The options at a glance
Product details checked 17 November 2026.
| Option | Who operates compute | Workflow changes | GitHub Actions still schedules jobs |
|---|---|---|---|
| Managed runner provider | Provider | Install the integration, select runner labels and check image and cache compatibility | Yes |
| Self-hosted VMs | Your team | Register runners, select labels and maintain images and capacity | Yes |
| Actions Runner Controller (ARC) | Your team, on Kubernetes | Configure the cluster, runner scale sets and execution environment | Yes |
| Runner software in your cloud account | Your team owns the cloud resources; the software automates provisioning | Deploy the integration and configure runner sizes, images and network access | Yes |
| Another CI platform | Your team or the platform, depending on its execution model | Convert configuration or validate a supported compatibility mode | No |
| Another engine using GitHub Actions syntax | Depends on the engine | Check supported triggers, actions, credentials and workflow location | No |
Moving CI does not necessarily mean moving the repository. CircleCI, Buildkite and Jenkins can build code hosted on GitHub. GitLab also supports CI/CD for GitHub repositories on its Premium and Ultimate tiers.
Managed runners: keep GitHub Actions scheduling
A managed runner provider supplies the machines that execute your GitHub Actions jobs. GitHub continues to create the workflow runs, resolve job dependencies and dispatch work. If build time or machine cost is the problem, this is usually the smallest change to evaluate.
After installing the provider’s integration, a compatible workflow can move by changing its runner label:
# Before
runs-on: ubuntu-latest
# After
runs-on: avrea-ubuntu-latest
Test a copy of the build and test jobs, with deployment and publication steps disabled, before switching required checks. Installed tools, CPU architecture, networking and cache integrations can differ. A workflow that needs an image-specific package or a particular action version may need more than a label change.
Providers also put build state in different places. A remote Docker builder, a persistent disk attached to the runner and a remote cache for individual compiler outputs solve different problems. Compare the setup your workflow needs, then measure the whole job with an empty cache and after the cache has been populated.
Our runner provider comparisons cover Avrea, Blacksmith, Depot, Namespace, WarpBuild, Ubicloud and GitHub-hosted runners. The runner pricing comparison separates minute rates, plan fees and included usage. Keep the machine’s memory and architecture in the comparison: the cheapest minute is not necessarily the cheapest completed job.
Self-hosted VMs and ARC: operate the runner fleet
Self-hosting gives your team control over images, machine placement and access to private services. It also gives the team responsibility for updates, scaling, job isolation and recovery when a runner stops accepting work. Include that work and idle capacity in the cost estimate.
VM-based runners fit teams that need a particular operating system, hardware or network boundary. Use ephemeral runners where practical, and decide explicitly which build state should survive a job. Reusing a machine without cleaning its workspace can carry files or credentials into later jobs.
Actions Runner Controller is GitHub’s Kubernetes operator for scaling self-hosted runners. It creates runner scale sets and adjusts capacity as jobs arrive. ARC is a reasonable option for a team that already operates Kubernetes and wants runner infrastructure there. Your team still owns the cluster and the runner environment, and GitHub Actions still dispatches the jobs.
Software such as RunsOn offers another way to run a fleet in your own AWS account. Its license and AWS infrastructure charges are separate. That is a different operating and billing model from buying minutes on a provider’s managed fleet.
The guide to GitHub Actions runner types explains the tradeoffs between GitHub-hosted, self-hosted and managed runners in more detail.
Another CI platform: review configuration and integrations
Changing platforms moves scheduling away from GitHub Actions.
| Platform | When to consider it | What to check |
|---|---|---|
| GitLab CI/CD | Your team wants GitLab's repository, CI and registry integration | GitLab CI configuration, runner ownership and the plan required to build GitHub repositories |
| CircleCI | You want another CI service while keeping code on GitHub | Native CircleCI configuration or its supported subset of GitHub Actions workflow files |
| Buildkite | You want a hosted control plane with a choice of hosted or self-hosted agents | Agent placement, pipeline configuration and any compatibility-preview limits |
| Jenkins | You need a self-managed automation server with custom integrations | Controller operations, plugins, agents and pipeline maintenance |
CircleCI’s migration guide describes both configuration conversion and running a supported subset of GitHub Actions workflow files. Buildkite offers hosted agents as well as self-hosted agents. Check those capabilities against your workflows before estimating a complete rewrite.
Inventory your uses: steps, secrets, deployment approvals and required checks. Native configuration may need replacements for marketplace actions; a compatibility mode may support some actions but not the GitHub services they call. Configure repository access through the new platform’s integration and verify its credential scopes and lifetime. Do not assume it supplies the same job-scoped GITHUB_TOKEN behavior.
Run both pipelines on the same commits before changing branch protection. Confirm that the new platform reports the intended check names and results, including failures and canceled runs. Required checks may need updating, but they do not all have to be renamed simply because execution moved.
Another workflow engine: check compatibility before moving
Avrea CI runs supported GitHub Actions workflow syntax using Avrea’s orchestration and runners. It is available in closed beta by request. Start with a build or test workflow: copy it into .avrea/workflows/, select Avrea runner labels and review its credentials and integrations. The migration guide walks through comparing runs before switching required checks.
The compatibility reference lists the supported features and differences. Secrets and variables are not imported automatically. Some triggers, including schedule and merge_group, are unsupported, and GitHub environment approvals are not enforced. Workflows that need those controls should stay on their existing engine until an equivalent is available and tested.
Avrea CI removes GitHub Actions scheduling from the execution path, but workflows can still depend on GitHub for repository access, events or APIs. Evaluate those dependencies separately from the scheduler. Keeping familiar YAML does not remove migration work or make a workflow independent of GitHub.
What outages tell you
A runner swap only changes part of the system. If GitHub Actions cannot dispatch a job, a third-party runner cannot start it either. ARC and self-hosted VMs also retain that dependency.
GitHub’s July 2026 availability report illustrates two different failures. On July 9, an unhealthy backend service prevented GitHub-hosted runners from being provisioned. On July 19, an expired internal certificate disrupted connectivity for self-hosted and larger runners while standard GitHub-hosted and macOS runners were unaffected. Operating a separate fleet avoids the hosted-fleet provisioning dependency, but it still needs a working connection to GitHub Actions.
If reliability is the reason for moving, identify which dependency caused your incidents. Then test the proposed replacement’s behavior when repository access, webhook delivery or runner connectivity is unavailable. Changing platforms changes these dependencies; it does not remove outages as a category.
How to decide
Start with one representative workflow. For a runner change, measure startup separately from build time, then compare cold and warm runs at the same commit. A fixed startup delay matters much more on a ninety-second job than on a twenty-minute build. Our build, CPU and cache benchmarks show why a CPU score alone cannot predict a complete build.
For self-hosting, add the cost of operating the fleet to the machine bill. For a platform or engine change, make a compatibility checklist from the workflow itself and test required checks and deployment controls before migrating production jobs.
Avrea’s managed runners support Linux, macOS and Windows. Install the GitHub App and test a compatible workflow using the setup guide. Current machine sizes and rates are on the pricing page. Avrea also supports GitLab runners; this guide focuses on alternatives for teams currently using GitHub Actions.
PromoRepublic cut execution from about 4 minutes to 42 seconds, about 5.7x faster. Focal measured 2x faster Turborepo CI against GitHub Actions plus Vercel Remote Cache, and Syncle cut job times roughly in half. Those are customer results on their own workloads; use your workflow to evaluate the change.
Avrea’s managed compute runs on Avrea infrastructure. If your requirement is to keep build compute inside your own cloud account, evaluate a self-hosted option instead. If GitHub’s standard runners already meet your needs within the included allowance, compare the value of faster feedback before paying another provider.

