tmux or Herdr?
Both keep your agents alive when the phone disconnects — they differ in what they know about those agents and what Moshi can do with that knowledge. A practical decision guide, with the honest cases for each.
Every serious Moshi setup runs a multiplexer — that part isn't in question. Anything long-running belongs inside one, or a dropped connection can take the agent down with it. The question is which one, and for agent work the field has narrowed to two serious answers: tmux, the battle-tested default, and Herdr, the multiplexer built specifically for coding agents.
The short version: they tie on durability. They differ in what they know. tmux keeps processes alive and tells you nothing about them; Herdr keeps processes alive and tracks which agent panes are blocked, working, or done — and Moshi can put that extra structure to work. Whether that's worth adopting a newer tool over a twenty-year incumbent is exactly the decision this guide walks through.
What they have in common
Worth stating, because it takes the fear out of the choice. With either one, Moshi gives you:
- Detection on connect and a session list in the picker — no typing attach commands
- A dedicated shortcut panel with a tab/window quick-access row, plus a configurable prefix key
- Swipe left/right to switch windows or tabs, and pinch to zoom a pane
- Sessions that survive sleep, dead zones, and the app backgrounding — inside mosh too
- Full moshi-hook context, so inbox events know which session they came from
- Tap-to-open from an agent event into the right session
Both default to the same Ctrl-B prefix — Herdr chose tmux compatibility deliberately — so the muscle memory transfers in either direction. You are choosing between two good options, not avoiding a bad one.
Where they differ
Three of those rows decide most cases.
Agent state. With tmux, "which of my four agents needs me?" is answered by Moshi's inbox — the kanban and notifications carry the state. With Herdr, the multiplexer itself also knows: the sidebar rolls every workspace up to its most urgent agent, so the answer is visible inside the terminal too, even for processes that never emit a hook event.
Where you land. tmux attach drops you in the session; you swipe or tap your way to the right window. Herdr attach goes through Moshi's workspace picker, which focuses your chosen workspace before you arrive — and an inbox event deep-links all the way to the agent's tab. The more parallel projects you run, the more that landing precision matters.
The launcher. tmux is the only multiplexer with the moshi <dir> attach-or-create launcher, and the only one where the picker's Recent tab can spawn a session anchored to a project directory. If your day starts with "open a shell in that project," tmux still has the smoothest cold start.
Pick tmux if…
- You already live in it. Your config, your muscle memory, your scripts — switching costs are real and the inbox already covers agent state from Moshi's side.
- You touch many machines. VPSes, containers, other people's servers: tmux is either present or one
apt installaway, and it behaves identically everywhere. It's also the most-tested path in Moshi's own detection and tooling. - You want the launcher.
moshi <dir>and Recent-directory sessions are tmux-only conveniences that add up across a day. - Copy mode matters to you. Moshi detects tmux copy mode and tunes touch scrolling around it — the strongest story for grabbing big chunks of scrollback on a phone.
The tmux guide turns this choice into a phone-shaped workspace.
Pick Herdr if…
- Agents are the point. Three or more agents across projects is where blocked/working/done in the workspace rollup, plus tab-level deep links from the lock screen, stop being nice and start being the workflow.
- You think in projects, not sessions. Workspaces anchored to directories, picked directly from Moshi's connect screen, fit "one workspace per repo" better than session-and-window bookkeeping.
- You'd rather tap than chord. Herdr's layout is mouse-aware end to end, and Moshi passes touch through — on glass, "tap the tab" beats
Ctrl-B 3. - You script your fleet. Herdr's CLI can create workspaces, run commands in panes, and wait on an agent's status — orchestration primitives tmux makes you improvise. (This is also what the fleet guide builds toward.)
The Herdr guide covers setup, the workspace picker, and the event deep-links in depth.
You can refuse to choose
Both at once on one host is a fully supported, even sensible, arrangement. Moshi detects each multiplexer independently and the picker groups their sessions under separate tabs, so nothing collides — including their prefixes, which Moshi configures per multiplexer.
A pattern that works well while you evaluate: keep daily shell work in tmux, where your config and habits live, and stand up one Herdr session for your multi-agent project. Two weeks of real use answers the question better than any matrix. Migration in either direction is shallow — same prefix by default, same durability model, and your agents don't care which one is keeping them alive.
Whichever you pick, the rest of the stack is identical: mosh for the transport, moshi-hook for approvals and events, and the inbox as mission control. The multiplexer decides how work is arranged on the host — not whether the loop works.
Where to go next
- Moshi with tmux — build the phone-shaped tmux workspace
- Moshi with Herdr — the agent-aware setup end to end
- Running a fleet of agents — the workflows this choice feeds into
- Multiplexers — the full support matrix, Zellij included