# Cronjobs

Settings → Cronjobs defines scheduled prompt runs: a prompt the workspace runs on a schedule, by itself, inside a session. The run is an ordinary agent turn, so its output shows up in the Agents tab like any other conversation — you can read it, continue it or ignore it.

A run happens inside the workspace, so it keeps its own schedule even while your browser is closed.

# The list

Each row shows the job's Name, its Schedule, its Session mode, when it last ran and when it next runs. A run that is in progress shows running now; the last outcome is done, failed or skipped.

Every row also shows where its definition comes from:

Source What it means
(no badge) A job you created here. You can edit, enable, disable and remove it.
Managed by the company settings Defined by your company's settings repository. Read-only here.
Managed by the addon settings Defined by an add-on attached to your workspace. Read-only here.

Definitions are merged by their id: an add-on's job is overridden by a company job with the same id, and both by one of yours. Re-read definitions picks up changes made to the company or addon definitions without waiting for the next automatic refresh.

If nothing is defined yet, the tab says so, and New cronjob starts the first one. Unusable definitions are skipped, and the tab lists what was skipped.

# Creating a job

New cronjob asks for:

Field What it is
Name A label for the job
Prompt What the agent should do — written as you would write it in a session
Schedule Either a cron expression (five fields: minute, hour, day of month, month, day of week, in UTC) or an Interval such as 30m or 6h
Directory Optional. The project or repository the run works in. Empty runs at the workspace's root folder.
Agent and Model Optional. Both default to the ones a new session in this workspace starts with.
Session per run Whether each run gets a new session (runs stay apart) or reuses one single session (runs collect in one conversation you can keep following)
When a run is still going Whether to skip the new run or queue it, when the previous one has not finished
Enabled Whether the schedule is active. A disabled job keeps its definition and can still be run by hand with Run now.

A job is stored under an id — a short name derived from the one you type — which is the key the definition layers are merged by. Saving a job with an existing id replaces that definition, which is how you override a company or addon job with your own.

# Running one by hand

Run now starts the job immediately, whether or not it is enabled. That is the way to test a job before you let it run on its own. When the run starts, its session appears in the Agents tab.

# While you work

  • The schedule is checked by the workspace itself, every 30 seconds. If that check is switched off on your workspace, the tab says so and only Run now works.
  • A run missed while the workspace was down catches up the next time the workspace starts.
  • Removing a job deletes its definition. A session an earlier run created stays where it is.
  • Runs are ordinary sessions, so they cost model usage like any other turn, and they are subject to the same permissions and company rules.

Agents
../../terminal/
Company
../company/
Settings
../