Herdr: The Rust Agent Multiplexer That Runs All Your AI Coding Agents in One Terminal

If you're running multiple AI coding agents simultaneously — Claude Code on one project, Codex on another, Copilot reviewing tests — you've hit the same wall: a sprawl of terminal windows, no idea which agent finished or got stuck, and losing everything when you close your laptop.

Herdr solves this. It's a terminal-native agent multiplexer written in Rust that gives every coding agent its own real terminal pane, tracks their state automatically, and keeps them all alive on a background server you can detach from and reattach to — even from your phone over SSH.

With over 9,200 GitHub stars and nearly 500 new stars daily, Herdr has become the go-to tool for developers who orchestrate multiple AI agents as part of their daily workflow. Let's break down what makes it special.

What Is Herdr and How Does It Work?

Herdr is a single Rust binary (~10MB) that runs inside the terminal you already use — iTerm, Terminal.app, Alacritty, WezTerm, whatever. Think of it as tmux rebuilt from scratch with first-class awareness of AI coding agents.

When you run herdr, it starts (or attaches to) a lightweight background server. That server manages:

• Workspaces — organize by repo, project, or folder
• Tabs and panes — click, drag, and split, mouse-native throughout
• Agent processes — each agent gets its own real PTY (pseudo-terminal), not a GUI imitation
• State tracking — every agent is automatically categorized as 🔴 blocked, 🟡 working, 🔵 done, or 🟢 idle

The key insight: Herdr doesn't replace your agents or wrap them in an app. It runs them in real terminals where they already run, then adds a thin observability and orchestration layer on top.

# Install Herdr
curl -fsSL https://herdr.dev/install.sh | sh

# Start Herdr (creates background server + opens workspace)
herdr

# Run your agents in panes
claude code ./project-a
codex ./project-b
copilot --review tests/

Why Herdr Instead of tmux or GUI Agent Managers?

You might be thinking: "I already use tmux" or "I've seen GUI tools for this." Here's the difference:

Herdr vs. tmux

tmux is excellent — but it was built before AI agents existed. It has no concept of agent state. When Claude Code finishes a task and waits for your input, tmux doesn't know. When Codex hits a permission prompt, tmux doesn't flag it. You have to manually check every pane.

Herdr adds zero-config agent awareness: process-name matching plus terminal-output heuristics automatically detect which agents are idle, working, blocked, or done. No hooks to wire, no config to write.

Herdr vs. GUI Agent Managers (Conductor, cmux, EmDash)

GUI managers show agent state — that's table stakes. But they're often Mac-only, closed-source, and they redraw your terminal inside an app wrapper. That means:

• Full-screen TUIs may not render correctly
• You can't SSH into a remote server and use them
• They require a GUI — no headless servers
• They're apps, not binaries — Electron overhead, accounts, telemetry

Herdr is a single Rust binary. It runs on Linux, macOS, and Windows (beta). It works over SSH. It has no GUI, no account, no telemetry. Your agents render in real terminals, and you can close your laptop and reattach from anywhere.

17+ Supported Agents — Out of the Box

Herdr ships with native integrations for the major AI coding agents. State detection works with zero configuration through process-name matching and output heuristics:

• Claude Code — full idle, working, and blocked detection
• OpenAI Codex — full state tracking with native session restore
• GitHub Copilot CLI — idle, working, and blocked
• Cursor Agent — full state detection
• Pi (pi.dev) — idle and working (partial blocked)
• Droid (Factory.ai) — full support
• Amp (ampcode.com) — full support
• OpenCode — full support
• Grok CLI — full support
• Devin CLI — full support
• Kilo Code CLI, Hermes Agent, QoderCLI, Kiro CLI — full support

Even agents without official integrations still work — Herdr runs them as a terminal multiplexer. You just don't get the automatic state badges.

# Install a native integration
herdr integration install claude
herdr integration install codex
herdr integration install copilot

# Or use the reusable agent skill
npx skills add ogulcancelik/herdr --skill herdr -g

The Socket API: Let Your Agents Orchestrate Each Other

This is where Herdr goes beyond any existing tool. A local Unix socket API lets agents (or your scripts) programmatically control the entire workspace:

• Create new workspaces and panes
• Spawn helper agents in sidecar panes
• Read output from any agent's terminal
• Subscribe to state changes (no polling needed)
• Zoom into specific panes

This means your lead agent can spin up helper agents, monitor their progress, read their output, and react — all without you touching the keyboard.

# Example: Agent creates a new workspace and spawns a helper
herdr socket workspace create "review-task"
herdr socket pane split --workspace "review-task"
herdr socket pane exec "claude code --review PR-142" --pane 1
herdr socket pane exec "codex --fix-tests" --pane 2

# Subscribe to state changes (great for scripts)
herdr socket state subscribe --json

Real-World Example: Multi-Agent Code Review Pipeline

Here's a practical workflow that demonstrates Herdr's power. Imagine you want to review a pull request with multiple AI agents working in parallel:

# 1. Start Herdr
herdr

# 2. Split into 3 panes (Ctrl+B then V, twice)

# 3. Pane 1: Claude Code does the main review
claude code
> Review PR #142 for security issues, performance, and code style

# 4. Pane 2: Codex fixes failing tests  
codex
> Fix all failing tests in the PR #142 branch

# 5. Pane 3: Copilot generates documentation
copilot --review src/new-feature/
> Generate JSDoc for all new public functions

The Herdr sidebar immediately shows you: 🟡 Claude Code is working on the review, 🟡 Codex is fixing tests, 🟢 Copilot is idle waiting for context.

Five minutes later: 🔵 Copilot is done, 🟡 Codex still working, 🔴 Claude Code is blocked — it needs your input on whether to approve an API change. You click the Claude Code pane, answer the question, and all three agents continue.

Close your laptop for lunch. Come back, run herdr, and everything is exactly where you left it. Nothing died.

Remote Sessions: Run Agents on Your Server, Control from Anywhere

Herdr's remote mode is a game-changer for developers who run agents on cloud servers or beefy workstations:

# Connect to a remote Herdr server
herdr --remote workbox
herdr --remote ssh://you@yourserver:2222

# Unlike plain SSH + tmux, Herdr preserves:
# - Image pasting into agent terminals
# - Mouse-native pane management
# - Real-time state sidebar
# - Full detach/reattach lifecycle

This means you can spin up Claude Code and Codex on a remote machine with 64GB RAM, walk away, and check in from your phone over SSH to see which agents are done.

Key Benefits

• 🎯 One terminal, all agents — Stop juggling terminal windows. Every agent gets a real PTY in one organized workspace.
• 🔴🟡🔵🟢 Instant state visibility — Know immediately which agent needs your attention. Zero config, works out of the box.
• 💪 Persistence that works — Close your laptop, lose WiFi, switch machines. Your agents keep running. Reattach from anywhere.
• ⚡ Blazing fast, tiny footprint — 10MB Rust binary. No Electron, no GUI framework, no account required.
• 🔌 Scriptable and extensible — Socket API lets agents orchestrate each other. Write plugins in any language.
• 🌍 Runs everywhere — Linux, macOS, Windows (beta). Local, remote, SSH. Any terminal emulator.
• 🆓 Open source — AGPL-3.0 licensed. Inspect the code, contribute, or get a commercial license.

Getting Started in Under 2 Minutes

# 1. Install
curl -fsSL https://herdr.dev/install.sh | sh

# Or with Homebrew
brew install herdr

# 2. Start
herdr

# 3. Run your agents in panes
# Use mouse to split panes, or Ctrl+B then V

# 4. Detach when done (agents keep running)
# Ctrl+B then Q

# 5. Reattach later
herdr

That's it. No config files to edit, no accounts to create, no plugins to install. State detection works immediately.

Frequently Asked Questions

Is Herdr free to use?

Yes. Herdr is open source under AGPL-3.0. You can use it freely for personal and commercial projects. Commercial licenses are available for organizations that can't comply with AGPL.

Does Herdr work with agents that aren't officially supported?

Absolutely. Any terminal-based agent works with Herdr as a multiplexer — you just won't get automatic state badges. Official integrations add native session restore and semantic state reporting. You can also build custom integrations using the Socket API.

How is Herdr different from tmux?

tmux is a general-purpose terminal multiplexer with no awareness of AI agents. Herdr adds automatic agent state detection (blocked, working, done, idle), mouse-native pane management, a Socket API for agent orchestration, and native integrations for 17+ coding agents. Think of it as tmux rebuilt for the AI-first development workflow.

Can I use Herdr over SSH on a remote server?

Yes — this is one of Herdr's key advantages. Run herdr --remote ssh://user@server and your local terminal becomes a client of the remote Herdr server. Unlike plain SSH + tmux, image pasting, mouse control, and state tracking all work seamlessly.

What happens to my agents when I disconnect?

Nothing. Herdr runs a background server that keeps all panes and agent processes alive. You can detach (Ctrl+B, Q), close your terminal, shut your laptop, or lose your network connection. When you reconnect and run herdr again, everything is exactly where you left it.

Does Herdr collect any data or require an account?

No. Herdr is a local binary with zero telemetry, no account requirement, and no cloud dependency. Everything runs on your machine (or your server). Your agent conversations and code never leave your infrastructure.

What platforms does Herdr support?

Herdr supports Linux and macOS fully, with Windows in beta. It works with any terminal emulator — iTerm2, Terminal.app, Alacritty, WezTerm, Kitty, GNOME Terminal, and more. The Rust binary has no external dependencies.