The change
- runs-on: ubuntu-latest
+ runs-on: tbench-2vcpu-ubuntu-2404Nothing else is added. No new step, no extra job, no action to install, and no workflow logic to restructure. GitHub resolves runs-on: by matching labels, so a runner that advertises this label simply becomes eligible for the job.
How it is set up
- Install the App. Authorize the Tbench Actions GitHub App on the single repository you want to run. This is the only authorization step; there is no account application and no approval queue.
- Start a runner. Run the worker on the machine that should execute the job. It registers itself as an ephemeral runner for one job, takes only work for the installation you authorized, and deregisters when the job ends.
- Change one line. Replace the runner image in
runs-on:. Everything else in the workflow — checkout, steps, secrets, matrix, artifacts — is untouched.
The public GitHub App is not published yet, so there is no install link on this page. The beta is currently reached through a scoped pilot; write to support@tbench.in and we will enroll your repository directly. Do not treat the absence of a button as the beta being generally available.
Runner labels
Each label is a size and an image. Use the smallest one that fits the job; a larger label does not run the job faster, it only reserves more of your machine.
| Label | Write this in the workflow |
|---|---|
tbench-2vcpu-ubuntu-2404 | runs-on: tbench-2vcpu-ubuntu-2404 |
tbench-2vcpu-ubuntu-2204 | runs-on: tbench-2vcpu-ubuntu-2204 |
tbench-4vcpu-ubuntu-2404 | runs-on: tbench-4vcpu-ubuntu-2404 |
tbench-4vcpu-ubuntu-2204 | runs-on: tbench-4vcpu-ubuntu-2204 |
tbench-8vcpu-ubuntu-2404 | runs-on: tbench-8vcpu-ubuntu-2404 |
tbench-8vcpu-ubuntu-2204 | runs-on: tbench-8vcpu-ubuntu-2204 |
tbench-16vcpu-ubuntu-2404 | runs-on: tbench-16vcpu-ubuntu-2404 |
tbench-16vcpu-ubuntu-2204 | runs-on: tbench-16vcpu-ubuntu-2204 |
Beta covers the Linux x64 SKUs above. Windows, macOS, GPU and arm64 are not part of this beta.
What the beta includes
- 5 runs to start, per organization. An opening credit, not a recurring free tier. After it is spent, the work is metered — nothing starts charging without a decision made in advance.
- Bring your own compute. Tbench charges nothing per minute on the bring-your-own-compute beta — the runner is your hardware. Electricity, wear and any cloud bill for that host remain yours, and we do not describe them as free.
- Ephemeral, one-job runners. The runner registers for one job and deregisters when it ends. Nothing persists between jobs on the runner itself.
What this is not
- Not a managed hosted fleet. You provide the machine; we provide the registration and the control plane.
- Not a cache. There is no equivalent of persistent dependency or build caches, and no speed claim attached to the beta.
- Not a security boundary for untrusted code. A container shares the host kernel, so run code you already trust on your own machine.
- Not Windows, macOS, GPU or arm64. See the label table above for the covered set.