~ ❯ cd compare
wrkstrms works like a terminal multiplexer underneath: a server holds real terminals, and your terminal attaches to it. What it adds is knowing that some of those terminals are agents: what each one is doing, which one needs you, and what they've all learned.
| Capability | wrkstrms | tmux | zellij |
|---|---|---|---|
| What it is | a terminal environment for agents, with a server | terminal multiplexer | terminal workspace |
| Runs inside your terminal | yes | yes | yes |
| Keeps running when you close the window | yes, a server holds every pane | yes, detach | yes, detach |
| After a reboot | offers to restore; agents resume their conversations | with a plugin (tmux-resurrect) | resurrects sessions |
| Knows what each agent is doing | blocked · working · done · idle | no | no |
| Takes you to the agent that needs you | ⌃␣ a, and a desktop notification | no | no |
| Memory shared by every agent | one local store, over MCP too | no | no |
| Hands work to another agent | ⌃␣ A, with the prompts, commits and diff | no | no |
| A worktree and branch per task | ws new, from templates | no | no |
| Pull request reviews | ws review: a worktree and a review agent per PR | no | no |
| Scripting | ws commands and a socket protocol | tmux commands | zellij actions |
| Default prefix | Ctrl-Space, changeable | Ctrl-b, changeable | modes (Ctrl-p for panes…) |
tmux and zellij are good at what they do, and wrkstrms borrows their model: named servers (ws -L, like tmux -L), detach and attach. Neither is built to know what an agent is.
tmux or zellij
Panes, tabs, a prefix key, detach and reattach all work the way you expect, and the prefix can be tmux's classic Ctrl-a if your fingers insist. You can even run wrkstrms inside tmux. What changes is that the frames, tabs and sidebar turn red when an agent is blocked, yellow while it works and green when it's done.
your terminal’s tabs and splits
wrkstrms isn't a terminal emulator; it runs inside the one you like, and each pane is a full terminal with true color and the kitty keyboard protocol. The difference is that closing the window no longer ends the work: the agents keep going, and wsputs you back where you were.
a desktop app for agents
Agents run as their own CLIs in real terminals, so everything the CLI does works as usual. Claude Code and Codex report their status through hooks; nine other agent CLIs are recognized by name, and you can add more. Reviews happen with an agent, the git panel and Neovim in each tab rather than a separate diff window.
agents →