Automations
An Automation is a repeatable task done by an AI agent, like refining a ticket into a plan, triaging a scanner finding, or generating a code change. You set it up once with the instructions, agent, model, and context it needs, then use it as often as you like: on its own, or as a step in any workflow.
Kinds of Automations
| Kind | What it does | Examples |
|---|---|---|
| Prompt | Returns a response, such as an analysis, a decision, or a plan. Doesn't change code. | Check whether a ticket is ready, triage a finding, draft a ticket comment |
| Repository | Changes code across repositories and opens pull requests. | Fix a vulnerability, upgrade a library, add a workflow file |
In a workflow, a prompt Automation's output can decide which branch the run takes next.
Create an Automation
- Go to Automations and select New Automation.
- Give it a name that describes the task, for example "Refine ticket into plan."
- Write its instructions: what the agent should do, and what a good result looks like.
- Choose the agent that does the work, from your Agent Registry, and the model it uses.
- Save the Automation.
Writing good instructions
- Describe the task and the expected output. For example, "Return a plan with the files to change and the tests to add."
- Say what the agent should do when it can't finish, such as asking for missing information instead of guessing.
- Keep each Automation focused on one task. Use a workflow to chain tasks together.
What an Automation runs against
An Automation doesn't store what it runs against. You choose that each time it runs:
- On its own: select the components to run against, and add any plan or extra instructions for this run.
- In a workflow: the workflow passes in what the Automation needs, such as a ticket's details or an earlier step's output.
This lets one Automation serve many workflows and teams.
Automations that change code
Automations that generate code, like the OpsLevel Coding Automation, work across each component's repository:
- Plan the change for each component.
- Make the change in an isolated environment.
- Open a pull request for review in each repository.
Pull requests go through your normal review process, branch protections, and required checks. Agents never merge to protected branches on their own.
Runs
Every Automation run appears under Runs, with its status, the components it ran against, and its results, including any pull requests it opened. See Automation runs for running an Automation on its own and reviewing results.
Usage and credits
By default, Automation runs that use OpsLevel's built-in agents consume OpsLevel credits.
You can also bring your own Anthropic API key. Runs that use your key are billed by Anthropic to your account and don't consume OpsLevel credits. Support for more model providers is coming soon.
Updated about 6 hours ago