Scheduling and reminders

Agent schedules

When you build an agent, Rookery asks how often it should run and you answer in words:

• “every weekday at eight”
• “every morning at seven, and again at ten”
• “the first of the month”
• “every twenty minutes during work hours”
• “only when I ask”

“Only when I ask” is a real answer. Plenty of agents are better run on demand.

The schedule appears on the agent’s page in plain language, and you can change it there. If Rookery cannot express your schedule exactly, it shows you the precise form rather than a friendly approximation — because you would have no way to tell a wrong friendly answer from a right one.

One limit worth knowing: a schedule cannot combine a day of the month with a day of the week. “The first Monday of the month” is not expressible — asking for both gives you every day in the first week and every Monday, not the one day where they meet. Pick one or the other: “the first of the month”, or “every Monday”.

When they run

The scheduler checks every minute and fires what is due. Runs happen whether or not you are logged in, which is the whole point of putting Rookery on a machine that stays on.

If the machine was off

Nothing is lost. On starting up, Rookery immediately runs everything that came due while it was off and delivers every reminder that should already have arrived — so a laptop closed on Friday and opened on Monday catches up within seconds. A reminder more than two hours late says so, arriving as a delayed reminder rather than pretending it is on time.

Missed runs collapse into one. An hourly agent that was off for three days runs once when you open the laptop, not seventy-two times. You get the current answer rather than a backlog of stale ones.

If the machine slept or shut down in the middle of a run, that run is retried once on the next start — exactly once, so a run that keeps failing can never become a loop.

Catching up is paced: a few agents run at a time and the rest queue. Opening your laptop to find every overdue agent starting at once is worse than waiting another minute for them.

Pausing an agent is still absolute. A paused agent is not caught up, not retried, and not run.

Caution

On a Linux server, a user service stops when you log out unless lingering is enabled:

sudo loginctl enable-linger "$USER"

Without it, scheduled runs stop when your SSH session ends, and the symptom is silence rather than an error. See Linux server.

Running by hand

The Run button on the agent’s page, or /run <name> from a chat app. A manual run is identical to a scheduled one — same access, same logging.

Reminders

Reminders are simpler: a one-off nudge at a time you describe.

remind me in 10 minutes to call the doctor
remind me tomorrow at 3pm to send the invoice
remind me next Tuesday at noon about the renewal

Set them from the interface or with /remind in a chat app. /remind also lists and deletes them.

They fire in your timezone, taken from your profile. If reminders arrive at odd hours, that is the setting to check first.

Reminders live in the database and the reminders view only — they are not written into your knowledge base.

Reminder or agent?

A reminder is a message to yourself at a time. It does no work.

An agent does something — checks, reads, writes, decides — and may or may not tell you about it.

“Remind me to check the server” is a reminder. “Check the server and tell me if it’s down” is an agent.