty, the note at the bottom of the editor follows your Serial / Parallel choice: it says in serial order or in parallel order.
- Open Select a recipe to trigger after completion..., pick a recipe and click Add Recipe. Do this for each recipe you want, then drag them into order. If C5 has not made any recipes yet, the list says No proven recipes available yet and there is nothing to add.
- Choose Serial to run them one after another, or Parallel to start them all at once. Once recipes are on the list, the line under them starts with Serial order: or Parallel fan-out: to match.
- Click Save Schedule.
Auto Review decides when they start. It is on for a new schedule. With it on, the badge at the top of the editor says Triggers after Automated deliverables approval, and the recipes start as soon as C5's review passes the task. Turn it off and the badge says HITL in place of Automated: the recipes wait until you approve the task. See Human review (HITL).
Each follow-up task gets the same review setting as the scheduled task. With Auto Review off, every follow-up task waits for your approval too, and in Serial the next recipe does not start until you approve the one before it. A schedule with three serial recipes then needs four approvals each time it runs, so turn on Auto Review for work you want to run unattended.
Each follow-up task shows up on the board as its own task, named [Recipe: recipe name] scheduled task's title. See Watching the chain and jumping between tasks to follow it, and If a chain stops to get it going again.
Options and Recipe Bundles work the same as on a task. See Running recipes after a task finishes.
Changing the list changes the tasks the schedule makes from then on. A task it already made keeps its own list, which you can still change on that task until it is Done.
Every follow-up recipe is one more task each time the schedule runs, so it adds to what the schedule costs.
Send Request schedules cannot have follow-up recipes. A request just sends back an answer, with nothing to approve. If you switch a schedule to Send Request, its follow-up recipes are dropped when you save.
Picking when it runs #
You can choose simple timing in plain words — every day, every Monday, every hour, the first of each month.
If you need something more exact, you can write a cron pattern. Cron is a short way of describing a repeating time. You do not need to learn it unless you want to.
| You want | Cron pattern |
|---|---|
| Every day at 9 in the morning | 0 9 * * * |
| Every Monday at 8 | 0 8 * * 1 |
| Every hour | 0 * * * * |
| First day of the month | 0 9 1 * * |
C5 shows you the next few run times before you save, so you can check you got it right.
Watching them run #
Each schedule has a history. You can see:
- When it last ran
- Whether it worked or failed
- How long it took
- What it produced
Click any run to open it like a normal task and see every step.
Pausing and stopping #
You can pause a schedule and turn it back on later. Pausing keeps the setup and the history — it just stops running for now.
Deleting removes the schedule. Work it already finished stays in your Library.
If a run fails #
A failed run does not stop the schedule: it runs again at its next normal time. Open the failed run to see what went wrong. Common causes are an expired API key or a spending limit that was already used up.
A Send Request schedule tries again sooner. When its request to the model fails, or
comes back empty, C5 tries it again about a minute later, up to 5 attempts in all. If the
fifth fails too, it gives up on that occurrence, and the schedule's recipients get
Scheduled Request Failed:
Each occurrence starts one run. A scheduled task is started once for each time it is due; versions before v327 could start two or three runs of the same occurrence.
Tip Set up an alert so you hear about failures right away instead of finding out later. See Settings.
ROI Trackers & System Audits #
At the top of the Schedules page, C5 includes built-in ROI Trackers. These run automated audits of your system's productivity, agent performance, and cost savings over regular intervals.
Cadences #
| Tracker | When it runs | What it measures |
|---|---|---|
| Weekly | Every Sunday at midnight | 7-day activity compared to the prior week. |
| Monthly | 1st of every month at midnight | Monthly deliverables, success rates, and cost deltas. |
| Quarterly | 1st of Jan, Apr, Jul, Oct | 3-month efficiency trends and total output. |
| Yearly | January 1st at midnight | Annual executive summary of total tasks and local token savings. |
How to use them #
- Pause or Resume — Click Pause on any tracker you do not need right now. Pausing keeps your historical numbers intact.
- Run & Notify — Click Run & Notify to run an audit immediately and send an executive summary to your in-app notifications.
- View Output — Click View Notification Output on any card to read the latest audit payload directly in the panel.
Each report shows your completed tasks, deliverables, agent success rates, tool usage, and an estimate of money saved by running models locally for free.
Things to watch out for #
- Costs add up. A schedule that runs every hour runs 720 times a month. Set a budget in Ops before you build a busy schedule.
- Your computer has to be on. C5 runs on your machine. If it is asleep, the schedule waits. C5 will catch up when it wakes.
- Start slow. Run something daily for a week before you move it to hourly.
Dictating a schedule #
The title, description and instructions boxes each have a small microphone. Click it, say what you want the schedule to do, and click again to stop — the words land in the box so you can read them over before saving.
The date and time controls have no microphone. Picking a date is faster with the calendar than by describing it, and a misheard date is the kind of mistake you only notice after a run happens at the wrong hour.
See Voice for turning it on.
Where to go next #
- Budgets & costs — spending limits and cost tracking.
- Analytics — charts showing agent quality over time.
- Recipes — save repeating jobs in Labs.
- How a scheduled AI agent fits the rest of your week — the plain-words guide on growther.si: what to schedule, and the trade-offs against cron.
