Claude Code's agent-team panes need tmux, and Anthropic says Windows Terminal is out

A vendor-authored post on dev.to makes a claim about parallel coding agents on Windows that is checkable rather than arguable: Claude Code's agent teams feature puts each agent in its own split pane, the backend for those splits is tmux with iTerm2 as the alternative, and Anthropic's own documentation says split-pane mode is not supported in Windows Terminal ^[4][5]. For anyone on a Windows toolchain, one line in a docs page decides the shape of the whole workflow.

Start with why panes matter. Four agents pointed at one repository overwrite each other's files and contend for the same working tree, the same branch and, across providers, the same credentials directory ^[2]. The standard fix is one git worktree per agent, one terminal pane per worktree, and some way to see all of it at once ^[3]. Two of those three are easy on Windows. The pane grid is not, because tmux does not run natively there; it runs under WSL, which puts the agents in a Linux filesystem talking to a Windows checkout across a boundary that is slow for file watching and awkward for native toolchains ^[6]. If the build is MSBuild, Unity or Unreal, that boundary is not a detail ^[7].

The author, who discloses building a competing product and says his platform claims come from each vendor's published documentation ^[1], surveyed the field as of August 2026: Conductor is macOS only on Apple Silicon ^[8]; cmux is macOS, with Linux in beta and Windows on a waitlist ^[9]; Superset is macOS, with Windows and Linux listed as coming soon ^[10]; Sculptor is macOS and Linux, with Windows through WSL ^[11]. That is zero of four with native Windows support ^[22]. All four shipped within the past year, and his read is habit rather than hostility: people build for the machine on their desk ^[12]. The demand sits elsewhere, in Unity and Unreal studios, .NET shops, and regulated fintech and hospital IT ^[13].

The useful part of the post is the requirements list, because it is short and none of it needs a multiplexer. One worktree per agent, sharing a single object store, so edits do not collide and merges happen normally afterward ^[14]. One isolated HOME per agent, which is the item he says people miss: Claude Code reads ~/.claude, Codex reads ~/.codex, Gemini CLI reads ~/.gemini, so two agents under one user account share configuration and, more consequentially, one authenticated session ^[15]. On Windows that means pointing USERPROFILE and HOME somewhere different per process ^[16]. A pane grid, which he argues is a 1980 problem that a desktop application can simply draw ^[17]. And a review path, since four agents produce four diffs and reading them one at a time is where the time savings go ^[18].

Then the pitch, which he labels as one: NestMux runs Claude Code, Codex, Gemini CLI, Copilot and OpenCode side by side in a resizable pane grid, each pane with its own account, isolated HOME and worktree, plus a diff viewer, MCP server management, broadcast prompting and voice input through a local Whisper model ^[19]. It runs on Windows 10 and up, macOS 13 and up, and Linux, local-first with no telemetry ^[20]. He also hands over the free version of the same idea: wt.exe split-pane produces a grid, and each pane can be driven into a different worktree with a different HOME, fiddly to set up and rebuilt every session ^[21].

Two things are worth tracking. Whether cmux's waitlist and Superset's "coming soon" turn into shipped Windows builds ^[9][10], and whether the split-pane backend stops assuming tmux or iTerm2 ^[4]. Until one of those changes, the honest Windows answer is a scripted pane layout you rebuild each morning ^[21].