cmux

cmux is an open-source native macOS terminal built for people who run several AI coding agents at the same time. The repository describes it as a Ghostty-based terminal with vertical tabs and notifications for agents. It uses libghostty as a rendering library and is implemented primarily with Swift and AppKit instead of Electron or Tauri, which makes it closer to a native terminal than a web wrapper. The product problem is familiar to heavy agent users: Claude Code, Codex, Gemini CLI, OpenCode, Cursor CLI, Copilot, and similar tools can each run long sessions, ask for input, or finish tasks while the developer is focused elsewhere. cmux organizes those sessions with workspaces, vertical tabs, split panes, status indicators, sidebar badges, a notification panel, and macOS notifications. cmux is more than a terminal window. The documentation describes embedded browser panes that can sit beside terminals and be driven through CLI or socket APIs. Agents and scripts can create workspaces, split panes, send input, read screen content, capture windows, and automate browser surfaces with actions such as DOM snapshots, clicks, form fills, JavaScript evaluation, console access, and network access. Remote work is also part of the project. cmux documents SSH workflows, remote localhost routing for browser panes, native tmux attachment, and newer local tmux and zellij persistence options. Session restore can preserve layout, working directories, best-effort scrollback, and browser state, with safeguards around untrusted imports. cmux is strongest for developers who already live in terminals and want one programmable workspace for multiple agent sessions. It is not a general no-code automation platform. The app is macOS-focused, the codebase is large, and users should expect fast-moving releases. The app repository is open source under GPL-3.0-or-later for the desktop app, while some web, worker, relay, and server pieces use BUSL-1.1 according to the repository summary. For OpenTools readers, the practical test is whether the project removes work from a real builder workflow. This listing focuses on documented setup paths, concrete capabilities, limits, pricing signals, and the kinds of teams that can safely use the product. Teams should still run their own review before using any agent-connected tool on sensitive data. Check what runs locally, what touches third-party APIs, which permissions are granted, and whether the workflow can be scoped to a test project before production use. The page avoids unsupported claims. When the source names a license, package, install command, browser, runtime, or feature, the listing uses that source-backed fact. When pricing or limits are not public, it says so instead of guessing.