DockTerm

---

• 🔔 munu: a notch mascot (or pin it anywhere) that reads Claude's state and surfaces permission prompts, even over a fullscreen app on another desktop. Pick your face: munu, nvurd, guru, or adanana.
• 🧭 Checkpoints for your Claude chat: see the prompts in the current session, jump back to them, and hand off restores to Claude's own /rewind.
• 🛰️ Live agent activity: when Claude spawns subagents, DockTerm shows the count, what they are doing, elapsed time, and finished results.
• ✂️ Send selection to Claude: select terminal text and send it into Claude as a referenced snippet, without building a separate chat UI.
• 📊 Real usage at a glance: your actual 5-hour and weekly limits from Claude itself, when they reset, and a line that shows whether you are above or below an even pace, live, in a panel, a top-bar pill and a floating widget.
• 🔍 Diff review + safe Git: see exactly what changed, then stage and commit, without leaving the terminal.
• 🗂️ Files, search, editor, MCP, skills & agents: they appear when you ask and vanish when you don't. Find any file by name or any text in any file, multi-select files, send paths to Claude, or Alt+drag real files into other apps.
• 🪟 A project per pane: a grid where each pane is a different repo; the side panels follow whichever you focus. Drag panes to reorder.
• 🔒 Local-only: no accounts, no telemetry; it never calls an AI of its own.

Running Claude Code means living next to a terminal: alt-tabbing to read a diff, to commit, to check whether it's stuck on a [y/n]. DockTerm keeps that terminal central and brings the rest to you, so you can let Claude work and actually step away. Claude Code does the work; DockTerm is the calm window around it.

It keeps your real claude. It doesn't replace it. Unlike Claude Code GUIs that swap your terminal for a custom chat UI (Claudia/Opcode), DockTerm wraps your actual session and builds views around it. Nothing to relearn, and it stays compatible as Claude Code evolves.

How it compares

	Raw terminal	Full IDE	Claude GUIs	DockTerm
Keeps your real claude session	✅	✅	❌ replaces it	✅
Highlighted diff review of Claude's changes	❌	✅	~	✅
Stage & commit without raw git	❌	~	~	✅
Tells you when Claude needs you (even fullscreen)	❌	❌	❌	✅ munu
Shows Claude subagents while they run	❌	❌	~	✅
Checkpoints for the current Claude session	❌	❌	~	✅
Send selected terminal text back to Claude	manual	manual	~	✅
Real usage limits, pace & reset times at a glance	❌	❌	❌	✅
Stays out of the way	✅	❌ heavy	~	✅
No telemetry, local-only	~	❌	~	✅

A terminal alone can't show a highlighted diff of what Claude changed, or let you commit safely, or tell you Claude is waiting while you're in another window. Opening a full IDE to review three lines breaks the flow. DockTerm sits in between.

munu

DockTerm reads Claude's state from the terminal output and shows it as munu, a small face near your menu bar (in the notch on a MacBook). At a glance you can tell whether Claude is working, finished, or waiting for a [y/n], even when DockTerm is behind another window. Sub-agents, background agents and team members show up as little munus too. When Claude pauses to ask permission, munu surfaces the prompt so you can answer with one click, including multi-choice and free-text answers, and never lose your flow. It infers everything from the terminal; it never auto-answers and never calls an API.

By default munu tucks into the notch, slides out on hover, and peeks for a few seconds when Claude's state changes. Prefer it always in view? Pin it and drag it anywhere on screen. It stays put and visible wherever you put it. On Windows and Linux it's a small auto-hiding (or pinned) pill at the top of the screen.

Choose your companion

…in every state

What you get

• Real terminal: xterm.js on a native PTY (your real shell). Tabs, splits, grids, true-color, unicode, search, and native, instant scrolling, tuned to stay smooth while Claude streams. Drag a pane to reorder the grid.
• Chat mode: press ⌘R / Ctrl Shift R on any pane to read Claude's session as a clean conversation, with a working timer, permission prompts as buttons, and a composer for attachments, smart paste, / and @ menus, history and drafts. Hold the mic to talk through Claude's own voice mode. The real terminal keeps running underneath; press the shortcut again to go back.
• Reading comfort: a docked or floating Reading view of the current conversation, a warm low-glare Reading theme, line height, letter spacing and padding controls with one-tap presets, and Zen mode (⌘. / Ctrl Shift .) that hides the top bar and side panels and leaves only your panes.
• Claude workflow helpers: one-click claude / claude --resume, a checkpoint rail for the current conversation, and a Send to Claude selection toolbar for turning terminal text into a referenced prompt snippet.
• Live agent activity: a top-bar count pill, Activity panel, and munu swarm show Claude Code subagents as they start, run, and finish, grouped by project and read from local transcripts.
• Terminal memory after quit: DockTerm restores each terminal's visible scrollback after a full app quit, so updating the app does not leave you staring at a blank shell. Use claude --resume to continue Claude's actual conversation.
• Real usage limits, live: a Usage panel, a compact top-bar pill (5h 11% by default) and a pop-out floating widget show your real rolling 5-hour and weekly percentages from Claude Code itself, with a reset countdown and warning colours. Show them as a percent, bar, ring or graph. An average pace marker (on by default, switchable) draws the point where your usage would be if you spent the window evenly: a tick on the ring and bar, a dashed line on the graph, a small arrow on the percent view. Until Claude has sent data it says "no data yet". Your own Claude settings and status line are never changed. Read locally; nothing leaves your machine.
• Diff review: see exactly what changed since your last commit, this session, or a pinned checkpoint, and open a side-by-side diff for any file before you trust it.
• Beginner-safe Git: grouped status, stage/discard, commit, push/pull, branches, with confirmations on the risky actions that show the exact command they'll run.

Split into a grid: a project per pane

Open any project: Claude in a real terminal

Your files, editor and image previews

Review what Claude changed, then commit when you're ready

MCP, skills & agents: read-only, secrets masked

Checkpoints, selections and live agents: built around your real Claude terminal

DockTerm does not replace Claude Code with a chat clone. It watches the real terminal and local Claude transcripts, then adds small workflow surfaces around them:

• Checkpoints rail follows the focused terminal's active Claude session. It lists your prompts, filters out system/tool noise, jumps back to the selected checkpoint, and opens Claude's own /rewind when you choose to restore.
• Send to Claude appears when you select terminal text. In a normal shell it uses the terminal selection; inside Claude Code it can use Claude's copied selection, so the same button still works.
• Agent Activity shows live subagents globally, grouped by project, with elapsed timers and result previews. munu can briefly reveal a small swarm when agent count changes.

See how much of your limits are left

Ten themes, light and dark

munu, pinned anywhere: reacting as Claude works

Keyboard shortcuts

DockTerm uses the shortcuts you already know from each platform's default terminal, and never steals keys the shell needs (plain Ctrl C with nothing selected and Ctrl W always go to the shell).

Action	macOS	Windows / Linux
New tab	⌘T	Ctrl Shift T
New window	⌘N	Ctrl Shift N
Close tab	⌘W	Ctrl Shift W
Command palette	⌘K	Ctrl Shift P
Open project	⌘O	Ctrl Shift O
Quick Open a file (fuzzy, Ctrl/⌘ Enter pastes its path)	⌘P	Ctrl Shift L
Find in files	⌘⇧F	Ctrl Shift F
Files / Git / Review panel	⌘B / ⌘G / ⌘E	Ctrl Shift B / G / E
Chat mode for the focused pane	⌘R	Ctrl Shift R
Zen mode	⌘.	Ctrl Shift .
Compose a long prompt	⌘⇧⏎	Ctrl Shift Enter
Split right	⌘D	Ctrl Shift D
MCP panel	⌘⇧M	Ctrl Shift M
Mini terminal	⌘J	Ctrl Shift J
Settings	⌘,	Ctrl Shift ,
Zoom in / out / reset	⌘ + / − / 0	Ctrl Shift + / − / 0
Copy / paste	⌘C / ⌘V	Ctrl Shift C / Ctrl Shift V (also Ctrl Insert / Shift Insert; on Windows, Ctrl V and Ctrl C with text selected too)
Copy the selection with the mouse

Install

macOS, with Homebrew:

brew install --cask munvard/dockterm/dockterm

Or download from Releases:

System	File
macOS (Apple Silicon)	DockTerm-<version>-macOS-Apple-Silicon.dmg
macOS (Intel)	DockTerm-<version>-macOS-Intel.dmg
Windows 10/11	DockTerm-<version>-Windows.exe
Linux (x86-64)	DockTerm-<version>-Linux.AppImage

macOS builds are signed and notarized, so they open normally. Windows builds are unsigned for now. If SmartScreen appears, choose More info → Run anyway. Installs per-user, no admin. After that, DockTerm can keep itself up to date from inside the app.

Privacy & security

DockTerm is built to be trusted with your code:

• No telemetry, no accounts, no AI of its own. It only ever runs your claude.
• Usage numbers are read from your local ~/.claude files, read-only: never message content, and nothing is uploaded.
• contextIsolation and sandbox are on; production loads over a custom protocol with a strict CSP and no remote content.
• Every IPC channel is an explicit, schema-validated verb with a sender check.
• Filesystem access is jailed to the open project (symlink-safe). Reading ~/.claude is a separate opt-in.
• Checkpoints and Agent Activity read local Claude transcript files under ~/.claude read-only; they do not modify, upload, or execute them.
• Every git call runs with core.hooksPath=, so a malicious repo's hooks can't execute.
• MCP and skill config is read-only and never executed; secrets are shown as key names only.

More in docs/SECURITY_MODEL.md.

FAQ

Does my code or any data leave my machine? No. DockTerm has no telemetry and never calls an AI of its own. Usage numbers come from your local ~/.claude files, read-only.

Do I keep using my normal claude? Yes. DockTerm wraps your real Claude Code session in a real shell. Nothing to relearn, and it stays compatible as Claude Code evolves.

Are the usage percentages exact to my plan? Yes. They are the real 5-hour and weekly numbers that Claude Code itself reports, and the reset times come with them. Until Claude has sent data you see "no data yet" instead of a guess. DockTerm never changes your Claude settings or status line to get them.

Windows says "unknown publisher." The Windows build is unsigned for now: choose More info → Run anyway. macOS is signed and notarized.

Does munu ever answer Claude for me? Never. It only surfaces the prompt; you click. It infers state from the terminal and never calls an API.

Build from source

git clone https://github.com/munvard/dockterm && cd dockterm
npm install
npm run dev        # run with hot reload
npm test           # unit tests (vitest)
npm run build      # production bundles

Requires Node 22+. Architecture notes are in CONTRIBUTING.md.

Status

Early but actively developed, used daily, and built and maintained with Claude Code itself. It's an Electron app, so the download is fairly large. macOS builds are notarized; Windows is unsigned for now. Bugs and rough edges are expected. Issues and PRs are welcome.

License

MIT. Built with Electron, xterm.js, Monaco, and simple-git.