Set up a runner

A connected project shows the Dashboard which experiments exist. To produce measurements, each experiment also needs a runner: an environment that checks out a commit and executes the experiment.

Choosing a runner is part of defining what a measurement means. Build time from controlled hardware and build time from a shared virtual machine are not interchangeable signals.

Open runner setup

From the connected project, open the command palette with ⌘K on macOS or Ctrl+K elsewhere. Select Set up a runner. The Dashboard opens the runner setup for that project.

Choose a runner for the measurement

Use a Wezel-provided bare-metal runner for measurements that are sensitive to the machine itself, such as execution time, CPU time, and memory usage. A stable host reduces noise and makes changes between commits easier to trust.

Bare-metal runners are not enabled automatically for new workspaces. In runner setup, find Wezel bare-metal and select Request access. The Dashboard opens a prefilled email to the Wezel team with the current workspace and project details.

Use GitHub Actions when convenience matters more than a tightly controlled machine. It is a good fit for deterministic measurements, experiment smoke tests, and other checks that do not depend heavily on host performance. Hosted GitHub runners can vary between jobs, so repeated sampling cannot make them equivalent to dedicated hardware.

Whichever runner you choose, Wezel stores its operating system and architecture with the result. Each platform gets a separate measurement history.

Add a GitHub Actions runner

With runner setup open:

1. Select Add GitHub Actions runner.
2. Enter a name and choose the runner’s platform.
3. Set the GitHub runs-on label and workflow path.
4. Select Add runner.
5. Select Open setup PR, then review and merge the generated workflow on GitHub.
6. Assign the new runner to the experiments it should execute.

The workflow lets the Dashboard dispatch experiments through the repository’s GitHub Actions configuration. Keep its platform selection consistent with the machines selected by runs-on, because the platform identifies the result’s comparison history.

Assign runners to experiments

Open the command palette and select Set up a runner if you have left the runner setup. Each experiment can be assigned independently, so a repository can send performance-sensitive experiments to bare metal and less sensitive work to GitHub Actions.

Select the runner or runners that should execute each experiment. An experiment without an assignment remains visible in the Dashboard but cannot produce a scheduled result.

With a runner assigned, the Dashboard can produce the experiment’s first recorded result. Next, learn how to track performance over time.