What is a runner?
A Runner is the Artemis Runner software, started on a workstation, server or cloud VM that you control. It connects to Artemis, picks up work, and runs your Project's Script there: the build, test and benchmark Commands. Each task gets its own copy of your code in the Runner's working folder, and the logs stream back to Artemis while it runs.
Artemis needs a Runner whenever code has to run: a Validation that checks your Commands work, a measurement of a change, or the measurements in a Discovery Run. Planning, chat and a Discovery Run with no Runner, where models judge the Versions, work without one. The Artemis CLI and your local coding agent are separate from the Runner.
A Runner goes through three separate states: registered with Artemis, online while the program is running, and selected for a Project or a Discovery Run.
Why the hardware matters
Performance depends on hardware. The same code can be quicker on one CPU and slower on another, and memory, disks and GPUs all change the numbers. Because the Runner is on hardware you chose, the results Artemis reports are measurements from that hardware. Pick hardware that looks like where the code will really run.
Artemis runs the Baseline, your unchanged code, and each changed Version on the same Runner, so a difference between them comes from the code. Several Runners on different machines can also be attached to one Discovery Run, to compare the same Versions across hardware.