Herder — Is This the Ultimate Agent Multiplexer?

1 What is Herder?

▶ 0:00

Herder (stylized herdr) is an agent multiplexer — a single Rust binary that runs inside your existing terminal and lets you manage multiple AI coding agents side by side in different panes or tabs. It shows you which agent is working, which is blocked, and which is done, complete with system notifications.

The name is a play on "herder" — as in a sheep herder, someone who herds multiple agents. It was built by developer Oğulcan Celik and is distributed as a single Rust binary built with Ratatui, meaning no Electron, no separate app — just text rendering in your terminal.

2 Why Not Just tmux?

▶ 0:28

Herder takes heavy inspiration from tmux — the classic terminal multiplexer with tabs, split panes, and persistent sessions. But tmux was built decades ago, long before AI agents existed. If you run a coding agent inside a tmux pane, tmux has no idea it exists and no concept of its status.

Tools like Warp and Cmux solve this by pulling you out of your own terminal into their own, forcing you to learn a different workflow. Herder's approach is different: it gives you the persistence of tmux with agent awareness baked in, while staying inside your existing terminal.

3 Architecture

▶ 1:26

Herder uses a client-server architecture communicating via a Unix socket:

• Server — manages tabs, panes, persistence, and session state
• Client — a thin TUI that sends keystrokes to the server and renders the output

Because it's a single Rust binary built with Ratatui, the whole thing is just rendering text to your terminal. This means it runs anywhere your terminal runs — including over SSH. There's also a socket API, which means agents themselves can drive Herder programmatically (creating panes, spawning other agents, etc.).

4 Installation

▶ 1:43

Herder can be installed via three methods:

After installing, just run herdr to start. You'll see a terminal view with workspaces and agents panels.

5 Workspaces, Tabs & Panes

▶ 2:02

Navigation works both via keyboard and mouse:

• Default prefix: Ctrl+B (same as tmux, configurable)
• New workspace: prefix + Shift+N
• Keyboard shortcuts: prefix + ? to see all
• Mouse: click to navigate tabs, right-click to close, create horizontal/vertical splits
• Rename: workspaces and individual panes can be renamed

Everything you'd expect from a modern multiplexer — tabs, splits, rename, reorder — with the addition of mouse support that tmux purists might actually appreciate.

6 Agent Awareness — The Killer Feature

▶ 2:41

This is where Herder differentiates itself. When you open Claude Code in one pane and Codex in another, Herder automatically detects them and shows their status:

• 🟡 Blocked — waiting for user input
• 🔵 Working — actively processing
• ⚪ Idle — ready for a prompt
• ✅ Done — task completed

In the demo, Claude Code is shown as "blocked" (waiting for directory access), while Codex shows as "idle." As soon as prompts are given, statuses update to "working" in real time.

7 Agents Driving Herder

▶ 3:14

The socket API means agents can control Herder themselves. In the demo, Claude Code is given a prompt to:

1. Use the Herder CLI to spawn two new panes with Claude Code
2. Run one pane with "summarize the largest file in the project"
3. Run another with "search the codebase for hardcoded API keys or secrets"

The agent creates the panes, assigns them appropriate titles ("Claude Secrets", "Disk Code"), runs both prompts in parallel, then summarizes the output of both panes and closes them — all orchestrated by the agent itself.

8 SSH Remote Mode

▶ 4:22

Because Herder is just a binary, you can install it on a remote server and SSH in. Sessions persist — you get back to the exact workspace and agent you had open. But there's a catch: your local configuration (keybindings, theme) doesn't carry over.

The solution is the --remote flag:

This runs the server on the remote machine (managing files, panes, sessions) but the client on your local machine (using your local keybindings and theme). Keystrokes travel via the Unix socket over SSH, giving you the best of both worlds.

How it works

• Without --remote: both server and client run on the remote machine → remote config applies
• With --remote: server on remote, client on local → local config + remote files

9 Themes & Configuration

▶ 5:55

Herder ships with multiple themes including:

• Nord, Gruvbox, Catppuccin, and a Terminal theme (uses your terminal's own colors)

Configuration options include:

• Prefix key: changeable from Ctrl+B to anything (demo uses Ctrl+Space)
• Sound alerts: system notifications when agents finish
• Toast menu: configurable between terminal or system notifications
• All settings editable via the in-app menu or config file

10 Other Features & Verdict

▶ 6:13

Features not fully covered in the demo but mentioned:

• Worktree integrations — git worktree support for parallel development
• Official harness integrations — give agents access to Herder skills, letting them resume exact sessions
• Session persistence — re-entering Herder restores your exact workspace, tabs, and agent sessions

The main selling point

The presenter's verdict: "The fact that this works in your existing terminal is the main selling point." Unlike tools that require a separate app, Herder integrates into whatever terminal you've already configured — just open a new tab and run herdr.

Limitations

• ❌ No Windows support — due to Unix socket requirements and terminal PTY dependencies. WSL works but isn't native.
• ❌ No built-in web browser — unlike Cmux, but the presenter notes he rarely used that feature anyway.

🎯 Key Takeaways

🔗 Resources & Links

• Herder (herdr.dev) — official website and documentation
• Cmux — alternative agent multiplexer mentioned in the video

⏱ Timestamp Index