Herdr: Run Claude Code, Codex, and Every Agent in One Terminal
Herdr: Run Claude Code, Codex, and Every Agent in One Terminal
If you run Claude Code and Codex at the same time, Herdr replaces your pile of six, seven, eight scattered terminal windows with one organized, agent-aware workspace — and it keeps your agents running even when you close the window. It's open source, it's lightweight, and it works on both Mac and Windows. I set it up and ran my whole multi-agent workflow through it, and here's exactly how it works.
What Is Herdr?
Herdr is a terminal multiplexer built specifically for AI coding agents — think tmux, but agent-forward. It's a single Rust binary (about 10MB, no Electron) that lets you control multiple terminals from one space: I ran two Claude Code terminals up top, OpenCode on the bottom left, and Codex on the bottom right, all inside one window.
Herdr buys you three things:
- Organization — spaces, tabs, and panes instead of terminal windows floating around your desktop
- Monitoring — an agents sidebar that shows which agents are working, which are done, and which are stuck waiting on you
- Persistence — Herdr acts as a server, so if you exit a window mid-task, the agent keeps running in the background and you can bring it right back
That third one is the sleeper feature. If an agent is halfway through building a feature and you accidentally close the terminal, nothing dies. The work continues.
The project is on GitHub, it's free, and it auto-detects 15+ agents out of the box — Claude Code, Codex, GitHub Copilot CLI, and more — with zero configuration.
How Is Herdr Different From tmux?
tmux has been around for years and does the same core job — controlling all your terminals from a single place. The difference is that Herdr is agent-forward: it has a dedicated agents tab that shows which agents are running, which need your input, and which are done.
That matters once you scale past three or four terminals. With five, six, seven, eight agents going — including ones that aren't even visible on screen — tmux gives you no signal about which one needs attention. Herdr does, out of the box. A working agent gets a little animated indicator in the sidebar; a finished or stuck one shows its state at a glance.
The other differences that pushed me over:
- It works on Windows. I'm on Windows daily, and tmux is not a native option there.
- Mouse support is first-class. You can click on spaces, right-click panes to split or rename them, and drag things around. It isn't purely keybinds like tmux.
- Persistence is built around agents. Close the window mid-run, SSH in from another machine, even get back to it from your phone — the session keeps going as long as your computer is on.
How Do You Install Herdr?
Installation is one line of code. Head to herdr.dev or the Herdr GitHub repo, copy the install command for your platform, and paste it into your terminal.
Once it's installed, open your terminal and type herdr. You'll see three things: spaces on the left, an agents tab, and your actual terminal. The terminal part works exactly like a normal terminal — type claude and Claude Code opens inside it, and Claude immediately shows up as a tracked agent in the sidebar.
How Do Spaces, Tabs, and Panes Work?
Herdr organizes everything in a three-level hierarchy. The mental model that clicked for me: spaces are folders, tabs are subfolders, and panes are the agents working inside them.
- Spaces = projects. The highest level. I have my main space plus a separate one for a web design project. Each space has its own agents, and the sidebar lists them all — so agents in different projects can work simultaneously without their terminals being visible.
- Tabs = subtasks within a project. Inside a space, tabs work like terminal tabs. I named one tab "research" for Claude and Codex to do research in, and a second tab just for running the dev server. Clean buckets for different workstreams.
- Panes = individual terminals inside a tab. Right-click any pane and hit split, and now Claude Code and Codex sit side by side in the same tab. You can have as many agents per tab as you want.
All of this exists to fight the chaos of a ton of terminals running a ton of agents doing a ton of different things. Spaces keep projects separated, tabs keep subtasks separated, and the agents sidebar tells you where to look next.
Can Your Agents Control Herdr Themselves?
Yes — and this is where it gets interesting. The Herdr skill teaches your agents how to use Herdr itself: they can spawn panes, create spaces, and organize their own work without you managing any of it. Installing the skill is another one-line paste, linked from the agent skills section of the Herdr GitHub.
Here's the real use case. I have a skill where Claude Code and Codex go back and forth on an implementation plan — Claude drafts, Codex critiques, Claude revises — through multiple iterations until they agree. Normally that whole exchange runs headlessly. I'm just trusting that the two models are actually talking.
With Herdr, I told Claude to use the Herdr skill to make that loop visible. Claude created the plan, spawned a new pane, and sent the plan to Codex — and I could watch the Codex critic list its issues with the plan in real time, watch the verdict come back as REVISE, and watch the revision go back for another round. The entire cross-model review happened in front of me instead of in a black box.
There's also a plugin marketplace where people have started publishing add-ons for the Herdr construct, so the ecosystem is bigger than just the core binary.
What Happens When You Close the Terminal?
Nothing — and that's the point. Herdr runs as a background server, so exiting the window doesn't kill your agents.
I tested this directly: with both Claude Code and Codex mid-task, I exited the terminal completely. Then I opened PowerShell, typed herdr, and both agents were still there, still working. As long as your computer is on, everything persists. If you're on a laptop, close the lid (as long as it doesn't sleep) and your agents keep going for as long as you want.
When you actually want to shut something down, right-click a pane and hit close, or close an entire space the same way. Shutdown is deliberate, never accidental.
What Are the Herdr Keybinds?
If you're a tmux person, the keybinds will feel familiar: Ctrl+B is the prefix, followed by a command key. So "prefix + V" for a vertical split means Ctrl+B, then V. Go to menu → keybinds to see the full list.
But here's the thing — you don't need them. Almost everything in Herdr can be done with right-clicks and the mouse: splitting panes, renaming, zooming in and out, changing themes, sounds, toasts, and integrations in settings. That alone makes it dramatically more approachable than tmux for anyone who never memorized the prefix-key liturgy.
Why This Matters for Multi-Agent Workflows
Claude Code's built-in agent view is great — but it only works inside Claude Code. The moment you bring in Codex or OpenCode, you're back to juggling raw terminal windows with zero visibility.
Herdr is the layer above all of them. One window, every harness, live status on every agent, and nothing dies when you close the lid. It's open source, free, and runs on Windows and Mac. If you're running more than two coding agents at once, try it.
Frequently Asked Questions
What is Herdr?
Herdr is an open-source terminal multiplexer built for AI coding agents. It lets you run Claude Code, Codex, OpenCode, and other agents in one organized window with spaces, tabs, and panes, plus a sidebar that tracks each agent's status — working, done, or stuck.
How is Herdr different from tmux?
Herdr is agent-aware and tmux is not. Herdr's agents sidebar shows which agents are running, done, or waiting on input out of the box — tmux can't do that. Herdr also runs natively on Windows and has first-class mouse support, while tmux is keybind-driven and Unix-only.
Do agents keep running if you close the Herdr window?
Yes. Herdr acts as a background server, so exiting the terminal doesn't stop your agents. Reopen your terminal, type herdr, and your sessions come right back mid-task. You can also reattach over SSH or from your phone.
Is Herdr free?
Yes — Herdr is open source and free to use. It's a single lightweight Rust binary, installs with one command from herdr.dev or the GitHub repo, and works on both Mac and Windows.
Can Claude Code control Herdr directly?
Yes, through the Herdr agent skill. Install it with one paste command and your agents can spawn panes, create spaces, and organize their own terminal layout — useful for making headless agent-to-agent workflows, like a Claude/Codex plan review loop, visible in real time.
If you want to go deeper into multi-agent coding workflows, join the free Chase AI community for templates, prompts, and live breakdowns. And if you're serious about building with AI, check out the paid community, Chase AI+, for hands-on guidance on how to make money with AI.
