Your agents. Your mux.
Our window.
Chat, diffs, code comments, and previews on top of the tmux or Herdr sessions you already run. 13 MB, no account, no relay — and nothing to migrate if you ever leave.
Moshi GoAlpha
GPU-rendered. Lighter, faster, still early.
Previous app · v0.4.23
Prefer a terminal? Install with curl on any OS.
The easiest app to uninstall.
Drag Moshi to the Trash. Your agents keep running in tmux or Herdr — mid-turn, exactly where they were. tmux attach from any terminal and carry on. No export, no migration, nothing to rescue.
Agent IDEs keep your sessions in their own daemon and reach you through their own relay. Moshi keeps nothing: the multiplexer holds your work, and it was there before us.
$ mv /Applications/Moshi.app ~/.Trash$ tmux attach -t work✳ Claude Code · still working · 14m Editing src/lib/workspaces.ts… Nothing lost. Nothing to migrate.
Everything you'd miss in a plain terminal.
- 01 · Chat
One chat UI for every agent.
Claude Code, Codex, OpenCode and 14 more look and work the same: messages, tool cards and approvals in one layout. Switch agents without learning a new screen. It reads the live terminal session, so nothing is copied or replaced.
- 02 · Term
One click back to the real TUI.
The chat is a view of the session, not a copy. Flip to Term and you’re in the same Herdr tab or tmux window, with full scrollback, typing straight into the agent. Each agent remembers which view you used last.
- 03 · Servers
Every machine in one sidebar.
Your laptop, the GPU box and staging, each with its workspaces and agents, and which ones need you. Add a host by typing any SSH destination. If
ssh hostworks in your terminal, it works here. - 04 · ⌘K
Jump anywhere.
Tabs, panes, previews and simulators on every machine, one ⌘K away.
- 05 · Previews
Remote dev servers, at localhost.
When an agent starts a dev server, Moshi detects it and tunnels it over your SSH connection, so it opens at
localhostbeside the chat. iOS Simulators and Android emulators show up there too, streamed live by Mata. - 06 · Files
Read anything on the box.
Browse any folder on the host. Code is highlighted and Markdown is rendered. PDFs, images and video open right in the app, over the same SSH connection.
- 07 · Annotate
Comment on anything. The agent gets it.
Press + on a line in a diff or a file, or pin a note to an element on a preview page. Your notes go to the agent with your next message, along with the file and line.
Moshi Desktop is the IDE that weighs nothing 🪶 — 13 MB, no projects to set up, no gigabytes to index.
A thin20% layer that does the big lift80%
Your agents already live in a multiplexer — Herdr or tmux, with Rex next once it ships. They stay there. Desktop doesn't move them, own them, or replace the TUI you already know.
moshi-hook is a thin layer on top: small web APIs for workspaces, transcripts, a live PTY, approvals, dev-server ports. Just enough to add what a terminal can't quite do — a chat view, diffs, previews, one-tap approvals. No daemon of its own, and Desktop is a native app drawn on the GPU, with no browser engine inside. Prefer no app at all? Run moshi and a web UI opens in your browser.
It's the 80/20 rule, on purpose: 20% of the surface for the 80% you actually reach for. No editor, no indexer, no agent framework of our own — that restraint is why it stays at 13 MB.
Two commands. Everything you run shows up.
- 1InstallOne line, or the Mac app. Already use moshi-hook for the phone? You have it.
- 2Run
moshiIt opens on your machine. No account, no pairing code, no project setup. - 3Pick up where you areYour tmux and Herdr sessions and their agents are already in the sidebar.
$ curl -fsSL https://getmoshi.app/install.sh | sh $ moshi → opens http://127.0.0.1:24544 in your browser
Step-by-step setup — server vs. client, agent hooks, remote hosts — in the Desktop install guide.
Against the agent IDEs, honestly.
The things everyone asks
Nothing, on your machines. Your agents run in tmux or Herdr, not in Moshi, so they keep going mid-turn. Attach from any terminal and carry on. There is no session format to export and nothing to migrate.
No. Keep your agents, your multiplexer, your SSH keys, and your editor. Desktop reads what is already running and adds chat, diffs, comments, and previews on top.
Yes. Everyone who bought the Moshi app funded this — so Desktop is free, for everyone. Thank you.
Yes. Desktop talks to a gateway that binds to 127.0.0.1 only; remote machines are reached over your own SSH connection — your terminal, transcripts, diffs, and previews never touch a relay. One honest caveat, on the phone side: the app's inbox, push notifications, and usage snapshots pass through Moshi's server, because Apple and Google only deliver push (and silent push) from a server, not from your dev box. Desktop needs none of that.
Yes — Moshi tunnels detected dev-server ports over the same SSH connection automatically. They show up in the Web tab; open one beside your session and pin comments to its elements for the agent.
Both get the same native app as macOS: a Windows installer and a Linux .deb. Or skip the app: run moshi and a web UI opens in your browser. On Windows, moshi-hook runs natively (experimental) or under WSL2.
No, though everything that holds your work is: tmux, Herdr, OpenSSH, and Mosh. The app stays closed so the time goes into building it rather than managing a project around it — and since your sessions never live inside Moshi, you are free to leave whenever you like.