Cilicon migrates 1.0 images to a default admin:admin guest login
🛠️ Self-Hosted ephemeral macOS CI on Apple Silicon
At a glance
- What is it?
- A macOS app that boots throwaway virtual machines and registers them as CI runners for three hosted services plus arbitrary scripts. The configuration reference is short and specific, and a handful of its defaults are worth reading before you point it at a repository.
- Who is it for?
- Use Cilicon if you need macOS runners on Apple Silicon hardware and you want the machine created and destroyed per job rather than kept warm. Three things to check first.
- Can I use it commercially?
- Yes. MIT is a permissive licence: you can use, modify and sell software built on it, as long as you keep its copyright and licence notices.
- Is it still maintained?
- Yes. The repository last received commits 62 days ago.
- What is it written in?
- Mainly Swift, according to GitHub's language statistics.
Answers come from the project's GitHub data, last synced on November 15, 2026, and from our analysis. They are not legal advice.
Star history
- traderepublic/Cilicon1,199 ★ · +9
| Repository | 2026-09-14 | 2026-11-15 |
|---|---|---|
| traderepublic/Cilicon | 1.2k | 1.2k |
9 Stars in the last 62 days, measured from the daily GitHub snapshot. Compare with other projects
Editorial analysis
A range of macOS releases is called unreliable for the SSH client
The first thing on the page is a warning, and it is narrow and actionable. There is an issue with the SSH implementation on macOS 15.0 through the 15.3 line, and the recommendation is to use 15.4 or newer for a more reliable experience. That range matters because the SSH client is the core of the 2.0 rewrite. In the previous major version, the virtual machine ran a login item script that the host triggered, and version 2 replaced that with a real client on the host that executes commands inside the machine directly. Everything after boot now depends on that connection working: the pre-run and post-run commands and the provisioner itself all run over it. So the compatibility note is not about the hypervisor, it is about the one component the whole design moved onto.
Migrating a 1.0 image means enabling SSH with a known default login
The migration notes are three items and the third is the one to read twice. The configuration schema changed slightly, and in most cases renaming one property to another is enough, so the migration is mostly a rename. When the app detects an image in the old format it offers to convert it for you. But converting requires SSH to be enabled on the guest and credentials to be set in the configuration, and the note then says you may use the default pair unless you do not. Those defaults are a conventional administrative username and its identical password, written as a placeholder in the documentation. That is fine for a machine that lives for one job and is destroyed afterwards, and it is not fine if the network configuration lets anything else reach the guest while it runs, so treat it as something to change rather than something to rely on.
The tag advice and the examples disagree about which tag to use
The important notes on choosing an image contain four warnings, and one of them is contradicted by every example in the documentation. The advice is not to use the moving tag but to pick the specific version of the toolchain you want, giving a version number as the example. Then all four provisioner examples use an image tag named after a macOS release rather than a toolchain version, and the recommended public images are named with both the system and the toolchain in them. So there are three naming conventions in play for one field. The fourth warning explains why this matters: images with newer system versions can be published carrying the same toolchain version, which means a toolchain-only tag does not uniquely identify what you will get, and upgrading may require deleting the image by hand and restarting the app.
A URL without a scheme is treated as a local path
The source field takes either a remote reference or a local one, and the disambiguation is by scheme. If you paste an image reference without the object-store scheme in front, the app assumes it is a path on the local filesystem and will go looking for it there, which produces a confusing failure rather than a download. Remote images fetched through the built-in client land in a cache directory under the home folder, and the documentation says that directory should be cleared of unused images periodically, which means the cache is yours to maintain. Configuration lives in a single file with a fixed name in the home directory of the host, not in the project being built and not in the application's own support directory, so every project on the machine shares one configuration.
Three hosted services and a shell script, with defaults on every field
The provisioner field decides who the machine registers with, and there are four options. One needs a GitHub App created with self-hosted runner permissions at the organization level plus the path to a downloaded private key, which is the heavyweight option because it means an app installation. One needs a GitLab runner token and then accepts five more settings, each with a stated default: the executor type, a maximum number of concurrent builds defaulting to one, a flag for whether to download the runner binary at all, the download location defaulting to the vendor's own bucket, and a configuration file path which, when set, tells the app to ignore the other runner settings. One needs only an agent token. And the last takes a shell block, for runners the app does not know about natively, which is the smallest configuration of the four and the only one with no external service:
source: oci://ghcr.io/cirruslabs/macos-runner:sequoia
provisioner:
type: script
config:
run: |
echo "Hello World"
sleep 10The guest is polled over SSH ten times before the app gives up
Boot is not the same as ready, and the app treats those as different moments. Immediately after the virtual machine starts it opens an SSH connection to run the pre-run and post-run commands and the provisioner, and some images take a while before the SSH service answers. A dedicated setting controls how many connection attempts are made before giving up, and the default is ten, with twenty given as the example change. There is a second compatibility knob for the same class of problem: console devices can be injected into the machine's configuration, which the documentation ties to guest agent tools that expect specific devices to be present, with one device name given as the example. Both settings exist because the guest is not yours, and neither the image nor the runner is under your control at that point.
The page documents 2.0 while the tags are on 2.4
The navigation links to a details block titled as what is new in version two, and that block is where the SSH client, the image format change and the built-in registry client are described. The tags tell a different story: the three most recent releases are in the 2.4 line, two of them on the same afternoon in December 2025, and the last push to the repository is dated 2026-09-14, roughly nine months later with nothing tagged in between. So the page you would read for migration guidance is describing a version two releases old, and the migration notes themselves have not been revisited. The last visible section of the page starts describing a liveness probe that monitors the health of the virtual machine during something, and stops there, so the probe's semantics are not documented on the page at all.
Editorial conclusion
Use Cilicon if you need macOS runners on Apple Silicon hardware and you want the machine created and destroyed per job rather than kept warm. Three things to check first. Set your own guest credentials, because the documented default for a converted image is a well-known username and password pair, and an ephemeral VM is still reachable over the network while it exists. Pin your image by content rather than by the newest macOS tag, since the documentation warns that the same toolchain version can be republished for a newer system and that a moving tag will not upgrade anything. And read the SSH retry setting, because the app connects to the guest immediately after boot and some images are not ready for it.
Frequently asked questions
What does Cilicon use the macOS Virtualization Framework for?
It creates, provisions and runs ephemeral CI virtual machines, and each provisioner registers that machine with a hosted runner service. The documented cycle is boot, provision, run the job, destroy.
What are the default SSH credentials for a converted image?
The migration notes say you must enable SSH and set credentials in the configuration when converting an older image, or use the documented default pair, which is a conventional administrative username with an identical password.
How long does Cilicon wait for the guest SSH service?
It makes ten connection attempts by default after the machine boots, before running the pre-run and post-run commands and the provisioner. The attempt count is configurable and twenty is given as an example.
Where does Cilicon keep its configuration and its images?
Configuration is a single file with a fixed name in the host user's home directory, shared by every project on the machine. Images fetched through the built-in registry client land in a cache directory under the home folder, which the documentation says to clear periodically.
Which CI services does Cilicon register with?
GitHub Actions through a GitHub App with self-hosted runner permissions, GitLab through a runner token with five further settings and their defaults, Buildkite through an agent token, plus a script option for a runner the app does not support natively.
Which macOS versions work reliably?
The page carries a warning that there is an issue with the SSH implementation on macOS 15.0 through 15.3, and recommends 15.4 or newer, which matters because the host-side SSH client replaced the previous in-machine script in version 2.
Official sources
Add this badge to your README
If you maintain this project, the badge below links readers to this analysis and shows its maintenance status from the daily GitHub snapshot. Paste the markdown into your README; add ?metric=license or ?metric=stars to the image URL for a different field.
[](https://hysenlabs.com/projects/traderepublic-cilicon)Star history chart for your README
[](https://hysenlabs.com/projects/traderepublic-cilicon)Was this useful?