This Repo Just Solved The #1 Claude + Codex Headache

CChase AI
Computing/SoftwareSmall Business/StartupsInternet Technology

Transcript

00:00:00If you're someone who uses Claude code and codecs at the same time, or really any combination
00:00:04of different coding agents, then you understand the pain of having six, seven, eight different
00:00:09terminals open at the same time, and trying to not only monitor all of them simultaneously,
00:00:14but figure out which ones need your attention first, and God forbid you accidentally exit out
00:00:19of one of those windows when they're in the middle of something. But there's an open source
00:00:22tool that can solve all of these problems for us. It's lightweight, it works on Mac and Windows,
00:00:27and I'm going to show you exactly how to use it today. Now the tool I'm talking about
00:00:31is Herder. Herder is a terminal multiplexer. That means I can control multiple terminals
00:00:35from a single space. And that looks something like this. I have two Claude code terminals
00:00:39open at the top, open code down here on the left, and then down on the bottom right, I have codecs.
00:00:44Now Herder really buys us three things. One, organization. Two, the ability to easily monitor
00:00:49multiple tabs. And then three, persistence. So in terms of organization, pretty clear,
00:00:54this is a lot easier to work with versus just having a bunch of terminal tabs just out in space
00:01:00on my desktop. This is also agent forward. You can see here over on the left, I have the different
00:01:05agents being listed. And when it comes to monitoring, it will show me exactly which ones are working,
00:01:10when they're done, or if they've been stuck. And then third, in terms of persistence,
00:01:14if one of these tabs is creating something, let's say it's working on a specific feature on my app,
00:01:19and I close this thing out, like I actually exit out. Well, it doesn't stop. In fact, it's Herder
00:01:25acts as a server. So even if I exit out of that window, it's still running in the background,
00:01:30and I can very easily bring it back up. And before I dive into those three sections of more detail,
00:01:34which is the organization, monitoring, and persistence, let me address the question a lot
00:01:39of you probably have right now, which is, how does this differ from Tmux and Cmux? Tmux has been
00:01:44around for years, and essentially does the same thing in terms of controlling all your terminals
00:01:48from a single place. Well, compared to Tmux, this is agent forward. Over here, we have this agents
00:01:53tab on the left. And like I alluded to earlier, this gives us the ability to see which agents are running,
00:01:59which need our input, and which are done. So it's very easy to control not just two, three, or four
00:02:05terminals, but five, six, seven, eight, even ones you can't necessarily see visually. Tmux doesn't do
00:02:09that out of the box. Now, when we compare it to Cmux, the biggest difference is persistence,
00:02:14and also the fact that I can use Herder very easily on Windows. Now, remember, in terms of
00:02:18persistence, it's the idea that I have it working on something, I exit out in the middle of its work,
00:02:22it keeps running in the background, and I can even get back to it from my phone or SSHN. So that's a
00:02:27brief overview of what Herder buys us and how it compares to the competition. Now, let's talk about
00:02:31actually installing and using it. So to install this thing, very, very easy. It is one line of code.
00:02:36So if you head to herder.dev, or if you just head to the Herder GitHub, you will find these commands
00:02:41here. So it's literally, you know, right here. This command right here, just copy this, open it in
00:02:47your terminal and paste it in. So let's actually do that from the beginning. So once you've installed
00:02:52it, all you have to do is open up your terminal and type in Herder. And what you're going to see is
00:02:56three things. We have spaces, we have agents, and then we have our actual terminal. So this just works
00:03:03like a normal terminal. If I open up Claude code inside of there, boom, I have Claude code. And I can
00:03:09see now that Claude is listed as an agent in the chase space. So what are spaces? Think of spaces as
00:03:16projects. So I have this space, this project right now that's inside of chase. I can create a new project
00:03:23right here. I can rename it. So let's call this, I don't know, web design, save it. And also the cool
00:03:32thing with Herder is I can actually like click on things with my mouse. It isn't purely keybinds.
00:03:37So let's say I have a new project I'm working on. It's called web design test. And I also have Claude
00:03:42code opened up in there. So I have my main space and I have my web design space. And you can see
00:03:48those agents both listed here. So they can be working on different things. I can see what they're doing
00:03:53and where they're at without those terminals necessarily being visible. So spaces are the
00:03:56highest level, right? They're almost like folders we're working inside. Inside of each space,
00:04:01we have panes and agents and tabs. So I'm on tab one here inside of the chase space. If I tab over
00:04:10again, I can name it whatever I want. But this is almost like just creating a new terminal tab like
00:04:13you would inside of your normal terminal or inside of PowerShell. So if I open up Claude again, what do
00:04:19I got? Well, now I have two tabs. Easy enough. You can see them listed over here on the left. Now inside
00:04:24each tab, I can have different visible panes. So if I right click here and I hit split right,
00:04:30well then I can open up Codex. So now I have very clearly Claude code and Codex inside the chase space
00:04:38inside of tab one. So it's almost just like folders are the spaces. The first layer of subfolders is the
00:04:45tabs. And within each tab, I can have as many agents as I want working. So if we take that, you know,
00:04:52for this case, maybe in this first tab, I rename this and this is for research. Okay.
00:04:59So I'm over here in this section, this is where I'm going to be doing all my research and I've been
00:05:02doing it with Claude and Codex. And tab number two, maybe this is just going to be the dev server.
00:05:09Okay. I'm just purely going to be using this to like mess with the dev server or something like
00:05:13that. Right. So this just allows you to like very clearly organize things and put them and put them
00:05:18in different buckets. And all this is meant to fight the chaos that happens when you have a ton of
00:05:22terminals with a ton of agents doing a ton of different things. I have my spaces to keep me organized.
00:05:26I have my tabs within the spaces to keep my like subtasks within that project organized. And over
00:05:33here on the left with the agents tab, I can always tab through and see which one is working and which
00:05:38one's done. And I find this super useful if you're using something like Claude to talk to Codex and
00:05:43Codex to talk to Claude, because oftentimes we do that in a headless manner, right? I'm just having
00:05:47Claude create this headless, you know, instance of Codex and it's sending stuff back and forth and you don't
00:05:53really see what they're actually doing. But with this, it's really easy to get sort of eyes on what
00:05:58they're doing. For example, I have a skill that has Claude, Code and Codex go back and forth when
00:06:04they're having a plan, right? They do a bunch of iterations till they get on the same page. Normally
00:06:07it's done headlessly, but inside of Herder, it's very easy for me to make that visible to me across
00:06:12the entire process. So for example, let me actually show you that in action. Hey, can you use the Codex
00:06:17review skill to create a, we're using this as a demo, a fake plan for creating some sort of web app.
00:06:26I want you to come up with a plan and then I want you to execute it using the Codex review. So you
00:06:32create the plan and you send it to Codex. I want you to use the Herder skill so that you create a new
00:06:38pane inside of Herder showing me the actual back and forth so it's visible. And you might've noticed me
00:06:44mention the Herder skill during that. So the Herder skill tells your agents how to use Herder. So your
00:06:51agents can actually spawn different panes inside of Herder. They can create different spaces. So you
00:06:56don't have to manually control all this organization. Your actual agents can do it for you. And all you
00:07:01have to do to install it is paste in this one line of code inside of your terminal. You can find a link
00:07:07to that on the Herder GitHub right here where it talks about agent skills. And of note, there's also an
00:07:12entire plugin marketplace for Herder. So if you follow the link right here, you can see people
00:07:17started to add a bunch of different, you know, sort of add-ons to this Herder construct. Now,
00:07:23while this is working, let me show you some other stuff. We can tell that this particular agent is
00:07:27working because we sort of have this little yellow snake thing going on over here on the left.
00:07:32If I hit menu and go to settings, I can change the theme, sounds, toast, labels, integrations,
00:07:38and experiments. If I right click on any of the panes, I can rename them, split them right up, down,
00:07:43zoom in, zoom out, all that good stuff. So now on the right, you can see the pane that Claude code
00:07:49opened up and we have the Codex critic here. So Claude created a plan. It sent it to Codex. And you
00:07:53can see right here, Codex coming up with essentially all of its issues with the plan that Fable came up
00:08:00with. So again, instead of it being headless, I can actually see this happen in real time. And then
00:08:04it's going to send its revised plan back to Codex. Because you can see here, the verdict is revised
00:08:10and they have this back and forth all the way till the end, till they get to a proper plan. So now let's
00:08:14show off the persistence aspect. We can see both of these agents over here on the left are in the
00:08:19middle of doing work. Now let's exit the terminal completely. And despite it being on, it's still
00:08:24working in the background. So if I click on PowerShell and I bring up the terminal again, and I just do
00:08:30Herder, well, there we go. They're still working, which is great, right? So as long as your computer
00:08:38is still on, all this is going to persist, which means if you're on like a laptop or something and
00:08:43you close it, but you still have it actually running and doesn't just put it to sleep, you know, you can
00:08:47have your agents going for as long as you want. Now, if you actually want to shut down those sessions,
00:08:51all you have to do is right click it and hit close pane. And you can also shut down the spaces in
00:08:56their entirety by just right clicking on them and hitting close. Now, the last thing I'll mention
00:09:00with Herder is the keybind. So if I go to menu and I go to keybinds, you see all the keybinds available
00:09:06to us. Like I said, the cool thing about Herder, a lot of it can be done with the mouse and keyboard.
00:09:11But if you want to use the keybinds, understand it is similar to Tmux. So you essentially have to hit
00:09:16like control plus B, which is going to be your prefix mode. And then from there, there's a series
00:09:22of keybinds that will do stuff. So if it says prefix plus V to split vertical, that means I would do
00:09:27control B, way to B, and then hit V. So that would look something like control B. So now my machine
00:09:34understands I'm talking about something I'm doing inside of Herder. I hit V, boom, there we go.
00:09:39But I understand a lot of this can be done with like right clicks and that sort of thing.
00:09:42But just wanted to mention that other than that, that's pretty much like 90% of Herder. You also
00:09:47have the ability to kind of like set this up with your phone and all that. But I think the real value
00:09:52add here is when you're on your actual machine, you're running a bunch of agents across different
00:09:56coding harnesses. Like agent view inside of cloud code is great, but that only works inside of cloud
00:10:01code, right? Oftentimes I want to use codex. Oftentimes I bring in something from open code. So
00:10:06this just is awesome for the monitoring piece, for the organization piece. And I really like the whole
00:10:12persistence aspect because I have exited out of terminals before when they're in the middle of
00:10:17doing work. So it's open source. You can use it for free and it works on Windows and Mac, like I said
00:10:23before. So try it out. Let me know what you think. Other than that, I'll see you guys around.

Key Takeaway

Herder resolves the chaos of managing multiple coding agents by providing a persistent, agent-aware terminal multiplexer that works on both Mac and Windows.

Highlights

  • Herder is an open-source, cross-platform terminal multiplexer for Windows and Mac that centralizes multiple coding agent terminals.

  • Persistent sessions keep agent processes running in the background even if the terminal window is closed.

  • The Agent-Forward feature provides a dedicated dashboard on the left sidebar to monitor which agents are active, stuck, or waiting for user input.

  • Users can split, zoom, and organize terminals into projects called 'spaces' using either mouse interactions or keybinds similar to Tmux.

  • An optional Herder skill allows coding agents to programmatically create and manage their own panes and spaces, eliminating manual setup.

Timeline

Problem and Solution Overview

  • Managing multiple coding agents simultaneously often results in cluttered terminal tabs and accidental process termination.
  • Herder acts as a terminal multiplexer that provides organization, monitoring, and process persistence.
  • Unlike Tmux, Herder includes an agent-specific interface to track status and input requirements.
  • Herder enables persistent background execution, allowing users to exit the interface without stopping active tasks.

Using multiple coding agents like Claude Code and Codex often requires monitoring six to eight terminals simultaneously. Herder solves the visibility and organization problems by centralizing these into a single interface. It improves upon traditional tools like Tmux by offering a native agent-forward dashboard and superior persistence, ensuring processes continue even if the window is closed.

Installation and Structure

  • Installation requires a single line of code available from the project's GitHub.
  • Spaces function as top-level containers, similar to project folders.
  • Each space contains multiple tabs and split panes to organize specific subtasks or agents.
  • The interface supports both mouse-based navigation and keyboard-driven commands.

Installation involves a one-line command from the Herder GitHub. The architecture uses 'spaces' for high-level project organization, while individual tabs and panes provide granular control. This allows users to bucket agents into specific areas, such as research or development servers, to combat terminal clutter.

Advanced Usage and Persistence

  • An optional Herder skill enables agents to programmatically manage their own terminal panes and spaces.
  • Visual monitoring of headless agent interactions is possible in real-time.
  • Closed sessions persist in the background and can be reconnected as long as the computer remains active.
  • Keybindings follow a prefix mode structure similar to Tmux, utilizing Control + B as the trigger.

Agents can use the Herder skill to autonomously create panes, making invisible agent-to-agent communication transparent. For persistence, tasks continue running after the user exits the terminal, accessible again upon reopening Herder. The interface also includes settings for themes and notifications, along with an extensible plugin marketplace.

Community Posts

View all posts