GitHub Actions Runners

Hippius runners run your GitHub Actions jobs on your Hippius account. Each job gets a fresh confidential VM that is destroyed when the job ends, so nothing is shared between jobs or accounts. Use them for builds that need more room than a hosted runner, or that you'd rather run on confidential hardware.

Open them from the sidebar: Confidential Computing → Runners.

Connect GitHub

Runners are picked up through the Hippius Runners GitHub App. You install it on a GitHub organisation or on your own user account, and link the installation to your Hippius account.

1. On the Runners page, in the GitHub card, click Connect GitHub.
2. On GitHub, choose the organisation or user account, and either all repositories or only the ones you select.
3. GitHub asks you to authorise the App too. That is how Hippius checks that you own the account the App was installed on.
4. You come back to the console, which says GitHub connected. Click Go to Runners, or wait a moment.

The connection link works for one hour, for the Hippius account that started it.

Who can link an installation. For a user account, the GitHub user who comes back must be that account. For an organisation, they must be one of its owners. If you ask to install the App on an organisation you don't own, GitHub sends the request to its owners. Once one approves it, connect GitHub again from the Runners page.

An installation belongs to one Hippius account at a time. Each linked installation is listed in the GitHub card, with Manage on GitHub and Disconnect. Click Connect another to link more.

What the App can do

Permission	Why
Administration: write (repositories)	Registers a single-use runner on the repository of each job, and removes one that never ran.
Actions: read (repositories)	Lets GitHub tell Hippius when a job asks for a hippius-* runner.
Metadata: read (repositories)	Required of every GitHub App.
Members: read (organisation)	Checks, when you link an organisation, that the GitHub user who installed the App is one of its owners.

Runners are registered on each job's repository only, never on the whole organisation.

Run a job on Hippius

Ask for a Hippius runner in the job's runs-on:

.github/workflows/build.yml

on: push

jobs:
  build:
    runs-on: hippius-medium
    steps:
      - uses: actions/checkout@v4
      - run: docker run --rm hello-world

There are three sizes. A runner has the vCPUs and memory of the VM size with the same name, and a smaller disk, which is enough for a CI job and boots faster:

Label	Resources	Price
hippius-small	1 vCPU, 4 GB memory, 20 GB disk	$30 a month ($0.0411 an hour)
hippius-medium	2 vCPU, 8 GB memory, 20 GB disk	$58 a month ($0.0795 an hour)
hippius-large	4 vCPU, 16 GB memory, 30 GB disk	$115 a month ($0.1575 an hour)

The Runners page shows the current sizes and prices.

To pin the region, add a two-letter country code: hippius-medium-fr runs only on servers in France. Without one, your Default region setting applies, or any region if it is empty.

A misspelled label is silently ignored

A job whose label isn't exactly hippius-small, hippius-medium or hippius-large, with an optional two-letter country, never reaches Hippius. It doesn't show on the Runners page, and it waits on GitHub until GitHub gives up on it, after about 24 hours. Check the label if a job sits in Queued and nothing appears in the console.

What's on the runner

• Ubuntu 24.04, x64, with the GitHub Actions runner.
• git, curl, jq, python3, unzip and Docker. Docker's data is on the VM's data disk.
• The job runs as a runner user that can use sudo without a password.

It isn't a copy of GitHub's hosted runner image: language toolchains aren't preinstalled. Use the actions/setup-* actions (actions/setup-node, actions/setup-python, …) or Docker images.

The runner VM has internet access, but no public IP address and no SSH access, and it is never added to your private network, so a workflow can't reach your other Hippius machines.

Settings

The Settings card applies to every installation you have linked. Click Save settings after a change.

Setting	Default	What it does
Run jobs on Hippius	On	Off: new hippius-* jobs are refused and stay queued on GitHub.
Runners at once	5	How many jobs run at the same time. More jobs wait for a free slot. It can't go above your runners quota.
Allowed sizes	All three	A job asking for another size is refused.
Default region	Any	The country used when a label has none. A label with a country wins over it.
Allow public repositories	Off	Whether jobs from public repositories run.

Think before you allow public repositories

Anyone can open a pull request on a public repository, and the workflow run for it would launch a VM billed to your account. That is why public repositories are off by default. Before you turn it on, require approval for workflow runs from outside collaborators on GitHub: in the repository, Settings → Actions → General.

Follow your jobs

The Jobs table lists each job with its repository, workflow, size and region, when it was queued, started and finished, its duration and an estimated cost.

Status	Meaning
Waiting	Every runner slot is taken. It launches as soon as one of your runners ends.
Launching	The VM is booting and the runner is registering with GitHub.
Running	The job is running on its VM.
Completed	The runner ran its job and the VM was destroyed. GitHub's result (success, failure, cancelled…) is shown under the status.
Cancelled	The job was cancelled or ran on another runner, the runner stayed idle 15 minutes, or the installation was unlinked.
Refused	Not launched. The Reason column says why.
Failed	The launch or the runner failed, and the VM was destroyed.
Timed out	The job ran over the 6-hour limit, and the VM was destroyed.

Boot time. Expect about 3 to 5 minutes from queued to the job starting: each runner is a fresh confidential VM that boots, then installs Docker and the GitHub runner.

6-hour limit. A job can run for 6 hours. A job still running after that is stopped within a few minutes, marked Timed out, and its VM destroyed.

"Ran on another runner". A single-use runner can take any queued job with matching labels on the same repository, not only the one it was launched for. When that happens, one row ends Cancelled with that reason while another job ran. This is normal.

A job that waits 24 hours without getting a slot is cancelled.

Why a job is refused

Refusals show on the Runners page only. On GitHub, the job just stays queued ("Waiting for a runner") until another runner takes it or you cancel it. Re-running the workflow creates a new job.

Reason	What to do
Waiting for a free runner slot	Not a refusal: the job starts when a slot frees up. Raise Runners at once if you need more.
Runners are turned off in your settings	Turn on Run jobs on Hippius.
Size not allowed in your settings	Add the size to Allowed sizes.
Public repository, not allowed in your settings	See the warning above before you allow public repositories.
GitHub installation suspended or unlinked	Unsuspend the App on GitHub (Manage on GitHub), or connect it again.
Installation linked to another Hippius account	The installation was moved to another account.
Over your compute quota	Runners count against your compute quotas. Free some resources, or contact support.
Balance too low for the runway	Top up: your balance must cover 24 hours of your compute, including the runner.
Unpaid compute usage	Top up to settle what your account owes.
Runners are turned off on Hippius	Runners are paused on our side. Try again later.

Billing

A runner VM is billed per second at its size's price while the VM is billed, like any VM, with a minimum of one minute. That runs from the moment the VM is running until it is torn down, so the runner's setup (installing Docker and the GitHub runner) is billed too, not only the job itself. The Est. cost column shows each job's billed time at today's price.

Runner VMs don't appear in your VM list. They count against your GitHub Actions runners quota (5 at once by default) and against your vCPU, memory and disk quotas. The 24-hour balance requirement is checked for each job.

Disconnect GitHub

Click Disconnect on an installation and confirm. Jobs still waiting are cancelled and runners that haven't taken a job yet are stopped. Running jobs finish. The App stays installed on GitHub: to remove it, uninstall it there (Manage on GitHub).