Runners

CI runner options, compared on checkable criteria

Every cell says what the source documents, or where to check it. No ratings, no guessed prices.

CriterionGraphide CIGitHub-hostedSelf-hostedOther runner services
Setup effortOne line: runs-on: graphide. Install the GitHub App on an organization and press Switch; it opens one pull request that changes the runs-on line.None for standard runners: runs-on: ubuntu-latest. Larger runners are created in organization settings first.You provision, register, patch and scale the machines, or run a controller such as Actions Runner Controller on Kubernetes.Check each vendor's setup docs for the app, labels and repo changes it needs.
Undo pathSwitching back opens a pull request that restores each runs-on line.Nothing to undo for standard runners; a larger runner's label is changed back.Change runs-on back to a GitHub-hosted label; then retire the machines.Check each vendor's docs for how its labels are removed.
CachingWarm machines. Graphide's agents tune each job's caches, machine and how it runs, and retune as the code changes.Fresh VM per job. Caches restore from GitHub's cache service via actions/cache; check the docs for size and eviction limits.Disks can stay warm between jobs if runners are not ephemeral; you own cleanup and isolation.Varies by vendor: check each one's caching docs.
SizingEach job runs on a machine sized to it, fresh for every job; up to 32 vCPU today.Fixed standard sizes per OS; larger sizes on paid plans. Sizes and plans: check the docs.Whatever you provision.Varies by vendor: check each one's machine sizes.
Scopex64 Linux jobs. macOS, Windows, Arm and self-hosted jobs stay where they run.Linux, Windows and macOS.Any OS and architecture you run. GitHub advises against self-hosted runners on public repositories.Varies by vendor: check supported OS and architectures.
PriceOne fixed monthly price, quoted from what you run. Trial: 7 days or $10 of usage.Per-minute billing beyond included minutes on private repos. Rates: check GitHub's billing docs.Your infrastructure and the time to run it. Check GitHub's billing docs for any platform charge.Check each vendor's pricing page.
SourcesGraphide CI docs · Graphide pricingGitHub-hosted runners · Larger runners · BillingAbout self-hosted runners · Actions Runner ControllerBlacksmith · Depot · Namespace · BuildJet · Ubicloud · RunsOn · WarpBuild

Follow each source link to check a cell: vendors change sizes and prices.

Which option fits which bottleneck

Start from what the analyzer flagged, not from a vendor list.

FlaggedStart withWhy
Undersized runnerGraphide CIWarm machine sized to the job; one runs-on line to move, one pull request to undo.
Cold cachesGraphide CIIts agents tune each job's caches and retune as the code changes.
Serial testsMatrix sharding, then a bigger runnerSplit the suite first; every test still runs. Then size the runner to the shard.
Queue waitConcurrency settings firstCancel superseded runs; check your plan's concurrency limit in GitHub's docs.

Not sure which case you are?

Paste one run and the analyzer tells you.

Open the analyzer