Grok Bot routines: six fields, 50 per bot, 20 runs kept
A Grok Bot routine has six fields and one time zone. A bot owns up to 50, the app keeps 20 run records each, and event triggers are Slack and GitHub only.
A Grok Bot routine is the schedule for a workflow. xAI’s skills and routines page says a routine specifies “when to run a workflow, on a schedule or, where supported, after an event,” and lists six fields to confirm: the owning bot, the schedule and time zone, the input source, the expected result, the approval boundary, and what happens when a source is missing. A bot can own up to 50 routines, and the app keeps the 20 most recent run records for each.
Those are the platform facts. The rest of this page is what they mean when you write one.
What are the six fields?
| Field | What to write |
|---|---|
| The owning bot | One bot. The FAQ says a routine "assigns a workflow to one Bot and tells it when to run." |
| The schedule and time zone | One clock time, one IANA zone. Set it deliberately. |
| The input source | The systems the run reads, and the skill that reads them. |
| The expected result | What a normal run produces, including "nothing" if silence is the normal case. |
| The approval boundary | What the run may do alone and where it stops for a human. |
| What happens when a source is missing | The no-data and stale-data policy, decided in advance. |
Scroll sideways for the second column on a narrow screen.
The docs recommend a routine keep to “a narrow matching rule and a clear response,” and they ask for “a no-data and stale-data policy” as part of every automation. The sixth field is the one most free templates leave blank, and it is the one that decides whether anyone finds out three weeks later that nothing ran.
What are the limits?
Two numbers, both from the same page:
- “A Bot can own up to 50 routines.”
- “The app keeps the 20 most recent run records for each routine.”
The first is generous. The busiest Botemplate bot, Backstop, runs nine. The second matters more than it looks. A daily routine’s history rolls over in under three weeks, so a routine that produces something worth keeping has to write it somewhere durable on every run. The run log is not the record. The register sheet, the CRM or QuickBooks is the record, and every Botemplate routine writes its outcome there rather than trusting the app to remember.
How do time zones work?
A routine carries one schedule and one time zone. That is the whole capability, and it has one consequence any template that sends messages has to face: a routine cannot resolve a send time per recipient. Ledger wants dunning emails to land at 10:00 in the customer’s zone, not the sender’s. So the template creates one dunning routine per time zone band, each holding a single fixed clock time and a single zone. A pack that claims one routine can time itself per recipient is describing software that does not exist.
Set the zone deliberately in every case. A collections email that lands at 03:00 reads as automated and gets ignored; a vendor-notice reminder that fires on Saturday still fires, which is why Quartermaster’s notice deadline guard runs seven days a week.
What is the missing-data protocol?
A missing-data protocol is the sixth field written out: what a routine does at 07:30 when the system it reads is down, empty, or stale, decided before it happens. xAI’s docs ask for a no-data and stale-data policy in every automation, and the app keeps only the 20 most recent run records, so a routine that fails quietly for three weeks leaves no trace that anyone will read. A protocol worth the name answers three questions. Which source could the run not reach, and was the data it did get complete. What does the run post instead of the number it would have posted, and does it name the gap rather than fill it with last week’s figure. And who is told, by name, when the gap persists. Every one of Botemplate’s 34 routines names its owning bot, schedule and time zone, input source, expected result, approval boundary, and what to do when a source is missing, in the ROUTINES.md that ships with the bot.
What about event triggers?
The docs say routines can run “where supported, after an event,” and the events they list are Slack messages and GitHub notifications. Nothing else is named. A template that claims to fire on a Stripe webhook, a new HubSpot deal or an inbound email is claiming a trigger the docs do not describe.
Backstop uses the one trigger that fits its job: a client message signal that fires on a Slack message event, subscribed per channel, with a narrow matching rule so it is not a broad listener. The other eight of its routines are on the clock.
What does a good schedule look like?
The time should have a reason, and the reason should be in the routine. A few from the pack:
- Quartermaster’s monthly spend close runs on the 4th, because card statements and bank feeds for the prior month have almost always landed by then, while the 1st is reliably too early and the 10th wastes a third of the month. A retry runs on the 6th.
- Quartermaster’s new recurring charge detection runs Thursday at 16:00, because most card charges have posted by then and the common refund window on a converted trial is 14 to 30 days, so catching it Thursday leaves a full week inside that window.
- Ledger’s reconciliation break watch runs daily at 06:30, all seven days, and on a normal day posts nothing. Silence is the expected result, and the field says so.
A schedule with a reason can be changed by a buyer who understands the reason. A schedule without one gets changed to 09:00 because 09:00 seemed fine.
How many routines does a template need?
Fewer than 50, and more than none. The five Botemplate bots carry 34 between them: five for Waterline, six for Quartermaster, seven each for Ledger and Bellwether, nine for Backstop. Each carries the six fields and a missing-data protocol, and each is on its bot’s page with its cadence. How it works covers what else is in the box, and the five bots together are the Operator Pack for $69.