Why Pitlane
Pitlane is for one situation: you run GitHub Actions for private repositories, you have a Linux machine to spare (a VM, a Proxmox LXC container, a server), and you want each job to get a fresh runner without starting from cold caches every time, and without running Kubernetes to get there.
What you get
On a real private monorepo (pnpm workspace, Nuxt + Prisma, Postgres and Redis services, three Docker image builds), with Pitlane in an unprivileged Proxmox LXC container with 12 vCPU:
GitHub-hosted (ubuntu-latest) | Pitlane | |
|---|---|---|
| Whole CI (longest job) | ~6 min 30 | ~3 min |
| Docker images & smoke test | 393 s | 127 s |
actions/cache save, 200 MB | about 9 s | about 1 s |
The speed comes from what a slot keeps between jobs: its Docker daemon with the images and build cache, the package manager caches and toolchains of the GitHub owner, the actions/cache entries on local disk and a Docker Hub mirror. Steps that only burn CPU (type-checking, unit tests) run at the speed of your hardware: in the same benchmark, lint went from 40 s to 38 s and unit tests from 161 s to 159 s. The full table is in the README.
Compared with the alternatives
Facts below come from each project's README or documentation, checked on 15 November 2026. If one of them is wrong or out of date, please open an issue.
| Runs on | Inbound URL | Kept between jobs | Web UI | Runner OS | License | |
|---|---|---|---|---|---|---|
| Pitlane | one Docker host | not needed | Docker layers and images, package caches, toolchains, actions/cache | yes, with roles | Linux | MIT |
| actions-runner-controller | Kubernetes | not needed | what you mount yourself | no | Linux | Apache-2.0 |
| runscaler | one Docker host, or Tart VMs on Apple Silicon | not needed | opt-in actions/cache receiver, optional shared volume | no (CLI) | Linux, macOS | MIT |
| GARM | a controller and a provider | webhooks for pools, not needed for scale sets | depends on the provider and image | yes | depends on the provider | Apache-2.0 |
| terraform-aws-github-runner | your AWS account | GitHub webhook to API Gateway | nothing on ephemeral instances by default | no | Linux, Windows, macOS | MIT |
| Hosted runners | the provider's machines | not needed | provider features | the provider's | Linux, others by provider | commercial |
actions-runner-controller (ARC)
ARC is GitHub's own controller, installed with Helm on a Kubernetes cluster. Its runner scale sets use the same API as Pitlane, so neither needs a webhook. Pods start fresh, and anything you want to keep between jobs (a Docker layer cache, a package cache) is a volume you set up yourself.
Pick ARC if you already run Kubernetes and want runners managed like your other workloads, with Kubernetes doing the scheduling across many nodes. Pick Pitlane if you would otherwise run a cluster only for CI.
runscaler
runscaler is the closest project to Pitlane: a single Go binary on a Docker host, built on the same actions/scaleset library, with a fresh container per job. It can also run macOS and Linux VMs with Tart on Apple Silicon Macs. It is driven by a TOML file and a CLI, and can opt into a built-in actions/cache receiver.
Pick runscaler if you need macOS runners, or prefer configuration files to a web UI. Pick Pitlane for warm Docker-in-Docker daemons per slot (images, layers and services: images stay between jobs), per-owner package caches, a web UI with roles and per-job resource usage, and strict mode for public repositories.
GARM
GARM manages pools of runners through providers: LXD, Incus, Kubernetes, OpenStack, AWS, Azure, GCP, Oracle and others, each a separate executable. It has a web UI, supports Gitea and GitHub Enterprise Server, and can use webhooks or scale sets.
Pick GARM if you need runners on several clouds, VMs per job, Gitea, or GitHub Enterprise Server. Pick Pitlane if one Docker host is enough and you want caches that stay warm without building runner images per provider.
terraform-aws-github-runner
terraform-aws-github-runner (started by Philips Labs) deploys runners on EC2, often spot instances, with Lambda functions that scale them on GitHub webhooks. It supports Linux, Windows and macOS, and ephemeral instances are terminated after their job.
Pick it if your CI belongs in AWS and you want an EC2 instance per job. Pick Pitlane if you have a machine of your own and want its disk to keep the caches.
Hosted runners and RunsOn
Blacksmith, Depot, Namespace, Ubicloud and WarpBuild sell runners billed per minute, advertised as cheaper or faster than GitHub-hosted ones, with their own caching (Blacksmith, for one, keeps Docker layers on persistent disks billed per GB). You change runs-on: and own no hardware. RunsOn runs in your own AWS account, free for non-commercial use and licensed for commercial use.
Pick a hosted service if you do not want to own, patch and secure a machine, need Windows or macOS, or run untrusted code from forks. Pick Pitlane if you already have the hardware and would rather not pay per minute.
When Pitlane is the wrong choice
- Windows or macOS jobs. Pitlane only runs Linux runners (amd64, arm64).
- Untrusted code. Every job controls a privileged Docker daemon and can take over the host. Strict mode keeps honest jobs apart; it does not contain a malicious one. See Security.
- GitHub Enterprise Server. Pitlane talks to GitHub.com only.
- A fleet of machines. A Pitlane instance drives one Docker host. Scheduling jobs across many nodes is what ARC on Kubernetes or GARM are built for.
- No time to run a host. Self-hosting means updates, disk space and a machine that can fail. Pitlane prunes its caches and, installed as a binary, updates itself, but the host is still yours.
What it costs
Pitlane is MIT-licensed and has no paid edition. You pay for the machine it runs on. Plan about 2 vCPU and 4 GB of RAM per parallel job, and 100–200 GB of fast disk for the caches (Getting started).