What is a CI runner?
A CI runner is the machine that executes the jobs in a continuous-integration pipeline. The CI system (GitHub Actions, GitLab CI, Buildkite, Azure Pipelines) decides what should run; the runner checks out the code, runs the build and test steps and reports back. Every CI platform has hosted runners of its own and also lets you attach runners from a third-party vendor, in your own cloud account or on your own hardware.
The two halves of a CI system
A CI system is split in two. The control plane stores your workflow files, listens for events such as a push or a pull request, decides which jobs to start and shows logs and results. The runner is the worker: a computer running a small agent program that asks the control plane for work.
When a job is assigned, the runner checks out the repository, runs each step in a shell or a container, streams the logs back, uploads artefacts and reports success or failure. GitHub Actions and GitLab CI call this worker a runner. Buildkite and Azure Pipelines call it an agent. The idea is the same.
The split matters because the two halves can come from different places. You can keep your workflows in GitHub Actions and run the jobs almost anywhere.
How a runner connects
A runner registers with the CI system for a repository, an organisation or a group, and advertises labels or tags that describe it: its operating system, architecture and size. It then holds an outbound connection to the CI service and waits. Because the runner dials out, it works behind a firewall with no inbound ports open.
A job states what it needs. In GitHub Actions that is the runs-on key; in GitLab CI it is tags. The control plane hands the job to an idle runner whose labels match.
The kinds of runner
- Platform-hosted runners. The CI platform runs the machines. GitHub-hosted runners and GitLab-hosted runners start a fresh virtual machine for each job from a maintained image. Nothing to set up, billed per minute beyond a free allowance.
- Third-party managed runners. A vendor hosts the machines and registers them with your CI system. You change a label in your workflow. They compete on price per minute, faster CPUs and caching.
- Runners in your own cloud. A product or a cloud service starts a virtual machine in your own cloud account for each job. Code and secrets stay inside your boundary and you pay the cloud bill.
- Self-hosted with open-source software. You run an autoscaler on Kubernetes, cloud instances or bare metal. No licence fee, and all of the operations.
- macOS runners. Apple platforms need Apple hardware, so Mac runners are a separate market of hosted Macs, dedicated Macs and virtualization tools.
The home page describes each kind with when to choose it.
Why the choice matters
- Speed. CPU generation, core count and cache location set how long developers wait for a result.
- Cost. Hosted runners bill per minute; rates differ several-fold between providers and between Linux, Windows and macOS.
- Security. A runner executes your code holding your secrets. Whether each job gets a fresh machine, and who else shares the hardware, is a supply-chain question.
- Reach. Only a runner inside your network can reach a private database or registry without opening it to the internet.
See how to choose a CI runner for a checklist.
Frequently asked questions
What is the difference between a runner and an agent?
Is a CI runner a virtual machine or a container?
Do I have to use the runners my CI platform provides?
Sources
Check before you rely on it. Products change quickly, and this page reflects public documentation as of November 22, 2026. Confirm anything that matters in the provider’s own documentation.
Cite this page
CI Runners. (2026). What is a CI runner?. cirunners.org. Last reviewed November 22, 2026. https://cirunners.org/guides/what-is-a-ci-runner/