# inbuxa-installer
> [!NOTE]
> Development happens on [git.coffeylabs.org/inbuxa/inbuxa-installer](https://git.coffeylabs.org/inbuxa/inbuxa-installer); the copy on GitHub is a read-only mirror.
> Report issues at **[git.coffeylabs.org/inbuxa/inbuxa-installer/issues](https://git.coffeylabs.org/inbuxa/inbuxa-installer/issues)**, join discussions at **[community.coffeylabs.org](https://community.coffeylabs.org)**, or chat on **[Discord](https://discord.gg/nqcY4TKfAn)**.
One program that installs the **inbuxa** suite on the machine you run it on:
the mail server, the administration console and the webmail, each as a
container or on the host, in any mixture.
inbuxa the interface: all of the below, a screen at a time
inbuxa survey what this machine is, as the installer sees it
inbuxa install --dry-run … what would happen, before any of it does
inbuxa install … do it
It only ever installs here. A second machine runs it too, and `inbuxa join`
points that machine at a server already running elsewhere.
Linux only, and it says so on any other system rather than reporting a
machine that does not exist. Debian, Red Hat and Arch families, and the
derivatives people run: Ubuntu, Fedora, Rocky, CentOS, CachyOS. It uses
whichever container runtime the distribution ships -- docker where there is
one, podman on the Red Hat family -- and refuses a shape the machine cannot
deliver, with the reason.
## What works today
This is early. What is built:
- **`inbuxa`** on its own (or `inbuxa install` with no options) -- the
interface: six screens, always in the same order. What it found on this
machine; the names; which components go here and how, with what the
machine can't do greyed out and the reason beside it; what answers on 80
and 443; the whole plan, with the command that does the same without the
interface; and the install, ticking off its steps, ending on the
credentials and addresses, which stay in the terminal afterwards. Nothing
happens until you answer y. It fills in the same options the flags do --
it is not a second installer -- and it is built on the standard library
alone, like the rest. Without a terminal it says which flags to use.
- **`--proxy snippets`** -- for a machine that already runs nginx or Caddy:
nothing here takes 80 or 443, and the installer writes
`/proxy/nginx-inbuxa.conf` with `/proxy/nginx-inbuxa-stream.conf`,
and `/proxy/Caddyfile.fragment`, for that web server to load, trusts
it in front of the suite (its X-Forwarded-For, and an exemption from the
auto-ban), and leaves loading and reloading to you. nginx routes 443 by
name: it terminates the console and the webmail with your certificates and
passes every other name, untouched and with the PROXY protocol header, to
the mail server's own HTTPS on 127.0.0.1:8443. So the server serves its own
certificates for every domain it has, a domain added later included, and
its `_443._tcp` TLSA records match. Caddy can't route TLS by name without
a plugin, so its fragment terminates the server's names itself. An install
made before nginx worked this way keeps its layout until `install` is run
again with `--key`. Once it is loaded, the mail server's own certificates
are turned on in the console. `--proxy none` writes nothing.
- **`survey`** -- the machine's facts: distribution, init, whether Docker is
usable *by this user*, Node's version, which of the suite's ports are free
and what holds the ones that are not, memory, disk, and whether an install
is already recorded here. It changes nothing.
- **`install --dry-run`** -- the whole plan: every file, unit, container,
port, DNS record and credential, and a refusal with a reason when the
machine cannot carry out what was asked.
- **`deps`** -- what a shape needs that this machine has not got, and, with
`--install`, the doing of it: the Docker daemon from the distribution's own
archive, the Compose plugin and Node from their official builds, both
pinned by version and checked against a checksum in the source before
anything is put in place. `install --install-deps` does the same as part of
a run. A missing dependency is an offer, not a refusal.
- **`install --yes`** -- carries the plan out: writes the deployment,
fetches the images, brings the mail server up in bootstrap mode with a
credential that exists only for that step, completes bootstrap, brings the
rest up without it, exempts the front ends from the auto-ban, creates the
first mailbox, writes `credentials.txt` and `dns.zone`, and then checks
that all three answer.
- **`install --server host`** -- the same, with the mail server as the
release binary under systemd instead of a container, checked against a
checksum in the source. It gets a user of its own, its configuration and
data in `/server`, and a unit, `inbuxa-server.service`. During setup
it is reachable from the machine itself and nowhere else; afterwards its
plain HTTP is on loopback for the proxy and the webmail, 443 is left to
the proxy, and the unit will not start without a configuration file.
- **`--console host`** -- the console's files from its release, checked
against a pinned checksum, unpacked into `/console` and pointed at the
server; the managed Caddy or your own web server (`--proxy snippets`)
serves them. Nothing runs, so there is nothing to start.
- **`--webmail host`** -- the webmail from its release, checked the same way,
in `/webmail`, owned by root and only read by the webmail, which runs
as its own user under `inbuxa-webmail.service` on the installer's own
Node 26 (in `/opt/inbuxa/node`), on 127.0.0.1:8080. With all three on the
host and `--proxy snippets`, no container runtime is installed at all.
- **`join URL --key KEY`** -- the console and the webmail on this machine,
for a mail server already running at URL on another one. It reaches that
server only over its own HTTPS, with an API key a server administrator
made in the console (`--key -` reads it from standard input, and
`INBUXA_JOIN_KEY` keeps it out of the shell's history; it is never written
down). Through it the front ends' addresses go into the server's
`inbuxa:FrontEnds`, which registers their sign-in and lets the console
call the server from the browser, with no restart and nothing changed on
that machine by hand. It asks first, dry runs included, and refuses with
the reason when the key is wrong, the server is too old, or it already
lists other front ends (`--replace` to put these in their place). Each
front end goes in a container or on the host (`--console host`, `--webmail
host`), as with `install`. Needs a server release 2026.10.12.1 or newer.
The interface does the same: skip the server on the Components screen and
give its address and the key.
- **`plan -f` / `apply -f` / `export`** -- a whole installation described in
one file, however many machines. Each machine acts on its own part and
prints the command to run on the others; it never reaches them. `plan`
diffs the file against what is actually installed here and changes
nothing; `apply` converges to it, adding and removing components;
`export` writes the file from what is already here.
- **`upgrade [server|console|webmail]`** -- moves what is installed here to
the releases this installer installs; a newer installer brings newer
releases. `--dry-run` shows each component's move first (so does
`status`). The server goes first, then the console, then the webmail,
each checked once it is running again and put back as it was if not,
which stops the upgrade there: a container by its old image tag, the
server on the host by its old binary (kept as `inbuxa-server.`),
the console and the webmail on the host by their old directory. Before
the server, which may convert its store on first start, it is stopped and
its own data archived into `/backups` (`--no-backup` to skip; a store
kept elsewhere, PostgreSQL or S3, is backed up where it lives).
- **`uninstall [--purge]`** -- takes away what the installer put on this
machine: containers, units, binaries (and the old ones kept for rollback),
the console's and the webmail's files, the webmail's user, its Node. The
data stays -- the server's volumes or directory, Caddy's certificates, the
credentials and secrets -- with the install record marked removed, so
`install` run again with the same options brings it all back. `--purge`
removes the data too, all but `/backups`, which it names for you to
delete. Docker, and the configuration your own web server loads, are left
alone. To take one component off, run `install` again with it skipped.
- **`certs [--key KEY]`** -- the mail server's own certificate, the one its
mail ports present: who issued it, until when, for which names, whether
it verifies, and where the mail host resolves. Reading it needs nothing.
Fixing it needs an API key made in the console, as `join` does (`--key -`
or `INBUXA_API_KEY`; never written down): a new ACME order, which the
server doesn't start again by itself once one has failed -- typically
because DNS wasn't in place during the install -- or, on a domain not yet
on ACME, ACME turned on where the managed Caddy or your own web server
passes the challenges. It waits for the mail ports to present the new
certificate. The console's and the webmail's certificates are the
proxy's, which renews them itself.
- **`status`** -- what is installed, what is running, and where those two
disagree: a container stopped by hand, or a deployment nothing recorded
installing. It exits non-zero when they disagree, so a machine can be
asked in a script whether it still matches itself.
- **`relay install --token -`** -- inbuxa Relay, on a machine of its own in
front of the server: the release's static `inbuxa-relay` binary, checked
against a checksum in the source and against the inbuxa release keys'
signature, owned by a user of its own so an upgrade from the console can
replace it, under `inbuxa-relay.service` with ports 25 and 443 and nothing
more. It joins with the token the console shows when you add the relay
(`--token -` reads it from standard input, `INBUXA_RELAY_JOIN_TOKEN` works
too), and refuses on a machine that runs the mail server or where port 25
is taken. `--link-listen` puts the link on one address, such as a WireGuard
one. Run again, it converges, and leaves a newer binary from the console
alone. `relay status` is the relay's own account of itself; `relay
uninstall` keeps its state, and `--purge` resets it, which it refuses
while it holds mail unless `--discard-held`.
- **Moving the server** -- `install` run again with the other `--server`
moves the mail server between a container and the host, with its
configuration and mail. It needs an API key made in the console (`--key -`
or `INBUXA_API_KEY`; never written down), because the server's listeners
and its auto-ban exemptions change with the shape. Everything here stops
for a minute or two. The data is archived into `/backups` first and
then copied, never moved: the old copy -- the volumes, or `/server` --
stays where it was until you delete it, and a failure on the way puts the
server back as it was. The interface doesn't offer it yet: it has no key
to give.
The design is in the inbuxa specification (§6.1 and the installer draft);
the phases are there too.
inbuxa install --local --domain example.test --install-deps --yes
is the shortest thing that works today: the whole suite on loopback, with no
DNS and no certificates, on a machine that starts with nothing. Without
`--local` it takes the real ports, puts Caddy in front and obtains
certificates -- which `e2e/cases/install-public.sh` proves against a private
CA, with no internet and no public name involved.
## Download
Each release has static binaries for linux/amd64 and linux/arm64, and a
SHA256SUMS to check them against:
V=v2026.10.15.2 # the newest: https://git.coffeylabs.org/inbuxa/inbuxa-installer/releases/latest
curl -fLO https://git.coffeylabs.org/inbuxa/inbuxa-installer/releases/download/$V/inbuxa-linux-amd64
curl -fLO https://git.coffeylabs.org/inbuxa/inbuxa-installer/releases/download/$V/SHA256SUMS
sha256sum --check --ignore-missing SHA256SUMS
install -m 755 inbuxa-linux-amd64 /usr/local/bin/inbuxa
SHA256SUMS is signed with [minisign](https://jedisct1.github.io/minisign/), so
you can confirm it came from us before trusting the checksums. `SHA256SUMS.minisig`
is the signature:
curl -fLO https://git.coffeylabs.org/inbuxa/inbuxa-installer/releases/download/$V/SHA256SUMS.minisig
minisign -Vm SHA256SUMS -P RWS1yFYwVuJ7MGJTZXLbhv9+hrEsjrC8Zg+7WmPlM86Sjy0qF/xkoCKq
The same public key signs every inbuxa release (installer, migrate and cli).
A release is a tag on main named with the INBUXA suite's release number:
`v27.1` for a series, `v27.1.3` for a patch, `v27.1-rc.1` for a release
candidate (before 27.1 the tags were dated, `v2026.10.15.2`).
`.gitea/workflows/release.yml` builds it with `scripts/build-release.sh`, which
builds the same thing locally. A release candidate is published as a
pre-release, so `releases/latest` stays on the last final release.
## Building and testing
go build ./cmd/inbuxa
The installer writes units, creates users and takes ports 25 and 443, so it
is tested on a throwaway virtual machine rather than on anybody's desk:
e2e/vm/up.sh a Debian 13 machine, in qemu, as you
DISTRO=fedora e2e/vm/up.sh or fedora, rocky9, ubuntu2404, arch, debian12
e2e/vm/run.sh e2e/cases/survey.sh what it says about a machine
e2e/vm/run.sh e2e/cases/deps.sh the offer, and taking it
e2e/vm/run.sh e2e/cases/install-local.sh a whole suite, and signing in to it
e2e/vm/run.sh e2e/cases/install-public.sh the same with real ports and certificates
e2e/vm/run.sh e2e/cases/install-host.sh the server on the host, the rest beside it
e2e/vm/run.sh e2e/cases/join.sh front ends joined to a server elsewhere
e2e/vm/run.sh e2e/cases/join-host.sh the same on the host, behind nginx, no containers
e2e/vm/run.sh e2e/cases/interface.sh the interface, driven key by key in a terminal
e2e/vm/run.sh e2e/cases/proxy-snippets.sh behind nginx, from the configuration it writes
e2e/vm/run.sh e2e/cases/webmail-host.sh the webmail on the host, and signing in to it
e2e/vm/run.sh e2e/cases/all-host.sh everything on the host, behind nginx, no containers
OLD_REF=v2026.10.12.1 e2e/vm/run.sh e2e/cases/upgrade-containers.sh
an older install moved to these releases
OLD_REF=v2026.10.12.1 e2e/vm/run.sh e2e/cases/upgrade-host-server.sh
the same on the host, and a bad release put back
CONSOLE_TARBALL=… WEBMAIL_TARBALL=… e2e/vm/run.sh e2e/cases/upgrade-frontends-host.sh
the console and the webmail on the host, moved on
OLD_REF=v2026.10.12.3 e2e/vm/run.sh e2e/cases/rerun-host-server.sh
install run again: restarted when it must be, only then
e2e/vm/run.sh e2e/cases/uninstall-containers.sh uninstalled, installed again on the data, purged
e2e/vm/run.sh e2e/cases/uninstall-host.sh the same with the server and the webmail on the host
e2e/vm/run.sh e2e/cases/certs.sh a failed order, reported, then fixed with a key
e2e/vm/run.sh e2e/cases/move-server.sh the server into the host and back, keeping its mail
e2e/vm/run.sh e2e/cases/topology.sh growing and shrinking from a file
e2e/vm/run.sh e2e/cases/status-export.sh intent against reality, and the file
OLD_REF= e2e/vm/run.sh e2e/cases/upgrade-volumes.sh
an install from before #5, upgraded
e2e/vm/down.sh remove it
`SERVER_BINARY=path/to/inbuxa e2e/vm/run.sh …` installs a server build
instead of the release wherever a case installs the server on the host (the
installer's `INBUXA_SERVER_BINARY`), for testing a server change before it
is released. `CONSOLE_TARBALL=` and `WEBMAIL_TARBALL=` do the same for the
console's and the webmail's release tarballs (`INBUXA_CONSOLE_TARBALL`,
`INBUXA_WEBMAIL_TARBALL`).
Each case starts from a copy of the machine taken when it was new, so a run
is free to break it and a failure is the installer's rather than the last
run's leftovers. `e2e/vm/up.sh` needs qemu, KVM and xorriso; nothing needs
root on your machine.
## License
AGPL-3.0-or-later. Some of this began as [ihasmail-oneshot], which is ours
and under the same license.
[ihasmail-oneshot]: https://git.coffeylabs.org/inbuxa/ihasmail-oneshot