Build
Automations
Work a space does on its own: on a schedule, when a file appears, when an app says so, or when a job finishes.
What an automation is
An automation belongs to a space. It starts a new job, or continues one you name, with an instruction you wrote. It keeps running when you are away, as the person who set it up, with their agents.
You can create one on the space's Automations page, or ask an agent to. Each automation is a small file in the space, so you can see exactly what a space does on its own by looking in one folder.
Every Monday at 8:00, build the customer-health report from /work/exports/crm.csv
and save it to /work/reports. Tell me anything that changed.Triggers
An automation has one trigger.
- Schedule. A time of day, chosen weekdays, a repeat such as every 15 minutes, or a cron rule, in a time zone you choose.
- File created. A new file appears in the space, optionally under a path you name.
- App event. An app in the space announces something it declared.
- Job finished. A job's turn ends. An automation never triggers itself this way.
- Webhook. Another system calls a web address that the space gets, with a secret shown once. It works only where your workspace can be reached from outside.
- Run now. Every automation can be run by hand from its page.
Who it runs as
An automation runs as the person who set it up, with the agents they signed in to, the same way their own jobs do. Make one in the app and it runs as you straight away. Ask your agent to make one and it runs as you too, but it waits for you to approve it once with Allow to run as me, so no job can put your login on a schedule without you.
For work that needs no personal login, set it to run as the space. It then uses the space's secrets and can use pi with local models, your providers and hosted models, and it keeps running if the person who made it leaves.
A teammate can make one run as someone else, but it waits until that person agrees. If they leave the space or withdraw consent, it pauses and the space's admins are told. When someone other than that person changes what it does, it asks them again; their own changes don't.
When things go wrong or overlap
- If the machine was off, a missed run happens once when the space is back. You can choose to skip it instead.
- If the last run is still going, the new one is skipped by default. You can queue it or let them overlap.
- A run that goes past its time limit is stopped and marked timed out.
- If an agent asks for approval, the run waits, or rejects after a time you set. You are told only if the automation's notify list includes approval. By default only failures notify the space's admins.
- A space runs two automations at once by default, and the rest queue.
Where you see them
The Automations page lists each automation with its trigger in plain words, such as Every Monday at 8:00, Chicago time, plus its next and last run, who it runs as, and its history. Each run opens its job. Activity shows every run across the space, and failures notify the space's admins by default.