Skip to content

Creating an automation

Choose the trigger, the project, the workspace and the chain of actions; use the pre-check so work is not wasted.

View Markdown

Before you start

  • A project in the app, with the repository the automation will work in.
  • If there is an agent in the chain, the agent installed and signed in.
  • Clarity about when it fires and what it does. A vague request running every hour produces an hour of vague work per hour.

Step by step

  1. Open Automations and use Create automation (or Create from scratch, on the empty screen). The empty screen also offers three ready-made templates: Review PRs in the morning, Tests on every commit and Notify when the agent stops.
  2. Give the automation a name.
  3. In the sentence, choose when this happens: the trigger. Each trigger opens its own fields. See Triggers.
  4. Confirm in (project) and in (workspace): where the actions run. See Where it runs.
  5. Choose do this: the first action. To chain more, use one more action. See Actions.
  6. If you want, open Advanced and set the pre-check.
  7. Create automation.

Editor with the Project Git trigger, a terminal command and a notification chained together

While the sentence is incomplete, the bar at the bottom says what is missing (“Choose when the automation fires.”, “Write the command.”) and the button does not save.

Where it runs

The sentence has two pieces about place:

  • in (project): the automation’s project.
  • in (workspace) or in a new workspace: the workspace the actions run in.
Choice What happens
An existing one The work happens in that folder, with what is already there
A new one for each run Each run starts in a new workspace, created from a branch (“Create from”; by default, the project’s)

Creating one workspace per run isolates runs from each other, and uses resources every time: disk for the workspace, time to prepare the environment. There is an option to run the project setup in each new workspace. Use it when independent runs are the point.

If the action is Agent in a workspace and the workspace is an existing one, you also choose the agent session:

  • New on every run: each run starts clean. It is the most predictable: nothing from one run contaminates the next.
  • Reuse the previous one: the next runs go to this automation’s previous session; if it no longer exists, Elyra opens a new one. It makes sense for work with continuity, such as a review that evolves.

Pre-check

In Advanced, the pre-check is a command that runs before every run:

  • Exits with code 0 within the time limit: the run happens.
  • Any other exit, or running out of time: the run is skipped and recorded in the history as “Skipped by the pre-check”.

The time limit is chosen among 30 s, 1 min, 2 min, 5 min and 10 min. Empty means no check.

It is how the automation avoids spending work when there is nothing to do. For example, a command that lists the open pull requests and fails when there are none.

Pre-check field under Advanced, with the time limit

Off from the start

To leave the automation ready without it running, create it and then turn the switch off. The state is kept and nothing runs until you turn it on.

From the command line

Janela do terminal
elyra automations create --name "Daily review" --trigger daily \
--prompt "Review the open changes and summarize" --provider codex

For triggers that are not a schedule and for chains of actions, use --trigger-json and --actions. See Triggers, Actions and the webhook walkthrough.

Scope and limits

  • The automation runs without you watching. Before turning one on, confirm what the agent or the command can do in that project: they act with the permissions they have.
  • The pre-check decides, it does not fix. It skips the run; it does not repair the environment.
  • One workspace per run has a recurring cost. At short intervals, that becomes disk and time.
  • Terminal command does not run on a remote project (SSH) yet. The editor refuses and asks for another action or a local project.

See also