wmux 2.7: A Windows terminal multiplexer that watches your AI agents and guards your edits

What wmux actually solves

The README positions wmux as a visibility layer for coding agents on Windows. The problem is concrete: when you run several AI agents in separate terminals, you cannot tell which one is waiting for your input, what it has changed, or whether a file you are editing has been rewritten underneath you. wmux gives you a roster of every agent in the window, ranked by how long each has been blocked, and a file tree that shows live diffs. It is not a general terminal multiplexer like tmux for Windows. It is built for the specific workflow of supervising multiple agents at once.

The mechanism: hooks, a CDP proxy, and screen detection

wmux observes agents in three layers. For Claude Code, it auto-configures hooks in `~/.claude/settings.json` that report agent and tool activity to the sidebar. A CDP proxy on `localhost:9222` lets Claude Code's native `chrome-devtools-mcp` plugin control the wmux browser panel. For agents that report nothing, wmux falls back to screen detection: it identifies which agent a pane is running from the command line or shell, then matches the agent's on-screen UI against bundled rules to classify it as blocked, working, or idle. The key design choice is that a detected state never overrides a reported one. If an agent says it is working, that wins over what the screen looks like. And if no rule matches, the pane reads as silent, never idle, because idle is a claim no one made.

Getting it running: install, launch, and the first keybindings

The README does not give an install command, but it describes a Windows desktop application built on Electron, so the likely path is downloading a release from the GitHub releases page or building from source. Once running, the important shortcuts are `Ctrl+Shift+X` to dock the file tree on the right, `Ctrl+Shift+A` for the agent navigator, and `Ctrl+Shift+B` to jump to the agent that has waited longest. Screen detection is on by default and can be turned off in Settings → Workspace. Detection rules live in `%APPDATA%\wmux\agent-detection` if you want to add or fix one. The `wmux detect explain` command names the rule that fired and the line that matched, and `wmux detect explain --file screen.txt` replays a captured screen offline.

The file tree: a jail, not a browser

The 2.7 file tree is deliberately constrained. The renderer sends only a pane id and a relative path, never an absolute one. The main process derives the root itself, so a folder above the pane's cwd is unreachable even if the renderer asks for it. Symlinks and junctions are resolved on every path segment. This is a security boundary, not a convenience. The tree shows a `+55/-22` column per row, rolled up so a collapsed `src/` shows the sum of everything beneath it, including folders you never expanded. In a git repo the numbers mean uncommitted changes; outside one, they mean changes since the session started. The panel header says which interpretation applies. Files an agent touched get a dot, read from the hook stream wmux already receives.

Saves that refuse to overwrite: the stale-buffer guard

A save in wmux can only land on a file you opened in a pane, in this window, in this session. The save carries the timestamp the buffer was read at. If an agent rewrote the file while you were typing, the save is refused rather than quietly picking a winner. This is a strong correctness choice, but it has a cost: you must re-read the file and reapply your edit manually. The README also promises that CRLF line endings and UTF-16/BOM encodings survive the round trip instead of being normalised into a whole-file rewrite. That matters for Windows users who share files with tools that expect specific line endings.

What is missing: agent coverage and the silent pane problem

The bundled detection rules currently cover Claude Code (blocked, working, idle), plus Codex and OpenCode for identity and idle. Anything else is identified but reads as silent until someone contributes rules. The README is explicit: a new prompt shape wmux has not learned yet shows as 'we don't know'. This is honest, but it means wmux is only as useful as the rule set for the agents you actually run. If you use Cursor or Copilot, you will get a roster entry but no state. The `wmux detect explain --file` command is the intended way to author rules, and PRs are welcome. Whether you want to maintain your own detection rules is a real adoption question.

Alternatives and the difference in approach

The README names cmux as inspiration and herdr as the source of the agent-visibility model. cmux is a terminal multiplexer for AI agents that focuses on managing sessions, while herdr is a separate project that models agent visibility. wmux differs by being Windows-specific and by adding a file tree with edit-dots and stale-buffer protection. A generic alternative is tmux for Windows via WSL, but tmux has no concept of an agent's blocked state or a file's edit history. If you only need to keep multiple terminals open and switch between them, tmux is lighter. If you need to know which agent is waiting and what it changed, wmux is the only project in this description that provides that.

Maintenance, licensing, and what to verify first

wmux is MIT licensed, which means you can modify and redistribute it with attribution. The repository is active, with releases v2.6.0, v2.7.0, and v2.7.1 pushed on consecutive days in August 2026. That pace suggests rapid iteration, but it also implies you should expect frequent updates. The README warns that with no hooks configured there are simply no agent dots, and every number still works. That is a degradation to honest silence, not a guess, but it means you must verify that the hooks were injected into `~/.claude/settings.json` for Claude Code, and that the CDP proxy on `localhost:9222` is not blocked by your firewall. Before relying on wmux for edit safety, test a save against a file you have open in two panes to confirm the stale-buffer refusal works as documented.

Editorial conclusion

Adopt wmux if you run multiple AI coding agents on Windows and need a single window that tells you who is blocked, what they changed, and lets you fix a line without leaving the terminal. Skip it if you never run agents, need a general-purpose multiplexer, or require per-project configuration beyond the global settings. Before adopting, verify that your agent's UI is covered by the bundled detection rules; otherwise panes will read as silent, and you will need to write your own rules with `wmux detect explain --file`. Also confirm that the auto-injected Claude Code hooks in `~/.claude/settings.json` do not conflict with your existing hook setup, and test a save on a file you have open in two panes to see the stale-buffer refusal in action.

Frequently asked questions

Official sources

1. Official documentation
2. Official README
3. Project repository
4. Release notes