A fast, beautiful, native macOS workbench for people who live on SSH.
저장소 README에서 가져옴
README
README 전체를 불러오는 중입니다. 아래는 저장된 발췌입니다.
A fast, native macOS workbench for people who live in remote terminals.
Portside is a session manager and terminal for operators who run fleets of
Linux servers — plus their containers, Kubernetes pods, serial consoles, and
legacy telnet endpoints — from a Mac. Connect over OpenSSH or mosh, plug into a
switch with a native serial bridge, and keep every target in one searchable,
foldered library. Multi-host broadcast with guardrails, SFTP, port forwarding,
and credential glue all live in the same native window.
A fast, native macOS workbench for people who live in remote terminals.
Portside is a session manager and terminal for operators who run fleets of
Linux servers — plus their containers, Kubernetes pods, serial consoles, and
legacy telnet endpoints — from a Mac. Connect over OpenSSH or mosh, plug into a
switch with a native serial bridge, and keep every target in one searchable,
foldered library. Multi-host broadcast with guardrails, SFTP, port forwarding,
and credential glue all live in the same native window. No Electron, no web
view: SwiftUI and a real terminal, built around the tools you already use.
Early days, moving fast. Built on the coast of Maine. ⛵
Install
Grab the latest notarized build from Releases, or:
brew install mcglothi/tap/portside
Builds are Developer ID signed and notarized by Apple, and keep themselves
current via Sparkle. Requires macOS 14+ on Apple silicon — Intel Macs
aren't supported. If that matters to you, please open an issue.
One foldered library for SSH and mosh hosts, containers, Kubernetes pods, serial consoles, and telnet endpoints — with transport and environment badges.
⌘K Quick Connect fuzzy-searches every saved target and recent connection at once.
Real terminal tabs over the OpenSSH you already have configured — keys, agent, and ProxyJump chains included.
MultiExec broadcasts your keystrokes to every included pane — with a loud banner counting what's live. Drop a host out for a single command and put it back (⌥⌘M, or the status bar under any pane) without disarming.
The SFTP browser rides the same SSH connection as your shell — no second login — with drag-and-drop transfer right beside the terminal.
Why
macOS has excellent terminal emulators — iTerm2 and WezTerm are superb. What
it lacks is a native operator workflow: a session library you can organize,
broadcast execution across a fleet with real safety UX, credential glue, file
transfer, and container/Kubernetes access — in one fast surface. Portside
exists to fill that gap without giving up native speed and macOS polish.
Opinionated bets:
Win on workflow density, not protocol count. SSH-first, with the console
transports operators still need close at hand.
Multi-host execution deserves safety UX as a flagship feature, not a checkbox.
Local-first, self-owned state. Your session library is a JSON file on
your disk, not a cloud sync subscription.
Lean on OpenSSH itself for transport — your ~/.ssh/config, agent,
keys, and ProxyJump chains work on day one, unmodified.
What works today
Session library with folders — organize hosts under prod, nonprod,
personal, nested as deep as you like. Everything is editable in place.
Containers & Kubernetes as first-class sessions — save a docker/podman
container, or a Kubernetes workload or pod through kubectl or oc (NKP,
GKE, OpenShift, any kubeconfig, including a per-cluster file), the same way
you save a host. It runs on this Mac or through an SSH jump host, and shows
up in the library, Quick Connect, and recents like everything else. Browse
contexts, workloads (deploy/web survives rollouts), pods and containers
from the editor so you never have to remember a churning name. When a
cluster sign-in has expired, the session says so and offers the platform's
own Sign In — Portside never holds a token.
~/.ssh/config import — seeds the library on first launch (follows
Include directives); re-import merges new hosts anytime.
Import from MobaXterm — bring over .mxtsessions / .mxtmacros files
with their folder structure intact. (Macros imported before 0.17 could pick up
literal key names — yumSPACEupdate — from MobaXterm's keystroke format.
Scripts/repair_moba_macros.py fixes those in place; re-importing does not,
since import skips macros whose name already exists.)
Export & import — back up or move your library as portable JSON;
sessions (with folders) and macros export separately and re-import into any
Portside install. Passwords stay in the Keychain and never travel.
Native terminal tabs — SwiftTerm
rendering across OpenSSH, mosh, serial, and telnet sessions. Local shells too
(⌘T). Find in scrollback with ⌘F, per-tab zoom with ⌘+/−/0. See the tested
compatibility matrix for exactly what's supported.
Terminal appearance & behavior — pick from hundreds of built-in themes and
Nerd Fonts (or import your own), set a configurable scrollback depth (default
10,000 lines), and optionally switch on GPU (Metal) rendering — all in
Settings, applied live to open terminals.
Inline images — imgcat a screenshot, a Grafana export, or a plot from a
remote host and look at it where you are, instead of copying it back first.
Sixel, iTerm2 (OSC 1337) and Kitty graphics all render; the
has the details. Try it without
installing anything: .
Roadmap
Shipped recently
✅ Configurable scrollback, GPU (Metal) rendering, a tested
terminal-compatibility matrix, native sidebar
multi-selection with drag-to-folder, session restore, and split panes.
✅ Tab bar polish — keyboard tab switching (⌘⇧[/], ⌘1–9, and a ⌘←/⌘→
alias), pane zoom (⌘⇧↵), reconnect-in-place from the "session ended" bar,
a tab context menu (rename / duplicate / close / close others), overflow
chevrons, and a configurable MultiExec alert color.
✅ Fully remappable keyboard shortcuts, cursor shape/blink, right-click
terminal copy/paste, the searchable welcome/start-page tab, app-wide and
bulk password handling, and configurable auto-update checking.
✅ SFTP browser polish — auto-refresh on host switch, delete confirmation,
a persistent drag/drop hint, and cd-following with one-click shell
integration install + shell auto-detection (bash/zsh). Plus MultiExec
arming in one step, a Grid View restore bugfix, working toolbar tooltips,
an optional auto-accept-new-host-keys toggle, arrow-key navigation in both
the welcome-screen search and the sidebar filter, and single-keystroke
session reconnect.
✅ Named credential profiles — multiple reusable identities (username +
SSH key and/or password), managed in Settings ▸ Profiles and applied in
bulk to a multi-selection or a folder, so a fleet split across a handful
of shared accounts (AD, Ansible/Nutanix service accounts, IPMI/vendor
consoles, etc.) doesn't need per-host credential entry. A host holds a
live reference to its assigned profile rather than a one-time copy, so
rotating a profile's password or key updates every host using it
immediately. The old single default password folds into this as the
first profile ("Default").
✅ Pinned favorites on the welcome/start page, shown alongside "Jump
back in" — hidden while actively searching, same as recents. Pin from a
sidebar right-click (single host or a multi-selection), a hover star icon
on each sidebar row, or a Favorite toggle in the session editor.
✅ App appearance (light / dark / follow system), independent of the
terminal's color theme; favourite macros pinned to the MultiExec bar;
browsable recently-closed tabs; ⌘W closes the tab; a "never
connected" coverage category; and inline images documented at last —
Sixel, iTerm2 and Kitty graphics all render.
✅ Host-to-host file copy — drag a file from the SFTP browser onto a
different pane and it lands on that host, in whatever directory that
pane's shell is currently sitting in. Drop it on a pane that is
broadcasting and it goes to every host in the MultiExec group at once;
files dragged in from Finder follow the same rule. The bytes relay
through a staging file on the Mac (there is no safe direct host-to-host
path without trusting one box with credentials for another), downloaded
once no matter how many hosts receive it, and uploaded to several at a
time — configurable in Settings ▸ Connection.
Next up
Nothing named. See Later for the standing list.
Later
Per-profile font/theme choices (appearance is global today).
Font ligatures are a current SwiftTerm limitation, tracked in the
compatibility matrix rather than promised here. (Shell integration / prompt
markers were previously listed here as blocked upstream. They aren't:
Portside parses OSC 133 itself from the raw byte stream, which is how command
history works — SwiftTerm never needs to see them.)
Touch ID gating for saved credentials (Vaultwarden references later)
Named / pinned layout presets
tmux control-mode (-CC) integration — native splits backed by a durable
remote tmux session
Building
Requires the Swift 6 toolchain (Xcode Command Line Tools are enough). The
package itself is still on swift-tools-version: 5.9 and builds in Swift 5
language mode — the Swift 6 concurrency migration hasn't happened yet. CI
holds the line with a per-file ratchet
(Scripts/strict-concurrency-check.sh) so the outstanding warning count can
shrink but never grow:
# Run for development
swift run
# Run against a throwaway library instead of your real one
PORTSIDE_LIBRARY_DIR=/tmp/portside-dev swift run
# Build a standalone Portside.app
Scripts/make_app.sh
open build/Portside.app
swift run otherwise shares everything with the installed app — the same
hosts, the same saved workspace, the same history — so a dev build is a small
risk to real data every time, and each launch stops to ask whether to restore
the session you left open in the other copy. PORTSIDE_LIBRARY_DIR points
it somewhere disposable instead; seeding from ~/.ssh/config is off in that
mode, so the library starts genuinely empty and your real hostnames stay out
of any screenshots.
A swift run build also has no Info.plist, so it has no Sparkle feed and
deliberately doesn't start the updater — no "couldn't check for updates"
dialog on launch, and "Check for Updates…" disables itself. Packaged builds
are unaffected: Scripts/make_app.sh writes the feed URL and public key.
Status & contributions
Pre-1.0 and evolving quickly; expect sharp edges. Issues and ideas welcome —
especially from fellow operators who live in the terminal.
Mosh roaming — opt any SSH host into mosh for sessions that survive sleep
and network changes. Portside respects your SSH alias, key, port, and saved
password during bootstrap, and falls back to SSH when mosh is unavailable.
Native serial consoles — connect directly to live /dev/cu.* devices
without spawning screen. Choose baud rate, data bits, parity, stop bits,
and flow control; logging, run-on-connect, and MultiExec work on the same path.
Telnet for legacy endpoints — save host and port, handle RFC 854 option
negotiation cleanly, and use the same terminal, logging, and MultiExec tools.
Telnet sessions carry a prominent UNENCRYPTED badge.
Field filters — the sidebar filter takes env:prod, kind:k8s,
folder:lab, profile:none, is:fav, - to exclude and /regex/; the
? beside the field lists them all, typing env: suggests values,
non-matching hosts dim rather than vanish, and filters can be saved.
ssh:// and portside://connect/<name> links — open hosts from a wiki,
dashboard or script. Unknown hosts always ask first; Settings ▸ Connection can
make Portside the default ssh:// handler.
Quick Connect (⌘K) — a fuzzy-search command palette over the whole
library; empty query lists recent hosts so it doubles as fast reconnect.
MultiExec — tile every session in a grid and type into all of them at
once. Arming it is one step: with several separate tabs open it gathers
them into Grid View automatically instead of requiring that first. A
broadcast command bar for deliberate one-shot commands, and a loud orange
banner so you always know when you're armed and how many panes are live.
Every pane carries a Broadcasting / Excluded status bar so you can drop
a host out for one command and put it back (⌥⌘M), with Include All
(⌥⌘A), Exclude All (⌥⌘E) and Invert Selection (⌥⌘I) one click or
one key away in the banner.
Environment badges & protected hosts — tag sessions prod / staging /
dev / personal for color-coded badges in the sidebar and tabs. Protected
hosts stay out of MultiExec by default and require explicit
confirmation to join a broadcast.
Macros — named command sequences, run in the active terminal or across
the whole MultiExec grid. Per-host run-on-connect commands too.
SFTP browser — per-session remote file pane riding the same SSH
connection (no re-auth), with drag/drop upload and drag-out download (a
persistent hint keeps that discoverable), automatic refresh when you
switch hosts, and a confirmation before deleting anything. It follows
cd in the live shell for bash/zsh (via the OSC 7 "current directory"
escape sequence) — one-click "Install Shell Integration" appends the
needed one-liner to the host's .bashrc/.zshrc over ssh (idempotent,
with an option to source it into the session immediately), and
auto-detects which shell you're running so you don't have to know.
Inventory coverage — for a large imported library, a view of what
Portside doesn't know about your fleet: hosts with no environment tag, no
credential profile, or no stored credentials, each fixable in bulk from the
same place. Deliberately framed as coverage rather than errors — a host
authenticating through ssh-agent isn't misconfigured, and the view says so.
Bulk classification — tag environment (prod/staging/dev/personal) or
apply a credential profile across a multi-selection or a whole folder.
History — per-host connection totals are kept automatically and rank
Quick Connect by frecency, so a host you use constantly outranks one touched
once yesterday; hosts you haven't touched in months surface in Coverage.
Optionally, a full log of every connection attempt with its outcome, and
(opt-in) a record of every command run with its timing and exit code, taken
from OSC 133 shell integration. Selecting a command shows the surrounding
session transcript, so history reads as a table of contents for your logs.
All of it is governed from one Settings ▸ Recording pane with a single
privacy rule.
Edit remote files in your own editor — double-click a file in the SFTP
browser and it opens in whatever app you'd normally use, with every save
uploaded straight back to the host. It's a private local copy rather than a
live mount, but the round trip is invisible: no manual download, edit,
re-upload. The browser shows exactly what's checked out and when it last
saved, and a protected host asks before the file is opened.
Per-host connection options — agent forwarding, X11 forwarding and
keepalive, each deferring to ~/.ssh/config until a host sets it. X11 is
trusted (-Y) or untrusted (-X) per host, and needs XQuartz
(brew install --cask xquartz); hosts with X11 on say so if it's missing or
not running.
Port forwarding — saved -L / -R / SOCKS tunnels with live status,
start/stop, and launch-at-startup, tunneled through any host in the library.
Best-effort by design: the status tracks the ssh process, not whether
traffic still flows, and nothing restarts a tunnel that dies. See
docs/port-forwarding.md — worth two minutes before
you depend on one.
A welcome screen that searches — the tab bar's + button opens a
"welcome aboard" tab with a live host search bar, arrow-key navigation and
Enter to launch, focused automatically so you can type right away (picking
a host or starting a local shell takes over that same tab); the
whole-window empty state and the "jump back in" recent-connections list
work the same way. The sidebar's own host filter supports the same
arrow-key navigation into its results.
Tabs, tuned for a lot of them — right-click a tab to duplicate it (same
host(s)/split layout, fresh sessions) or reopen the last one you closed
(⇧⌘T); an overflowing tab strip grows </> chevrons to page through it
instead of requiring you to know a keyboard shortcut. A dropped session
reconnects with a single keystroke (R) from its "session ended" bar.
Every shortcut remappable — Settings → Shortcuts lists every keyboard
shortcut with a click-to-record rebind, conflict detection, and per-row or
one-click reset to defaults. New default shortcuts: Reopen Closed Tab,
Toggle MultiExec, Toggle Grid View, Clear Buffer, and the MultiExec pane
controls (Toggle Pane ⌥⌘M, Include All ⌥⌘A, Exclude All ⌥⌘E, Invert
Selection ⌥⌘I).
Cursor styling — block, underline, or bar, with independent blink,
in Settings → Appearance with a live preview.
Copy/paste from a right-click — a context menu in every terminal,
alongside the existing ⌘C/⌘V.
Session logging — per-host log folders with compression and search.
Keychain passwords — per-host saved passwords supplied to ssh
automatically; nothing ever lands in the JSON library. An app-wide default
password (Settings → Connection) covers hosts that opt in to saving a
password but don't have one of their own yet — pairs with the sidebar's
bulk "Save Password in Keychain" action for a multi-selection of hosts, so
a big batch of imported hosts sharing one login doesn't need per-host
passwords typed in one at a time.
Auto-updates — Sparkle-powered in-app updates from GitHub Releases,
with the check frequency and on/off switch configurable in
Settings → Updates.
Host key handling — an optional "automatically accept new host keys"
toggle (Settings → Connection) skips the first-connection prompt without
weakening protection against a known host's key changing later, which
still hard-fails exactly as ssh normally does.
✅ Changelog on update — the Sparkle update prompt now shows a
cumulative changelog covering everything since whatever version you're
updating from, not just the latest release's own notes, since releases
ship fast enough that auto-updaters often jump several versions at once.
✅ Agent Access — Claude Code, Codex or your own scripts can list hosts,
open sessions, and (with a second switch) type into and read them, through
the portside command or as MCP tools (portside mcp). Every program is
approved by name; protected hosts and multi-line input ask every time;
commands are staged for you to run unless the agent is told otherwise; an
agent never arms MultiExec or types at a password prompt. Optional, scoped
"Don't Ask" mode for trusted setups. See docs/agent-api.md.
✅ Shared inventory — subscribe to a team's hosts, and the containers
on them, in a git repo, shown read-only beside your own, with your own
settings layered on top. Plain
git, fast-forward only, manifest sanitized. See
docs/shared-inventory.md.
