Pick the one that matches your platform. There is nothing to sign up
for, and no installer to review.
macOS
A Homebrew cask from a tap. This is the whole install, and the same
command upgrades it later.
Homebrew
brew install tkoizumi/tap/otter
Linux
A .deb per architecture. Set
VERSION to the release you want — the
filenames carry the version, so there is no unversioned URL to bookmark.
Ubuntu / Debian
VERSION=0.5.3 # or any tag from the releases page
ARCH=amd64 # amd64 or arm64
curl -LO "https://github.com/tkoizumi/otter/releases/download/v${VERSION}/otter_${VERSION}_linux_${ARCH}.deb"
sudo apt install "./otter_${VERSION}_linux_${ARCH}.deb"
Tarball
Both binaries in one archive, for a host where you would rather place
them yourself. The archive name is a published interface.
manual install
VERSION=0.5.3
OS=linux # linux or darwin
ARCH=amd64 # amd64 or arm64
curl -LO "https://github.com/tkoizumi/otter/releases/download/v${VERSION}/otter_${VERSION}_${OS}_${ARCH}.tar.gz"
tar -xzf "otter_${VERSION}_${OS}_${ARCH}.tar.gz"
sudo install -m 0755 otter otterd /usr/local/bin/
From source
With Go 1.22 or newer. Useful if you want to read the code you are
running, or build for a platform we do not publish.
go install
go install github.com/tkoizumi/otter/cmd/otter@latest
go install github.com/tkoizumi/otter/cmd/otterd@latest
Installing by hand after using a package? Remove the
hand-placed binaries from ~/.local/bin.
A directory earlier in PATH wins, and
otter version would then disagree with
dpkg -l otter.
Verify
Check what you installed.
Both binaries report their version, and the runtime can print the
contract it implements before you commit to it.
otter --json version is the machine-readable
form: schema, contract, SDK and manifest versions in one object. A client
uses it to establish what it is talking to before it holds a credential.
verify
otter version
otterd --version
otter --json version
otter init my-project
cd my-project
otter validate my-project
otter release my-project
otter start --detach
otter run my-project
otter state get my-project count
otter stop
otter start loads
otter.env and
otter.daemon.env if they exist, and records
where it listens in .otter/serve/, so every
later command finds the runtime on its own.
Requirements
What the host needs.
Requirement
Notes
Operating system
Linux or macOS
amd64 and arm64. Windows is not published.
Python
Python 3 on the host
Unless a job sets python.mode: managed, which prepares its own interpreter and locks dependencies.
Disk
A data directory you choose
Holds otter.db, the extracted SDK and release snapshots. Defaults to ./tmp, or /var/lib/otter in a service install.
Network
None required
The daemon binds loopback by default. Binding elsewhere requires an API token.
Memory
Budget for the daemon plus your runs
Roughly 256 MiB for otterd, plus the peak RSS of each concurrent run.
Upgrading
Same command, newer version.
Releases publish otter and
otterd together from one archive. The
binary pair is the unit — do not mix versions.
Every job must be released once after upgrading: the release digest
format is versioned, and a deploy prunes inactive snapshots while always
keeping the previous release as the rollback target.
# macOS
brew upgrade otter
# Linux — download the new .deb and install over the old one
sudo apt install "./otter_${VERSION}_linux_${ARCH}.deb"
# then re-release each job once
otter release --all
Without installing
Other ways to run it.
Remote host
Deploy from your machine
otter deploy --host user@host
cross-compiles locally, pushes the binaries, creates an unprivileged
service account and installs the systemd unit. Nothing needs to be
installed on the target first except the SSH access you already have.
Otter Cloud · alpha
Skip the host entirely
otter login then
otter deploy --cloud promotes a release
you already built and validated to a pooled tenant we operate. Alpha
access is invite-based.
Every release, including the notes for the one you are installing, is on
the releases page.
The
compatibility
policy says what counts as a change to a frozen interface.
Installed. Now write a job.
The quickstart walks from an empty directory to a released, scheduled job.