Run the Monday Hex refresh without anybody pressing the button
The project triggers with the parameters it actually accepts, polls until it lands, and reports at 09:14 that it finished or exactly which error stopped it.
The weekly numbers come out of a Hex project. Every Monday somebody opens it, sets the date range, presses run, waits eleven minutes and copies three figures into a deck. It is a button and a wait, and it lives on the calendar of whoever was slowest to say no.
A companion takes that Monday. It reads the input parameter schema so the values go in correctly, triggers the run, polls until the state stops being pending, and reports the terminal state with its timestamp and result link. The thing that actually goes wrong on a Monday is not the run failing. It is nobody noticing for six hours.
Trigger the Monday Hex run on a schedule
Fetching the project first is how the parameters come out right: the project record carries the input parameter schema, so a companion reads which keys the project defines before passing any values into a run. Then it triggers, receives a run id, and polls until the state stops being pending or running.
The terminal states are the interesting ones: completed, errored, killed. A report naming the error with its timestamp and result link beats a calendar reminder every week. On a schedule this is an operations chore that stops needing a person to hold it.
Why is triggering a run marked destructive?
Because it is a real run against real compute, and because of one parameter.
The published-results option replaces the published app output, the thing your colleagues open when they want the number. A run with the wrong date range changes what everybody else sees until somebody runs it again correctly.
So the tool is flagged destructive, the call waits for your approval, and there is a dry run for when you only need to know whether a configuration is valid: it validates without executing cells. Between the parameter schema, the dry run and the approval, the sequence is read, validate, ask, run.
What can a companion see inside a project?
Title, creator, status and the input schema it accepts, plus the runs you can trigger and poll. Cells, SQL and Python stay inside Hex, alongside its own notebook assistant, and nothing here creates, edits, publishes or reschedules a project.
Results arrive as links rather than rows: state, timestamps and result links. So a companion reports that the Monday refresh finished at 09:14, then opens what those links point at in the session you are signed into and lifts the three numbers into the deck. That last step is data extraction, and it is where the button-and-wait finally ends.
- Projects: list them, and get one with the input parameter schema it accepts.
- Runs: trigger with overrides, poll status, list recent runs newest first.
- No cell, query or code is readable; no project can be created, edited or scheduled.
- Status returns state, timestamps and result links rather than result data.
Plan around the seven-day Hex token
Hex API tokens rotate every seven days.
That is how the tokens are issued rather than anything this integration chose, and it has a blunt consequence: a connection built on one starts returning 401 about once a week until somebody re-mints it in workspace settings and reconnects. The integration says exactly that when it happens instead of failing quietly.
So the sweet spot is work you ask for and work on a cadence you maintain. Run the cohort project with last quarter’s dates and tell me when it lands, every Monday, with the token refreshed when the message says to. Put the reminder in the same routine and the maintenance stops being a surprise.
Experience Strawberry for free
DownloadTrusted by fast-growing companies worldwide
Frequently asked questions
Strawberry is free to download and includes AI credits to start. Paid plans begin at $20/month. See pricing. · Reviewed · Canonical facts for AI agents