2 to 32 vCPU on Linux x64, ARM64 or Windows. Secrets, OIDC and matrix builds keep working.
Change one line, runs-on:, and the job runs here: 1.2–2.1× sooner than GitHub-hosted at 2 vCPU, at $0.004 a minute against GitHub’s $0.006. Each job gets its own fresh virtual machine, in Europe.
Card required. 3,000 minutes a month at 2 vCPU are included.
Three open-source builds on GitHub’s 2 vCPU runner and on ours: each time is the whole job, and each cost is what the build was billed. Why your own gain depends on the job is on the measured builds page.
eslint/eslint @ c27bc92, built with node Makefile mocharust-lang/cargo @ e8cb624, built with cargo build --release --lockedprometheus/prometheus @ d15adb9, built with go test ./tsdb/ -test.tsdb-isolation=falseEach time is the median of at least 21 approved rounds that both runners completed.
Each size is a label: choose it in runs-on for the job that needs it.
runnersio-1vcpu-ubuntu-2404runnersio-2vcpu-ubuntu-2404runnersio-4vcpu-ubuntu-2404runnersio-8vcpu-ubuntu-2404runnersio-16vcpu-ubuntu-2404runnersio-32vcpu-ubuntu-2404Like GitHub-hosted runners, there is nothing to install, patch or scale. Unlike them, you choose the size of every job, up to 32 vCPU, and a job waiting in the queue costs nothing.
Three runner types, in the sizes above. Running a fleet of your own today? This is what replacing it looks like.
The size is part of the label, so the jobs that need cores get them and the rest stay small.
The Actions cache is included. A Docker layer cache is an add-on at $0.50/GB-month of what it stores.
You pay for the minutes a job runs, and nothing while it waits for a runner.
Installing the App moves nothing: no job runs here until a workflow asks for one of our labels. Change runs-on yourself, or pick a repository and we open a pull request that changes it and nothing else. Its checks run on our runners, so you compare times on your own code before you merge. There is no per-seat fee, and a monthly spend cap you set stops new jobs when it is reached; running ones finish.
runs-on:, and the size is part of the label, so you choose it for each job. A compile or a test suite spread over many files gains from more cores; a job that waits on the network does not. The labels page has every label.While a job runs, its code and the secrets GitHub hands it are on a machine we operate, as on any runner you do not host yourself. Each job gets its own virtual machine, starting fresh and shared with no other job. Nothing a job leaves on its machine can be seen by another job.
The repositories you choose. GitHub will list read and write permissions, and each write has one use. Administration registers a single-use runner for each job and removes it. Contents, Workflows and Pull requests open the migration pull request. Actions cancels a run we refused, or one holding a job you stopped. Checks and Commit statuses report Terraform plans, if you connect Terraform. Your secrets, environments and branch protection stay on GitHub.
Change the label back and the job runs on GitHub again: the runs-on label is the only change a workflow needs, and your actions, secrets and triggers were never ours to untangle. Uninstall the App on GitHub and we see nothing more. We delete build output and test results 90 days after each job.
Card required. 3,000 minutes a month at 2 vCPU are included.