cat ./posts/herdr-ai-agent-terminal-multiplexer-guide.en.md
# A primary-source introduction to Herdr for organizing Claude Code and Codex in workspaces, tabs, and panes, with installation, controls, and practical workflows.
grep -R tag ./posts
No comments yet.
Herdr is an open-source terminal multiplexer built for AI coding agents. It lets you run tools such as Claude Code and Codex in separate panes while organizing projects into workspaces, tabs, and panes.
According to the official Quick start, Herdr detects agents automatically and shows each agent as working, blocked, done, or idle in the sidebar. That makes it easier to see which project needs attention without cycling through every terminal.
As of July 23, 2026, the latest official release is v0.7.5, published on July 21, 2026. This article follows the Stable documentation and the official GitHub repository.
Herdr does not provide an AI model or coding agent. Install and authenticate each agent CLI separately. Herdr is the workspace that helps you operate several of them without losing track of their state.
The official install script is shown below. Open https://herdr.dev/install.sh and review it before execution when your environment requires that control.
curl -fsSL https://herdr.dev/install.sh | shNix users can run the verified Stable release directly:
nix run github:ogulcancelik/herdr/v0.7.5A Homebrew formula is also available:
brew install herdrLaunch Herdr from a project directory:
cd your-project
herdrIf the session has no workspace, Herdr creates one automatically. Start your normal agent inside a pane:
claude
# or
codexHerdr is mouse-native: click panes and tabs, drag split borders, and right-click to split panes or create tabs. For keyboard control, press ctrl+b to enter prefix mode, then use an action key.
| Action | Key |
|---|---|
| Split right | prefix+v |
| Split down | prefix+- |
| New tab | prefix+c |
| Next/previous tab | prefix+n / prefix+p |
| Workspace navigation | prefix+w |
| Detach | prefix+q |
Assign clear roles instead of letting multiple agents edit the same files. For example, use Claude Code for implementation in the left pane, Codex for tests or review in the right pane, and another tab for server logs. Check only the pane marked blocked when an agent needs input.
Simultaneous edits to the same file can still create conflicts. Separate branches, workspaces, or ownership boundaries remain important.
The main benefit is not merely running more agents. It is seeing the growing workload in one place and returning to the pane that needs you. Start with one project and two panes.
*Diagram: Claude Code, Codex, tests, and each agent state are monitored from one terminal workspace.*