GitHub Actions Runners
Avrea runners are ephemeral VMs on dedicated hardware, with colocated caching and step-level observability.
Pricing
Section titled “Pricing”Runners are billed per minute with no minimum. See avrea.com/pricing for current rates.
Runner images
Section titled “Runner images”Avrea runner images are based on the official GitHub Actions Runner Images. They come pre-installed with the same set of software as GitHub-hosted runners, including Docker, Node.js, Go, Python, Java, .NET, and more.
For the full list of installed software, see the GitHub-maintained image readmes:
Avrea images include additional tooling on top of the base GitHub images (e.g. build cache adapters). If your workflow runs on GitHub-hosted runners today, it will run on Avrea without changes.
The latest alias
Section titled “The latest alias”avrea-ubuntu-latest-* labels always point to the most recent supported Ubuntu LTS (currently 24.04). The bare avrea-ubuntu-latest maps to avrea-ubuntu-latest-2-vcpu — 2 vCPUs, 8 GB RAM.
Use latest unless you need to pin a specific Ubuntu version — see Pinned Ubuntu versions below.
Linux x86_64
Section titled “Linux x86_64”| Label | vCPUs | Memory |
|---|---|---|
avrea-ubuntu-latest-1-vcpu | 1 | 4 GB |
avrea-ubuntu-latest-2-vcpu | 2 | 8 GB |
avrea-ubuntu-latest-4-vcpu-8gb | 4 | 8 GB |
avrea-ubuntu-latest-4-vcpu | 4 | 16 GB |
avrea-ubuntu-latest-8-vcpu-16gb | 8 | 16 GB |
avrea-ubuntu-latest-8-vcpu | 8 | 32 GB |
avrea-ubuntu-latest-16-vcpu | 16 | 64 GB |
avrea-ubuntu-latest-32-vcpu | 32 | 128 GB |
The -4-vcpu-8gb and -8-vcpu-16gb sizes have the same vCPUs as -4-vcpu
and -8-vcpu with half the memory, at a lower per-minute rate. Use them for
CPU-bound jobs that do not need the extra memory. They are available on Linux
x86_64 and ARM.
Linux ARM
Section titled “Linux ARM”| Label | vCPUs | Memory |
|---|---|---|
avrea-ubuntu-latest-arm-1-vcpu | 1 | 4 GB |
avrea-ubuntu-latest-arm-2-vcpu | 2 | 8 GB |
avrea-ubuntu-latest-arm-4-vcpu-8gb | 4 | 8 GB |
avrea-ubuntu-latest-arm-4-vcpu | 4 | 16 GB |
avrea-ubuntu-latest-arm-8-vcpu-16gb | 8 | 16 GB |
avrea-ubuntu-latest-arm-8-vcpu | 8 | 32 GB |
avrea-ubuntu-latest-arm-16-vcpu | 16 | 64 GB |
Supported Linux runners get 3 GiB of disk-backed swap by default, so a job survives a short memory peak instead of being killed. Turn it off under Settings → Swap for the organization or for a single repository; the change applies to jobs that start afterwards. Disabling swap does not give its disk space back to the job.
| Label | vCPUs | Memory |
|---|---|---|
avrea-macos-26-8-vcpu | 8 | 24 GB |
avrea-macos-26-16-vcpu | 16 | 48 GB |
avrea-macos-27-8-vcpu | 8 | 24 GB |
avrea-macos-27-16-vcpu | 16 | 48 GB |
avrea-macos-latest-* labels follow GitHub's macos-latest and currently
point to macOS 26. Use an avrea-macos-27-* label to run on macOS 27.
Windows
Section titled “Windows”| Label | vCPUs | Memory |
|---|---|---|
avrea-windows-2025-2-vcpu | 2 | 8 GB |
avrea-windows-2025-4-vcpu | 4 | 16 GB |
avrea-windows-2025-8-vcpu | 8 | 32 GB |
avrea-windows-2025-16-vcpu | 16 | 64 GB |
Pinned Ubuntu versions
Section titled “Pinned Ubuntu versions”Pin a specific Ubuntu version when you need stability across runner-image rollouts. Sizes match the latest table above, including the memory-qualified sizes such as avrea-ubuntu-22.04-8-vcpu-16gb:
avrea-ubuntu-26.04-*-vcpuavrea-ubuntu-24.04-*-vcpuavrea-ubuntu-22.04-*-vcpu
ARM images are built for 26.04 and 24.04 only, including the memory-qualified
sizes, for example avrea-ubuntu-24.04-arm-4-vcpu-8gb and
avrea-ubuntu-26.04-arm-8-vcpu-16gb. There is no ARM build of 22.04.
Concurrency
Section titled “Concurrency”Avrea scales with your workloads, so you can run as many jobs in parallel as you need. For safety, the number of concurrent jobs and vCPUs has caps that depend on your usage, and they grow as your usage grows. Organizations on their free monthly minutes start with lower caps, which rise automatically once you move on to paid usage.
If you need more capacity or are planning a large rollout, email support@avrea.com and we'll raise your caps.
Set the runner label in your workflow:
jobs: build: runs-on: avrea-ubuntu-latest-4-vcpu steps: - uses: actions/checkout@v7 - run: make buildPrivate certificates in Node.js
Section titled “Private certificates in Node.js”Linux runners and hosted Linux VMs set NODE_USE_SYSTEM_CA=1. Node.js 22.19
and later 22.x releases, and Node.js 24.6 and later, then trust the
certificates in the system store in addition to their built-in roots. A
private CA you install with update-ca-certificates is trusted by those
versions without further configuration.
Older Node.js versions ignore that variable. For them, combine your private CA
certificates and /opt/avrea/certs/avrea-ca.crt into one PEM bundle, then set
NODE_EXTRA_CA_CERTS to that bundle before starting Node.js. If your workflow
overrides NODE_EXTRA_CA_CERTS on one of those versions, include the Avrea CA
in the replacement bundle; otherwise Node-based cache and artifact operations
can fail TLS verification.
The variable is set for processes running directly on the runner or VM, not inside container jobs. On hosted Linux VMs it applies after their next start.
Linting your workflows
Section titled “Linting your workflows”actionlint only recognises GitHub's own
hosted runner labels, so it reports label "avrea-ubuntu-latest-4-vcpu" is unknown after you switch. Register the labels you use in
.github/actionlint.yaml:
self-hosted-runner: labels: - avrea-ubuntu-latest-4-vcpuLabel matching is case-sensitive, and glob patterns such as avrea-* are
accepted if you prefer a single entry.
If you run actionlint through
super-linter or
MegaLinter, they pass their own
-config-file, so put the same content in .github/linters/actionlint.yml
instead.
Avrea's workflow conversion keeps this up to date for you. When it converts a
workflow it looks for how your repository runs actionlint, then adds any Avrea
labels the matching configuration does not already accept to that same
configuration, in the same pull request. If you have no actionlint
configuration yet, it creates .github/actionlint.yaml. Nothing is written
when every label is already accepted, or when your configuration is in a shape
Avrea cannot extend without risking your formatting or comments.