By The Botseon team
Routines, and what a run costs
"Standing work on a clock: how to write a routine you can trust to run while you are away, and how to read the credit meter before and after."
The promise
By the end of this guide a routine runs on its own clock, you have watched one test run before trusting the schedule, and you know how to read what a week of runs cost before the bill does.
Who this is for
Anyone who wants standing work off their own plate: a morning summary, a weekly report, a job that should just happen. One bot is enough to start with.
Set up
Chief of Staff ships as one of the five starter bots, and it ships with one routine already written. Sign in and create a bot from the five starter templates; pick Chief of Staff. Open Routines in its details pane to see what it starts with.
The jobs
Run the morning digest
A bot's job description says what it does. A routine says when: the same job, run on a schedule, without anyone opening the app. Every routine run is an ordinary run, with the same transcript, the same approval cards and the same meter as a job started by hand.
The single bot the member talks to first. Before doing a task itself, it checks whether another
bot already owns it and hands off to that bot rather than duplicating the work. Otherwise it
coordinates the day: surfacing what needs attention and keeping a running digest.
Morning digest, at 08:00 every day: Summarise what happened across every bot overnight and flag
anything that needs the member's attention today.
Read the overnight transcripts of the bots it coordinates: Always allow
Post the digest to the member: Always allow
Hand a task off to another bot: Ask first
Shot: routine-editor (the routine editor, desktop and phone).
7 credits per run, measured against six bots' overnight transcripts.
Write the weekly report
Write the instruction for one run, not for a career: what to read, what to produce, and, by saying nothing else, when to stop. "Read the tickets that arrived since the last run, tag each one, draft a reply for the routine ones, and list the rest with a line on why they need a person" is the shape a good routine takes.
10 credits per run, measured over one week of the digest's own history.
Set a budget per run
Beyond the week's allowance, every run carries a budget of its own: a number of steps, an amount of spend and a time limit. A run that reaches any of the three stops there, so a routine that wanders cannot spend the week in one night.
Not measured: this job needs a person, a paid account or a real mailbox, so no run of it can be
automated.
Now let it run
Every routine has a Test run: it runs the instruction now, once, and writes the transcript where every other run goes. Read it before the schedule takes over. A routine that ran well says nothing: the product notifies for two things only, a blocking approval and a failure, so a morning without a notification is a morning where everything ran.
On the managed service every run costs credits, and the Usage & Billing tab shows the week's spend at each route's published price, beside the allowance it draws on.
Shot: usage-tab (the Usage & Billing tab, desktop and phone).
What the bot may not do
Approvals in a routine work the same as anywhere else. A card that needs you waits for you, and the run waits with it. Handing a task off to another bot carries the write permission tag on that bot's own queue, which is why it stops at a card until you have watched it happen a few times.
A hard cap is on by default. When the week's allowance is used up, runs wait for the next week rather than spending beyond it.
Where your data is
A routine reads only what its own job description already reads, through your own connections. On the managed service that data follows the same rule as the rest of the product: EU/EEA storage, no non-EEA egress by default, US sub-processors under SCCs.
What can go wrong
| Symptom | Cause | What to do |
|---|---|---|
| The week's usage will not refresh | A transient error on the usage read | Use the refresh control and try again |
| A plan shows no weekly allowance | The plan does not include one, by design | Check the plan's own usage rows instead of the missing bar |
| A routine stopped mid-run | It reached its own step, spend or time budget | Read the transcript for which limit it hit, and adjust the instruction |
refreshFailed: 'Could not refresh your usage. Try again.',
noAllowanceLead: 'This plan does not include a weekly credit allowance.',
perRoutine: 'Per routine',
Do it yourself in 10 minutes
- Create a bot from the five starter templates and pick Chief of Staff.
- Open Routines in the bot's details pane and read the Morning digest it ships with.
- Run a Test run once, and read the transcript it writes.
- Open the
routine-editormoment on the desktop and on a phone. - Edit the instruction to say what to read, what to produce, and when to stop.
- Save the routine and confirm its schedule reads back correctly.
- Open the Usage & Billing tab and read the
usage-tabmoment for this bot. - Compare the week's spend against the allowance the plan includes.
- Set a per-run budget and confirm a run that exceeds it stops rather than continuing.
- Let the schedule run once unattended, and check that a quiet morning means it worked.