Can I run several coding agents at the same time on Windows?

Yes. You can do it by hand, opening each agent in its own terminal tab and its own git worktree. AgentVera does it natively on Windows: it opens 18 coding agents side by side in a grid per project, each in its own real terminal.

Do I need WSL to run coding agents on Windows?

No. Claude Code, Codex, Gemini CLI, Copilot CLI, Cursor CLI and most others run natively in PowerShell. WSL helps when your project depends on Linux tooling or when you want Claude Code’s sandbox.

Does Conductor work on Windows?

Not as of November 2026. Conductor is macOS only and keeps a waitlist for Windows and Linux.

How do I manage parallel agents on Windows without tmux?

Open split panes in Windows Terminal, or use a desktop app that shows the agents as a grid in its own window. tmux has no native Windows build, so tools built on it, such as Claude Squad, only run inside WSL.

Managing coding agents on Windows: parallel agents

Search for “parallel coding agents” and the download page of most tools you find has a single button: Download for macOS. If you write code on Windows, the agents themselves work; Claude Code, Codex and Gemini CLI now open natively in PowerShell. What’s missing is a tool to manage them together.

This post fills that gap: how to set up, isolate and keep an eye on several coding agents on Windows, which tools actually run there, and when you need WSL. We build AgentVera, one of those tools, so we’ll cover it too, limits included.

Do the agents themselves run on Windows?

They do. As of 2026, most major coding agents have a native Windows install:

Agent	On Windows	Note
Claude Code	Native (PowerShell, CMD, winget, npm)	Git for Windows optional; sandbox only in WSL 2
Codex	Native, with its own Windows sandbox	Windows 11 is the main target
Gemini CLI	Via npm	Node.js 20+
Copilot CLI	winget or npm	PowerShell 6+
Cursor CLI	PowerShell install script
opencode	scoop, choco, npm	Its docs recommend WSL for the best experience
Kiro CLI	PowerShell install script	Windows 11 only
Amp	Official docs say WSL

Every install command is in installing coding agents on Windows.

What about agent managers?

This is where the real problem is. Most tools built to run several agents in parallel are Mac-only:

Conductor is macOS only. Its official docs point to a waitlist for Windows and Linux.
Paradev (formerly Superset) runs on macOS; its FAQ says Windows isn’t available yet. The Linux build is experimental.
Sculptor (Imbue) is for Apple Silicon Macs and Linux; Windows is “on the way”.
Claude Squad is built on tmux. tmux has no native Windows build, so on Windows it only runs inside WSL.
Nimbalyst (the successor of Crystal) runs on Windows, macOS and Linux and supports Claude Code and Codex.
Anthropic’s Claude desktop app runs on Windows and can open parallel sessions in worktrees, but it only runs Claude.

So if you want to manage agents from several providers in one window on Windows, the options are few. If you’re coming from a Mac and looking for something to replace Conductor, Conductor for Windows goes through the comparison.

Three things parallel agents need

Whatever tool you use, parallel agents on Windows rest on these three.

1. A separate working copy per agent

Two agents working in the same folder overwrite each other’s files. The fix is git worktrees: separate folders from the same repo, on separate branches.

git worktree add ..\project-auth -b agent/auth
git worktree add ..\project-tests -b agent/tests

Two Windows-specific things to watch:

Long paths. Deep node_modules folders can hit the 260-character limit. Set git config --global core.longpaths true and put worktrees under a short root such as C:\src.
Line endings. When agents write files with LF and Windows tools save them with CRLF, diffs get noisy and some edit tools fail with “text not found”. Add a .gitattributes with * text=auto to the repo.

For details, see parallel development with git worktrees.

2. A place to see the agents at once

On a Mac this usually falls to tmux. On Windows you have two options:

Windows Terminal panes. Open a pane with Alt+Shift+D and start an agent in each. Free and fast, but you find out which agent finished and which one is waiting for approval by checking the panes one by one.
An agent manager. A desktop app that shows the agents in a grid and labels each pane with its status (working, waiting for approval, done).

3. A way to collect approval requests

If three agents ask for permission to write files at the same moment, you become the slowest link. Seeing approval requests in a single queue, or setting permission rules in advance for work you trust, is what makes parallel work actually parallel. A Windows notification when an agent finishes or asks a question does the same job.

Parallel agents on Windows with AgentVera

AgentVera is a desktop app that runs natively on Windows, not a port running through WSL. Download the x64 installer and run it; later versions download in the background. (The installer isn’t signed with a Microsoft certificate yet; in SmartScreen, click “More info”, then “Run anyway”.)

What it does on Windows:

18 coding agents, one grid. Claude Code, Codex, Gemini CLI, Copilot, Cursor, opencode, Qwen Code, Kiro and others sit side by side in a grid from 1×1 to 5×4. Each pane shows its status and live context size. See the workspace.
Real terminals. Each agent runs in a real terminal session on Windows’ ConPTY layer, so the agent’s own interface looks the way it does in Windows Terminal.
.cmd shims aren’t a problem. On Windows, npm installs shims such as claude.cmd and codex.cmd, and cmd.exe re-parses their arguments. AgentVera unwraps the shim and starts the script inside it directly, so roles and JSON settings containing & or % arrive intact.
Worktrees. Create a worktree from the Git panel and start an agent directly in it.
One approval queue. Permission requests from every agent land in a single queue, and you get a Windows notification when an agent finishes.
Limits. When a Claude limit runs out, account switching (Pro) moves the agent to the next account.

The limits, plainly: there’s only an x64 installer for now (it runs under emulation on Windows on ARM), and it doesn’t manage agents inside WSL directly. You can run an agent installed in WSL by opening a Terminal agent and running a command such as wsl claude; you get a terminal pane, but not agent-specific features such as session resume.

Everything Windows-specific is summed up on the Windows page.

Should I use WSL?

Short answer: yes if your project depends on Linux, otherwise no.

Native Windows is the right choice for .NET, C#, Unity, Windows desktop apps and any project whose code lives on the C: drive. Setup is short and file access is fast.
WSL 2 is better for Docker-heavy projects, shell scripts and tools that only run on Linux. Keep the code in the Linux filesystem; working over /mnt/c is noticeably slower.

The full comparison is in WSL vs native Windows.

Common problems on Windows

The agent command isn’t recognized. The install folder hasn’t been added to PATH yet. Open a new terminal; where claude or where codex shows what gets found.

The agent runs shell commands in the wrong shell. Claude Code uses Git Bash if Git for Windows is installed, and PowerShell otherwise. To point it at a specific Git Bash, set CLAUDE_CODE_GIT_BASH_PATH.

A worktree can’t be created. Usually a long path or a locked file. Turn on core.longpaths, put the worktree under a short root, and close open editors and dev servers.

Two agents want the same port. Give each worktree’s dev server its own port.

Wrapping up

Running a coding agent on Windows is no longer the problem; managing several is still harder than on a Mac, because most tools skip Windows. A worktree per agent, a side-by-side view and a single approval queue close that gap. If you want that ready to go, download AgentVera for free; it works the same on Windows, macOS and Linux.