Cheap & fast
GitHub Actions runners
Unlimited, flat price

Yes, most do. Skipbox runs the workflows of public and private repositories on GitHub organizations and personal accounts. The VM is Ubuntu 24.04 on x64 with Docker, git, build-essential, Node 22, and Python 3.12, and the job runs as a user with passwordless sudo, as on GitHub's runners. GitHub's ubuntu-latest image has more preinstalled, so add actions/setup-java, actions/setup-go, and the like for other tools.

With AI writing more of the code, CI runs more every month, and per-minute billing grows with every run. So Skipbox chose a different model on purpose: you rent a VM for your account alone, by the month. GitHub doesn't bill self-hosted runner minutes, so however many jobs you run, the price stays a flat $39 a month for S. Fluxenic, the company behind Skipbox, runs servers from low-cost cloud providers and its own, and serves them as VMs of a useful size. Skipbox does the tedious part for you: it registers the runners, cleans up after every job and checks the machine's health. It also rebuilds a broken machine and keeps the image current. At full use, the price is 1/12 of GitHub's and 1/8 of Blacksmith's.

Because the machine is your account's alone and keeps running between jobs. Docker layers and package caches from your last job are still on its disk, so a build that uses them doesn't fetch or rebuild them. A GitHub-hosted job starts on a new VM with an empty disk every time. Even S keeps its caches on its own 80 GB disk. That's over 7x the cache GitHub gives a repository by default, and nothing in it expires after 7 days unused. S also has 4 vCPU, more than the 2 cores of ubuntu-latest on a private repository.

Rent a VPS of the same size from a major host and you still won't get a self-hosted runner that is cheaper, faster and less work than Skipbox. Skipbox registers a fresh runner for every job and removes it afterwards. After every job it wipes the runner's credentials and keeps your checkout and build output for the next one. It checks the machine's health, rebuilds it when the check fails and keeps the image current. Code from people without write access (such as pull requests from forks) runs on a throwaway VM. You get all of that for $39 a month on S, by changing one runs-on line, with nothing to maintain.

Yes, public and private repositories alike. Code from people without write access (such as pull requests from forks) runs on a fresh VM that reads your caches but can't change them; your own machine runs only your team's code.

Yes. Plans cost $39, $79, or $149 a month, billed in US dollars monthly, or yearly upfront for the price of 10 months, with no usage charges. A plan limits how many jobs run at once and how big each is, not minutes: a job that doesn't fit waits until a running one ends, and waiting adds nothing to your bill. Your organization's page in Skipbox shows how many jobs are waiting and how long the oldest has waited. A job can run for up to 6 hours.

Concurrent jobs are counted in 4 vCPU jobs. A plan runs jobs up to its vCPU total at once (S 4, M 8, L 16 vCPU), so a job of 8 vCPU counts as 2, and one of 16 vCPU as 4. The largest single job a plan can run is its vCPU count: 4 on S, 8 on M, 16 on L. Add plans to run more jobs at once.

Each Skipbox S job gets 4 vCPU and 16 GB; GitHub's ubuntu-latest on a private repository is 2 cores and 8 GB. Your own machine also keeps Docker layers and package caches from your last job, so a build that uses them doesn't fetch them again; a GitHub-hosted job starts on a new VM every time. docker build and docker compose build reuse the layers. The default builder of docker/setup-buildx-action starts empty in every job, so set driver: docker on it to use the machine's own cache. After your first job runs on Skipbox, the dashboard shows its time beside its median on GitHub over the last 30 days, when GitHub ran it then.

16 GB on plan S, on your own machine: 2× the 8 GB of ubuntu-latest on a private repository, and the same as GitHub's 4-core larger runner. Plan S used to have 8 GB, and that was not enough for openai/codex (rust-v0.159.3): its cargo clippy --workspace --tests got through only by pushing about 14 GB into compressed swap. On today's 16 GB the same build swaps about 20 MB. Plan L has 32 GB for a job that needs more.

Only what it needs to register runners and check jobs. On an organization, the Skipbox CI app asks for three permissions: Self-hosted runners (write), to register a runner with GitHub for each job and remove it when the job ends; Actions (read and write), to see which jobs are waiting, check their status, read how long your jobs ran in the last 30 days for your quote, and re-run a job whose Skipbox machine stopped mid-job; and Metadata (read). The Skipbox CI app never edits or pushes your workflows, and it can't read your code: each job checks out your code itself, on the VM it runs on. On a personal account, GitHub registers a runner for a repository only with the Administration permission, which also covers the repository's settings, collaborators and deploy keys. Skipbox (personal) asks for it and uses it only to register and remove runners. One feature is the exception, and only when you install it: the dashboard's "Open a pull request" button uses a separate app, Skipbox Migrate, which you install only on the repositories you pick. There it asks for write access to Contents, Workflows and Pull requests, pushes one branch (skipbox/run-ci-on-skipbox) that changes only runs-on lines, and opens a pull request from it. It never merges or touches your default branch. When you sign in, Skipbox also reads your email address, for trial and billing notices. As on any runner, a job's VM holds your workflow's secrets and GitHub's token for that job, so code the job runs can read them.

On your organization's own machine: a VM of your plan's size (during the trial too), one more with each extra plan. No other organization uses it. It keeps running between your jobs, so Docker images, caches, your checkout and its build output (Cargo's target/, and Turborepo's cache with turbo 2) stay for your next jobs. It runs on servers that Fluxenic, the company behind Skipbox, rents from cloud providers or runs itself. The privacy policy lists each provider and country. After each job, the runner's credentials, the git credentials in your checkout and common credential files are wiped. Jobs beyond your own machines, code from people without write access (such as pull requests from forks) and every job on Clean every job run in a new VM, deleted with its disk when the job ends. A job still running after 6 hours is stopped. Nothing can connect in from the internet. Outbound mail (TCP port 25) is blocked. The rest of the internet is open, as on GitHub-hosted runners. Traffic isn't charged.

Yes. Press Reset machine in Plans and billing, any time. Turn on Clean every job there, and each job runs on a new VM.

Cancel anytime; your plan runs to the end of the month, or the year, you paid for. If your jobs aren't faster on Skipbox than they were on GitHub, you get everything you paid back, on monthly and yearly plans alike: ask on your account's Plans and billing page within 30 days of your first payment. Skipbox takes each job's median time on Skipbox and on GitHub (the 30 days before you installed Skipbox), weights each job by how often it ran on GitHub, and compares the two totals; a job counts once it has run at least 3 times on each, and with no job to compare there is nothing to refund. If the first job you run after subscribing doesn't start within 10 minutes, not counting time it waits behind your own jobs, we refund your payment in full automatically and end the plan. Otherwise a paid month or year isn't refunded, except in the cases the Terms list.

Nothing is charged: the trial takes no card. It starts at your first job on Skipbox, not when you sign in or install the app, and runs one 4 vCPU · 16 GB job at a time for 7 days or 3,000 minutes, whichever comes first. There's 1 trial per account and 1 per person. When it ends, new jobs wait on GitHub until you choose a plan.

Change runs-on to skipbox to move, and back to ubuntu-latest to go back. Sign in with GitHub, install the Skipbox CI app on your organization or Skipbox (personal) on your personal account, change the job's runs-on line to skipbox, and commit; for a larger VM on plans that include one, use skipbox-8x16 or skipbox-16x32. To go back, change runs-on: skipbox back to runs-on: ubuntu-latest. Nothing else in your repository changed.