fghj and OrbStack
OrbStack is a Docker Desktop replacement for macOS,
and it ships a feature set that overlaps fghj’s squarely: every container
gets a domain (service.project.orb.local), there’s a reverse proxy so you
don’t type a port, certificates are minted per hostname by a CA it installs
into your keychain, and its root CA is injected into containers so TLS
verifies from the inside too.
If you have read Local CA & TLS proxy and Split DNS, that list will sound familiar. The overlap is real and this page does not pretend otherwise.
What the two do differently is the shape of what gets a name. OrbStack names containers. fghj names nodes in a resolved graph — and that difference turns into five concrete capabilities below.
Where OrbStack is the better answer
Section titled “Where OrbStack is the better answer”Be clear about this first, because it’s a large share of real cases.
If what you want is one HTTP port, on one container, in one Compose
project, reached from the host, OrbStack does it with less machinery than
fghj and you should use OrbStack. It owns the VM and the network stack, so
it needs no loopback alias, no port forwarder, no /etc/resolver files and no
root daemon. It even detects your listening port by probing it, so you don’t
declare one.
fghj needs all of that machinery because it runs your containers on your
Docker daemon rather than inside a VM it controls. That’s a genuine cost:
most of fghj doctor exists to diagnose a host integration
OrbStack doesn’t have to build. fghj is also not a container runtime at all
— it has nothing to say about VM performance, file-sharing speed or
Kubernetes, and OrbStack is a reasonable thing to run fghj’s containers on.
Where they run differs. OrbStack is the VM, so on Linux there would be
nothing for it to be — Docker runs natively there. fghj was only ever
macOS-specific in its mechanisms (/etc/resolver, the keychain), and
each has its Linux counterpart now: systemd-resolved for the zones, the
system CA store and the browsers’ own databases for trust. It runs on macOS
and Ubuntu; see installation.
1. Non-HTTP services, by name, on their real port
Section titled “1. Non-HTTP services, by name, on their real port”OrbStack’s domains route HTTP. Connecting to a database by domain —
psql -h postgres.orb.local — resolves and pings but fails to establish a
TCP connection, despite the documentation saying that non-HTTP services are
reachable by host:port. The fallback is the container’s IP, which changes
when Compose recreates it, or a published host port, which is the
port-juggling the domain was supposed to remove.
fghj’s raw zone is built for exactly this. Each raw node gets its own virtual IP on a loopback alias, where fghjd listens on the port its author declared and forwards to the container, so the canonical port is the port:
psql -h db.web.storefront.shop.fghj.raw.internal -p 5432psql -h db.api.billing.shop.fghj.raw.internal -p 5432Two Postgres instances, both on 5432, no mapping table. The address is derived from graph identity, so it survives a container being recreated.
2. More than one named port per service
Section titled “2. More than one named port per service”A container that serves two HTTP things — an app port and a Vite dev server,
or an API and an admin console — can have only one OrbStack domain.
dev.orbstack.http-port is singular, and there is no syntax for binding a
second domain to a second port. The second port falls back to
https://name.orb.local:5173, which produces a certificate error.
In fghj, ports are a map and each entry can carry its own name:
services: storefront: ports: "3000": primary: true "5173": name: vite "9229": name: adminThat’s storefront.shop.fghj.internal, vite.storefront.shop.fghj.internal
and admin.storefront.shop.fghj.internal — three hostnames, three
certificates minted from the same local CA, one container. See
ports for name and wildcard.
3. Cross-repo names that work from inside a container
Section titled “3. Cross-repo names that work from inside a container”This is the one that matters most for fghj’s actual job.
Under OrbStack, each repo is its own Compose project. A container in project
A cannot reach svc.projectB.orb.local — it resolves from the host and fails
from inside a container, and OrbStack has declined this as out of scope
rather than filed it as a defect. So the multi-repo case is precisely the
unsupported one.
fghj puts a sidecar inside each run’s network that
serves DNS and terminates TLS for every node in that run, whatever repo it
came from. That’s why a storefront container can call
https://api.catalog.shop.fghj.internal and get a valid certificate, with
the same address the host uses — see
tutorial chapter 2.
4. A TLD that survives a hostile resolver
Section titled “4. A TLD that survives a hostile resolver”.local is reserved for Multicast DNS by RFC 6762, so any network is
entitled to answer for it. When an ISP or corporate DNS server does,
orb.local resolution hangs — the system resolver wins. That isn’t
straightforwardly fixable; it follows from the TLD.
fghj uses .internal, which ICANN designated for private use, and routes it
by zone: a per-zone /etc/resolver/fghj.internal file on macOS, a
routing-only domain on its own link in systemd-resolved on Linux. Either way
the OS sends the query to fghjd by zone and never to the default resolver,
so an upstream server claiming the zone never sees it.
One honest caveat on our side: fghj accepts author-declared
additional_hosts under .local as well as
.test, .internal and .localhost. A single alias is written into
/etc/hosts, which beats DNS outright and is unaffected by any of this. A
wildcard: true alias needs a resolver file, so a wildcard under .local
inherits the same exposure. Prefer .test for wildcard aliases.
5. Your own TLS, and your own certificates
Section titled “5. Your own TLS, and your own certificates”OrbStack’s proxy expects to be the thing terminating TLS. A container that generates a self-signed certificate at startup and serves HTTPS itself wins over OrbStack’s certificate, so the domain presents the untrusted one with no documented override. There’s also no way to get a leaf certificate from its CA for your own server to present, which rules out testing your app’s own TLS path or mTLS. And its port detection probes your backend with a TLS ClientHello, which makes some servers log spurious “connection that looks like TLS received on a clear channel” warnings.
fghj declares the port rather than probing it, and
its CA is on disk under
/var/lib/fghjd/ca, so a certificate for your own server is something you
can mint.
Summary
Section titled “Summary”| OrbStack | fghj | |
|---|---|---|
| HTTP by name, host → container | Yes, zero config | Yes, declared port |
| Automatic per-hostname certs | Yes | Yes |
| CA trusted by the system | Yes | Yes |
| CA available inside containers | Yes, injected into the image | Yes, via the sidecar |
| Port auto-detection | Yes | No — primary: true |
| Non-HTTP by name, canonical port | No | Yes, the raw zone |
| Several named ports per service | No | Yes, ports[].name |
| Cross-project names from inside a container | No, declined | Yes, the sidecar |
| TLD safe from upstream DNS | No, .local | Yes, .internal |
| Serve your own TLS behind the name | No | Yes |
| Container runtime | Yes, that’s the product | No |
| Platform | macOS only | macOS, Ubuntu |
| Multi-repo dependency graph | No | That’s the product |
| Licensing | Free for personal use, paid for commercial | MIT |
Reading this page later
Section titled “Reading this page later”Every “no” above describes OrbStack as of October 2026, established from its documentation and its public issue tracker rather than from a local install. It is actively developed, so treat each one as a thing to re-check rather than as a fixed property — and if you find one has been fixed, this page is wrong and worth a pull request.
Deliberately no issue links: a link to a report that has since been fixed argues the opposite of what it was cited for, and a page of them rots faster than the prose around it.
Two of the gaps look structural rather than scheduled. .local follows from
the TLD, not from a bug. And cross-project container-to-container naming was
declined as out of scope, which is a decision rather than a backlog item.
What won’t change by OrbStack shipping fixes is the shape difference. OrbStack names containers you have already defined. fghj names nodes in a graph it resolved from per-repo config across many repos, which is where items 2 and 3 come from and is what the rest of these docs are about.