tmux vs herdr
herdr is an agent multiplexer: a terminal multiplexer built around AI coding agents rather than bare shells. Where tmux thinks in sessions, windows, and panes, herdr adds first-class agents (it knows which pane runs which agent and what state it’s in), workspaces, built-in git worktrees, and persistent sessions that survive a restart — no resurrect/continuum plugins required.
I’m migrating from tmux to herdr while keeping both in parallel. herdr keeps
its default prefix Ctrl-b; tmux is on Ctrl-a, so the two multiplexers
don’t collide. That leaves Ctrl-a free for readline and for agents running
inside herdr panes (and as tmux’s prefix when one is nested). Everything below
reads prefix+X as Ctrl-b X.
Note
This page is the narrative comparison. The complete, current binding list lives on the herdr shortcuts page — treat that as the source of truth for keys.
Key differences at a glance
| Concept | tmux | herdr |
|---|---|---|
| Prefix | default Ctrl-b (mine: Ctrl-a) | Ctrl-b (herdr default) |
| Top-level grouping | Sessions → windows → panes | Workspaces → tabs → panes |
| Splits | -h side-by-side / -v stacked | split_vertical = / split_horizontal stacked |
| Agents | None | First-class: per-pane labels, state, sound |
| Worktrees | Manual / scripted | Built in (Ctrl-b Shift+g) |
| Persistence | resurrect + continuum plugins | Built in |
| Theme | catppuccin via TPM | Built in (kanagawa) |
| Extensibility | TPM plugin ecosystem | Fixed feature set + native (manifest) plugins |
| Copy / scrollback | copy-mode + plugins | edit_scrollback + built-in mouse/copy UI |
What carries over
These habits survive the move unchanged — same physical key, same intent:
| Action | tmux | herdr |
|---|---|---|
| Focus pane left/down/up/right | h j k l | Ctrl-b h/j/k/l |
| New tab/window | c | Ctrl-b c |
| Jump to tab 1–9 | 1…9 | Ctrl-b 1…9 |
| Next tab/window | n | Ctrl-b n |
| Zoom / fullscreen pane | z | Ctrl-b z |
| Split side-by-side | = | Ctrl-b = |
| Split stacked | - | Ctrl-b - |
The split keys (= / -) are preserved on purpose, even though I had to
override herdr’s defaults to get = (see below).
What’s remapped
Prefix: stays on herdr’s default Ctrl-b. herdr keeps Ctrl-b; my tmux is
on Ctrl-a, so the two don’t fight over the same prefix even when both run. That
also leaves Ctrl-a free for shell line-editing and for agents inside panes.
(I trialed F12 to keep Ctrl-b free too, but went back to the default.)
Splits keep the keys but the action names invert. tmux and herdr disagree on what “horizontal” and “vertical” mean, so I bound by physical key + intent:
| Intent | tmux key (action) | herdr key (action) |
|---|---|---|
| Side-by-side | = (split-window -h) | Ctrl-b = (split_vertical) |
| Stacked | - (split-window -v) | Ctrl-b - (split_horizontal) |
Tip
split_vertical is bound to the list ["prefix+=", "prefix+v"] — = is my
added tmux-muscle key, and herdr’s original v is kept as an alias.
A handful of management keys also moved layer:
| Action | tmux | herdr |
|---|---|---|
| Rename tab | , | Ctrl-b , |
| Rename workspace | $ | Ctrl-b . |
| Close tab | & | Ctrl-b Shift+x |
| Detach | d | Ctrl-b q |
| Reload config | r | Ctrl-b Shift+r |
| Scrollback / copy | Esc / [ | Ctrl-b e |
| Resize | H J K L | Ctrl-b r + arrows |
| Previous tab | Ctrl-a Ctrl-a | Ctrl-b p |
Ctrl-b , (rename tab) keeps tmux’s , exactly; Ctrl-b . (rename workspace) is the
adjacent key, a small mnemonic I added since tmux’s $ had no obvious herdr home.
Note r is deliberately herdr’s resize, not reload — reload moved to
Ctrl-b Shift+r.
herdr-only
Features tmux simply doesn’t have:
- Agent awareness — herdr detects the agent in each pane, labels it on the
border, tracks its state, and can notify/sort by state and resume agent
sessions on restart.
previous_agent/next_agent/focus_agentship unbound; I put them onCtrl-b {/Ctrl-b }/Ctrl-b Alt+1–9— the curly brackets step agents (plain[/]step workspaces), mirroringCtrl-b Shift+1–9for workspace jumps. - Built-in git worktrees —
Ctrl-b Shift+g(new_worktree) spins up a worktree-backed workspace; no scripting. - Workspaces —
Ctrl-b wpicker,Ctrl-b Shift+nnew,Ctrl-b .rename, a layer above tabs. - Persistent sessions — survive restart out of the box; no resurrect / continuum.
- Built-in theming —
kanagawaships in-app instead of a TPM theme plugin. - Plus conveniences like
Ctrl-b bsidebar,Ctrl-b ggoto/jump,Ctrl-b Shift+oopen the latest notification’s target, andCtrl-b ?keybinding overlay.
No herdr equivalent (dropped)
herdr has no tmux/TPM-style plugins, so the plugins I relied on have no replacement and are gone:
| Dropped | What it did |
|---|---|
tmux-thumbs (Space) | Hint-based on-screen copy |
tmux-copycat (/ n N) | Regex search in scrollback |
tmux-open (o C-o S) | Open selection / editor / web-search |
tmux-logging (P Alt-p/P) | Pane logging / screen capture |
cpu (#{cpu_percentage}) | CPU% in the status bar |
tmux-fzf (F), resurrect’s save/restore keys, mouse-toggle keys, and pane-swap
(< >) likewise have no herdr binding (persistence, mouse, and copy are handled
internally instead).
tmux’s F12 nested OFF-mode (passthrough that sends all keys to a
remote/nested tmux) is unaffected by the move: herdr sits on Ctrl-b, so F12
stays free for tmux and the two don’t collide.
Rebuilt as herdr custom commands
A few tmux workflows I wasn’t willing to lose were rebuilt as herdr custom
commands (helper scripts in scripts/dot-local/bin/). Each opens a temporary
pane that closes when the command exits, and drives herdr over its CLI / pane
API (herdr pane read / send-text, herdr workspace …):
| Key | Command | Replaces | What it does |
|---|---|---|---|
Ctrl-b u | herdr-extract | extrakto + urlview | Fuzzy-grab URLs/paths/IDs from scrollback → copy / open / insert (Ctrl-G opens a GitHub ref in ghzinga) |
Ctrl-b f | pick-files-herdr | pick-files | Pick files (fd+fzf), inject paths into the triggering pane |
Ctrl-b y | pick-files-yazi | — (yazi UI variant) | Same, but with yazi’s full file-manager UI |
Ctrl-b t | exec $SHELL | C-t display-popup | Disposable scratch shell in a temp pane |
Ctrl-b o | herdr-sessionize | sessionx (prefix o) | Fuzzy-pick a project/notes dir → open or switch a workspace for it; leads with the already-open workspaces |
Note
herdr has no floating popups; these are real (disposable) panes that
close on exit, not display-popup overlays.
Native plugins
Unlike the TPM plugins above (which have no herdr equivalent), herdr does have
its own manifest-based plugin system — these have no tmux analogue and are
installed with just herdr-plugins:
| Plugin | Key | What it adds |
|---|---|---|
| herdr-plus | Ctrl-b Up / Ctrl-b Down | Project workspace templates + a fuzzy quick-action launcher |
| ghzinga | Ctrl-click / Ctrl-b u → Ctrl-G | Open a GitHub issue/PR in a side viewer pane |
| reviewr | Ctrl-b Shift+v | Code-review sidebar for the focused agent’s changes |
| vim-herdr-navigation | Alt-h/j/k/l | Seamless pane ↔ Neovim-split nav (the vim-tmux-navigator pattern) |
When to use which
- herdr for AI-agent-heavy work: anything where I want per-pane agent state, worktree workspaces, and persistent sessions doing the heavy lifting.
- tmux over plain SSH and anywhere the plugins still matter — thumbs, copycat, logging, or nested/passthrough sessions on remote hosts that don’t have herdr installed.
Both stay configured; the prefix split (herdr Ctrl-b vs tmux Ctrl-a) lets them coexist
without stepping on each other.