Is This The Best TMUX for Running Multiple AI Agents?

Added: 2026-05-21

[00:00:00]Hi, and welcome in this new video, and I'm super excited to present one of the coolest projects that I came across recently, which is Erder. So, Erder is a terminal multiplexer. If you don't know what it is, it allows you to create and manage multiple terminal programs in one terminal. This is especially important when you have to manage and orchestrate multiple agents. So, let me show you Erder, what it is. If you take a look at the home page and scroll down a little bit, you can see that Erder defines itself as a lightweight TUI process with real terminal panes. We will come back at that in a second. Then clickable layout, agent state, detach, reattach, remote attach, and a runtime API. That's a lot, but you can summarize Erder as the terminal workspace manager for AI coding agents. And we will see why in a second. First thing first, let's install Erder, and for that you just need to run this command in your terminal, and then run the command Erder. As soon as you do that, you will land on this view where you have access to the settings of Erder. So, here you can pick the theme that you prefer. You can also choose if you want to play a sound to get notified when your agent has done anything in background. Same thing for the notifications, and then you have also some additional features for the pane lovers. But, let's stick to the default settings, and you land on that view. So, on the left you have spaces, which are workspaces, and think of a workspace at the top-level project container. So, typically, you will have one workspace per repo, task, or investigation. It owns panes and tabs. To create one, you just need to press N, and then by default, the name of the workspace will be the folder in which you run the command Erder. So, by default, that creates one tab with the label one and a pane, which is the terminal that you can see right here. So, now, what you can do is to create multiple panes so that you have multiple terminal programs, and for that, it's rather simple. You can just hit the control B shortcut, and now you are in navigation mode. So, in navigation mode, you can actually use your mouse to click on different items here. So, for example, the menu, or if you want to create a new workspace, you can click on new right here. But, most importantly, you can create multiple panes and so terminal sessions by pressing V or dash, depending on the layout you want for your panes, either vertically or horizontally or both, as you can see here. I have three panes, three terminals. To switch between the different panes, you just need to click on whichever you want, and you can also create a new tab just by clicking here.

[00:02:35]And then, let's call it two. And just like that, you have a new tab with just one pane. If you want to come back to the previous tab, you just click on it, and then you are back to the three panes. This is one thing that I enjoy with Ergo is that I can use the mouse or the keyboard. Speaking of the keyboard, if you want to use the shortcuts, you can do that. Enter in navigation mode with control B, and then if you press question mark, then you will see here the key bindings that you can change if you want. In addition, if you want to zoom on a specific pane, you just select whichever you want, and then you go back in navigation mode, and you press F. And now, I just have this pane in full screen.

[00:03:12]And if you want to come back, again, navigation mode and F again. Okay, this is pretty cool, but what about agents?

[00:03:18]Because I think this is where Ergo is actually pretty nice.

[00:03:22]Well, if you want to start an agent, it's pretty easy. You just select the pane that you want. And for example, in this one, let's say I want to use Codex.

[00:03:30]So, here I'm going to type Codex.

[00:03:32]And then, in this pane, I want to use Cloud. So, here I can just type Cloud.

[00:03:38]And again, as you can see right here, I have Codex in this pane, and on the right one, I have Cloud. But, most importantly, if you carefully take a look, on the left, I have the agent list. So, here, whenever I start an agent in one of the bands, I will see the corresponding agent in that list.

[00:03:57]So, here I have Codex Claude, but for example, if I'm on another tab, so let's say this one, and here I open another Codex agent, you can see here that on tab two, I have Codex that is running.

[00:04:10]So, this is great because I don't know about you, but if you have many agents running in parallel, you will be able to see all of them right here even if they are across different tabs. And another thing that I truly love about Erda is the state of your agent. So, you see here Codex is idle, same thing for Claude, and Codex here on tab two is done. Erda gives different states for your agents. So, typically you have blocked, working, done, idle, and unknown. So, let's play a bit with them just by typing a prompt in Codex, for example. So, let's say wait 5 seconds and say hi.

[00:04:51]So, as soon as you do that, you see here you have working, and we are waiting a little bit, and now the state is idle. Why it is idle? Because the agent returned an output, and we saw that output. In fact, if we go to the second tab, and here do the same thing, so wait for 5 seconds and say hi, and then come back to tab one, you will see that Codex in the second tab is working because it is going through the prompt that we sent. But now we have done because we haven't seen the output of Codex yet. If we click on tab two, now it is idle. Let's do something different so that Codex needs an input from us. So, we can say create a repo in that folder.

[00:05:39]And then let's go back to Codex one.

[00:05:42]So, by the way, you can just click on the agent that you want right here. And did you hear the sound? Like, I'm not sure if you heard it, but basically whenever an agent is done with something, you will hear a notification bell. And I don't know about you, but damn, I wasted so much time just because my agent needed my inputs and I didn't know about that. So now with Erder, you can just hear a sound that tells you your agent is waiting for you. And you see here you have blocked, which means Codex is waiting for your input. So let's click on the agent and here either we approve or this is approved to create that repo. I'm not going to create that repo, this is just for the example, but I think this is a really cool feature.

[00:06:23]Speaking of cool features, there is another one about persisting sessions.

[00:06:27]So basically a session has multiple workspaces, tabs, uh agents, and so on.

[00:06:34]And so you don't have to keep Erder in foreground to have your agents running.

[00:06:40]So for example, let's say here again in Codex, I'm going to say wait for uh 15 seconds.

[00:06:51]And then Codex is working. I can just hit control B and then Q to quit Erder. And as you can see, now I just detached Erder to my terminal, but I can come back at it if I want by typing Erder like that. And I go back to where I was. So you see here Codex is now idle. So you can have multiple sessions like that. If I hit control B and then Q again and then Erder session list, you can see here the default session. So you have multiple options with the sessions. So for example, you can either attach an existing one or stopping one or even deleting a session if you want. Also, you have access to many other commands using the command line interface of Erder. So in fact, if you type Erder and then {dash} H, you can see all the commands that you have access to.

[00:07:48]And speaking of those commands, this is where it's going to get very interesting because remember that I told you Erder is a terminal workspace manager for AI coding agents. So, you just discovered that you can manage all of your coding agents through Erder. But guess what?

[00:08:04]Your coding agents can also manage Erder for you. How? By using a dedicated skill. Here it is. You can find it in the Erder repo. Basically, it tells your agent that it can control Erder from inside it. Remember, you have to use that skill inside Erder so that your agent can manage workspaces, tabs, split pans, spawn agents, and so on. So, let's see that skill in action, but first we have to install it in Codex and Clooder or whatever agent you want. For that, you can run this command.

[00:08:39]And then wait a little bit, then pick Clood code if you are using Clood, and hit enter, then global, and proceed, and wait a little bit until the skill is installed, and it's done.

[00:08:52]Back to Erder. What I'm going to do here, I'm not going to spin up AI agents. In that case, I'm going to let Codex to use the Erder skill in order to spin up agents for me and do whatever, uh, you know, I want using Erder. So, let's do it. Here, I'm going to ask, let's say something like, uh, using the Erder skill, I want you to uh, create three, let's say, pans with one on another tab, and then in two buns, not this one, I want you to spin up Claude and Codex for the other one, or respectively.

[00:09:57]And for each and each agent should generate, let's say, two Italian dishes. Okay, so obviously that's a pretty weird prompt, but in the real world that might be very useful for you. So, let's hit enter and wait a little bit. So, for now we have only one agent. And you can see that here is using the Erder skill, as expected. I actually like to emphasize that it has to use the skill.

[00:10:28]So, let's wait a little bit. Now the state is blocked because Codex is waiting for me to give an input. So, I'm going to say, yes. You see, so here it is listing for the buns that we have.

[00:10:38]So, we only have one, which is the current one. So, here we have a second bun and then a third one. And then a new tab, you see Italian extra, and we can click on it. We can see here another bun as well. Okay, so now we are getting Claude on this bun. And you see we are getting the different agents as well on the left, as expected. Okay, so now we have three agents. One which is using Claude and another one which is using Codex, as expected. And you can see here the prompt that we asked to use for Claude as well as for Codex.

[00:11:08]And we just got a notification bell, so you see that Codex has successfully generated the dishes. We can take a look at that. So, if we eat the tab, here we have the two dishes.

[00:11:19]And then if we go back to Claude right here, we have the two dishes as well. And Codex is still working, just to check if everything has been successfully executed. Okay, so So, all I wanted to show you about Erder. I think it's a great tool. I'm definitely going to keep using it, especially as I have to use different agents.

[00:11:39]You know, it helps a lot. Also, I highly recommend you to take a look at the documentation. For example, you can see here how to get remote access over SSH.

[00:11:48]So, there are plenty of things that you can still do. Also, the CLI has a lot of commands that I didn't cover in that video, but if you want to learn more about Herder, please let me know in the comments below. If you want me to test anything else, again, let me know in the comments below and I will be happy to help. So, that's it for this video. I hope you enjoyed it and I see you for the next one. Thank you.