Conductor vs Claude Squad vs Hermes IDE: 3 Worktree Managers for Parallel Agents

Book: System Design Pocket Guide: Interviews The series: Pocket Guides for Developers — 8 short books on agents, RAG, prompts, observability, events, databases and system design — the whole series My project: Hermes IDE | GitHub — an IDE for developers who ship with Claude Code and other AI coding tools Me: xgabriel.com | GitHub A small category of tools now exists just to keep coding agents out of each other's folders, and three of them could hardly look less alike. Claude Squad is a terminal UI, Conductor is a Mac app, and Hermes IDE is a desktop app for macOS, Windows and Linux. They solve the same problem. Put two agents, say Claude Code and Codex, in one checkout and sooner or later they edit the same file. The usual fix is a git worktree per agent: a separate folder and branch for each task, all backed by one repository. Doing that by hand gets old by the third task, and that's the chore these tools take over. I build Hermes IDE, so weigh this with that in mind. I haven't run Conductor or Claude Squad myself. Everything about them below comes from their own docs and repository, linked where it's used. What all three do, and where they split The core loop is the same in every tool. You describe a task, the tool cuts a branch, puts it in its own worktree, and starts an agent there. Agents can't overwrite each other's files because they don't share a folder. When the work is done, you review the diff and get it back into your main branch. The differences sit at both ends of that loop. At the start: which agents it can run, on which operating system, and how a fresh worktree gets the .env file and dependencies it doesn't have yet. At the end it's about review. You need to read the diff and send fixes back, and then the branch has to get merged. One repository, three worktrees, three agents that never share a folder. Claude Squad: tmux and worktrees in one terminal window Claude Squad describes itself as "a terminal app that manages multiple Claude Code, Codex, Gemini (and other local agents including Aider) in separate workspaces." Its README explains the build in three lines: tmux for an isolated terminal session per agent, git worktrees so each session works on its own branch, and "a simple TUI interface" on top. The command is cs. Claude Code is the default program, and the README shows the others with the -p flag: cs cs -p "codex" cs -p "gemini" You can also define profiles in its config file, each a name plus a launch command, and set a default_program. Anything that runs in a terminal can be a profile. The keyboard drives everything. n starts a session and N starts one with a prompt. Enter attaches to a session and ctrl-q detaches. tab switches between the preview and the diff of what the agent changed. When you're happy, s commits and pushes the branch to GitHub, and c commits the changes and pauses the session so you can check out the branch yourself. r resumes it. There is an experimental -y / --autoyes flag that auto-accepts prompts for Claude Code and Aider, for tasks you want running in the background. It's open source under AGPL-3.0, it's a single binary, and it lives where many agent users already are: a terminal. The latest release, v1.0.20, shipped on August 20, 2026 with archives for macOS, Linux and Windows. It needs tmux installed, plus the GitHub CLI. Review is thinner. You get a diff view and a commit-and-push key. Line comments, per-turn history and checks before merging aren't part of what the README describes, so those happen in your editor or on GitHub. Conductor: a Mac app built around the pull request Conductor says it lets you "run Claude Code, Codex, Cursor, and OpenCode in parallel" (docs). It is a Mac app. The installation page says plainly that "Conductor is not available for Windows or Linux yet," and that "You need GitHub and at least one agent provider to use Conductor." Workspaces are worktrees, usually created under ~/conductor/workspaces// (worktree docs). For the files a new worktree is missing, Conductor "uses .worktreeinclude-style patterns and only copies files that Git already ignores," and setup scripts handle dependency installs, code generation and database setup. Where Conductor goes furthest is the GitHub end of the loop. Its workflow page lets you start a workspace from a branch, a pull request, a GitHub issue or a Linear issue (Command+Shift+N). The Diff Viewer (Command+Shift+D) turns inline comments into attachments you send back to the agent, and a Review action asks an agent to review the work. The Checks tab collects git status, CI, deployments, GitHub review threads and todos in one place. Command+Shift+P creates the pull request, and Conductor can draft the description, answer review comments and fix failing checks. Finished workspaces are archived and can be restored later with their chat history. It's also the only one of the three with a hosted option. The pricing page has a free tier with local Mac workspaces where you "bring your own subscriptions and keys," and a $50/month Pro plan that adds cloud workspaces, multiplayer, an API and the mobile app. Teams is $60 per user per month. The changelog lists 0.90.0, Conductor for iOS, on October 2, 2026. So: Mac only for now, GitHub required, four agents, and paid plans for the cloud features. If your team already lives in pull requests and Linear, none of that may bother you. Hermes IDE: any CLI agent, with a Review Desk and Land Hermes is a desktop app for macOS (Apple Silicon and Intel), Windows and Linux (.deb and AppImage), per its README. Every agent runs its own interface in a real terminal. The built-in list covers Claude Code, Codex, Antigravity CLI, Gemini CLI, GitHub Copilot CLI, OpenCode, goose, Aider, Kiro CLI and Hermes Agent, plus any command-line agent you add. Each agent uses its own sign-in, and Hermes never asks for your tokens. ⌘N (Ctrl+Shift+N on Windows and Linux) starts a task on its own hermes/ branch in its own worktree, with the task as the agent's first prompt. New worktrees get their dependencies and their own ports, from worktree settings you keep per repository (docs) and can approve or skip. Your own checks can decide when a task is done. The ⌘N New task sheet: Claude Code, project chip "orders-api +2", a new hermes/ worktree, Accept edits, the extra-folder chips and the "Hermes will run" preview of the claude command. While agents work, every session shows one status: working, needs approval, asked you, done, error or idle. An attention inbox counts the agents blocked on you, and ⌘I jumps to the one that has waited longest. You answer in that agent's own terminal. Hermes never answers a permission prompt for you. Review is where Hermes differs most from Claude Squad. Every agent turn is saved, so you can diff or restore a single turn. The Review Desk (⌘G) shows changes by turn or by file, with risk flags, and a line comment goes back to the agent that wrote that line. Land then commits, opens a pull request or merges locally, into the branch the task started from, and can be undone. A local merge doesn't need GitHub at all. Hermes is young: it has shipped several releases in October 2026 alone, and everything runs on your own machine, with no cloud workspaces. It's source-available under BSL 1.1, not an OSI open-source licence: you can't use it to build a competing code editor, terminal emulator or IDE offered to third parties, and each release converts to Apache 2.0 three years after it ships. A desktop app is also more to install than one binary in tmux. Side by side, and which one fits you Claude Squad Conductor Hermes IDE Shape TUI (cs) on tmux Mac app Desktop app Platforms macOS, Linux, Windows archives macOS only macOS, Windows, Linux Agents Claude Code default, any CLI via -p Claude Code, Codex, Cursor, OpenCode 10 built in, plus custom CLIs Worktree setup Worktree + branch only (no copy or setup step in the README) Files to copy, setup scripts Per-repo settings: dependencies, ports, checks Review Diff tab Diff Viewer, comments to agent, Checks tab Review Desk by turn or file, comments to the author Merge s commits and pushes Pull request on GitHub Land: commit, PR or local merge, with undo GitHub gh is a prerequisite Required Optional Licence AGPL-3.0 Not stated BSL 1.1 Price Free Free; Pro $50/month Free Same destination, three ways in: how much ceremony you want between the agent and your main branch. Pick Claude Squad if you already work in tmux, want an open-source tool you can read end to end, and are happy to review and merge in your editor or on GitHub. Nothing here is lighter to install: one binary plus tmux. Pick Conductor if you're on a Mac, your work starts from GitHub or Linear issues and ends in pull requests, and you'd use cloud workspaces or the mobile app. It has the deepest pull-request flow of the three. Pick Hermes IDE if you mix agents from several vendors, need Windows or Linux, or want to review by turn and merge locally before anything reaches GitHub. And if you only ever run one agent, you may not need any of them. A couple of git worktree add commands and a second terminal cover it. If this was useful I build Hermes IDE, and it's free on macOS, Windows and Linux: download it, and if it saves you a tangled diff, star it on GitHub. Hermes IDE is not affiliated with Nous Research, Anthropic, OpenAI or any agent vendor. Picking between these three came down to trade-offs: platform against reach, ceremony against control, local against hosted. If you like working through choices that way, System Design Pocket Guide: Interviews does it for whole systems, 15 real designs taken step by step, trade-offs included. It's one of the Pocket Guides for Developers, 8 short books on agents, RAG, prompts, observability, events, databases and system design.