Herdr is a terminal multiplexer built for AI coding agents — basically tmux for agents. In this video, I walk through how Herdr helps solve one of the biggest AI coding workflow problems: running long coding sessions, detaching, walking away, and coming back later without losing your work.
Herdr gives you persistent terminal sessions, workspaces, panes, tabs, keyboard shortcuts, mouse-friendly TUI controls, remote SSH sessions, and built-in agent status tracking for tools like Claude Code, Codex, and other coding agents. It also exposes CLI commands through a Unix socket, making it possible to use Herdr as an agent orchestration layer for spawning and managing multiple AI agents inside the same terminal session.
In this video, I cover:
- What Herdr is and why it feels like “tmux for agents” - How Herdr compares to tmux and Cmux - Persistent terminal sessions for long-running AI coding tasks - Workspaces, panes, tabs, splits, and keyboard shortcuts - Agent status updates, blocker notifications, and macOS alerts - Remote Herdr sessions over SSH - Using the Herdr CLI for agent orchestration - Spinning up multiple Claude Code / Codex instances from inside Herdr - Current limitations, bugs, and tradeoffs
Herdr is especially useful if you use AI coding agents and want a better way to manage long-running sessions across your local machine, laptop, server, or remote Ubuntu box.
I just discovered Herder, which is marketed as a TMUX for agents.
And what that means is that it's a terminal multiplexer, so you're able to manage multiple
terminal programs from a single screen. And this is solving a critical AI workflow
problem that I've been having, and that is being able to start a really long coding session
on my terminal, close my laptop, walk away, come back, and resume that session.
And not only that, Herder has this really nice UI, well, TUI that's interactable with the mouse,
but it also has all these TMUX primitives so that you can interact it with your keyboards
as well. It's like a win-win situation. It's fantastic. But instead of me just
raving about this, why don't I just show you what it looks like?
So yeah, this is the default screen for Herder. It's a terminal application,
so it'll work on just about any application. And you can see here that you've got workspaces,
that they've shortened to spaces here on the left-hand side. You can create multiple of these
using the new button or keyboard shortcuts, if you will. And then you can even right-click to
rename them, to close them, all that kind of stuff. Inside of these spaces, you have your
panes. Now, you have tabs inside of these panes too, or you can just right-click and split vertical,
split horizontal, all these really cool stuff. They can drag these borders too to resize them.
It's really cool. But the real kind of cool part about Herder is its integration with AI agents.
And the way that that works is you just spin one up. So let's do Claude. Let's do Codex.
And you can see here on the left-hand side here that the nav bar has an agent's tab. And when I
created these coding agents, they showed up. So they have these really cool integrations,
so that when your agents are working, it updates the status so that when it finishes,
you get a cool little notification that says, hey, come back and continue your work.
So let me show you how that works. Let's do say hello back to me after four seconds.
And immediately when I sent that off, the status will update it to working. And now
there's a few different statuses here. Idle is the default one, obviously, but it does switch
to working. And then if you have questions that are demanding your attention immediately,
then it has a blocker status as well. So let's try something like that.
Now the macOS notification usually requires a permission on Codex. And so here you can see that
it's asking for permissions. And this is a super common problem that a lot of people who are using
coding harnesses face. You just kind of start a prompt, you walk away, you come back, you're like,
shoot, it hasn't done anything. So in these cases, you get a ping from the notification. I'm not sure
if you guys heard that, but there was a ping. And not only that, the visual status updates to block
so that it's super clear that, hey, this agent needs attention. So yeah, that's a really cool
feature to have in your tmux-like instance. And speaking of tmux, like I said, it shares the same
primitives as tmux, so you're able to use your prefix keys. If you're not familiar with tmux,
basically, you're able to navigate the entire terminal application without ever lifting your
hands from the keyboard. So here I can do Ctrl-B, and that puts me into navigate mode. So now I can
use my arrow keys to kind of move back and forth from different panes, move up and down my spaces
here. You can see on the top left here. And if I do question mark here, while I'm in the command
mode, it shows me the keyboard binds. And these are all configurable as well. So if you're a
subscriber to my channel, and if you're not, you totally should, because I make a lot of content
that covers fresh software releases like this. So if you're into that kind of stuff, definitely do
subscribe. But if you are a regular viewer, you're probably aware that I have a video on cmux that
looks pretty much like this. And this is what my cmux looks like. You've got tabs on the left-hand
side, and inside of panes here, you're able to create multiple terminals. It's very similar.
But the main difference here is that cmux is a native macOS application. It has its own application
process. Herder is a terminal application, so it will run on pretty much any terminal application
that you're using. So you can pretty much use whatever you're most comfortable with. The other
thing is that cmux has a built-in browser that you are able to programmatically control. Herder
doesn't have that, but the nice thing about Herder is if I do Ctrl B and Q, I'm able to exit my Herder
session and type Herder back to get back into this session. And that makes it a persistent session.
Now, the really cool thing about this is that now I am able to start a Herder session on my Ubuntu
laptop that I have running full-time. It's kind of off the screen to my left here. It handles some
of my server workloads, and now it can handle all my coding tasks that I know is going to take a
couple of hours to finish. I can spin that off, go take lunch, come back, or if I wanted to, I can
spin it up, go to a coffee shop, use VPN to connect back to my home network, and then use Herder to
connect back into that session. So let me show you how that works. So I have a computer on my network.
I can SSH into it. It's HS Macaroni. I name all my computers in my home after pastas. I've got
macaroni, linguini, spaghetti, tortellini. I'm a huge fan of pasta, so I name all my servers after
pasta. But if I SSH into it, I can also use Herder. And this is a session that I spun up way before I
started recording this video. In this case, I've obviously SSHed into it first and then started
the Herder instance. But here's the cool thing. Let me go back out, exit out of my SSH session,
and then use Herder dash dash remote HS Macaroni. Now, this is reading through my SSH host config.
So your scenario is going to be a little bit different if you don't have this set up. But when
I do that, it creates or it hooks into that session, but directly through my host machine
without having to go through that SSH process. So this is a super nice way for me to be able to
asynchronize my coding sessions. And I'm super happy I found the software because this is exactly
what I'm looking for. So now I can hear all of you guys asking, hey, this sounds great. How do
I get started? The install is super easy. The downside, unfortunately, is that Herder exposes
a Unix socket. So that kind of means that you are limited to Unix operating systems. So that's going
to be your Mac and Ubuntu or any other Linux servers. But to install this, just go to the
website, you run the curl command, and that's it. It's really simple. But just for fun, let me show
you what it looks like. So I've copied the command. And once it's completed, all you have to do is
type Herder, H E R D R. Now, you'll notice that there is no second E to Herder. And that's because
unfortunately, in this time and age, inflation has kind of caught up to a lot of people here.
The second E is really expensive. So a lot of companies are dropping that E. So just be mindful
if you do talk to the maintainers, it is a hard time for all of us. But yeah, to start your session,
you just run Herder. And that will take us back to that session that we have created. But if you
don't have a running session, it will create a default one for you. You can list all the existing
Herder sessions through Herder session list. And if you look at the Herder session commands,
you'll see that it has all the operations to be able to create and delete and all that kind of
stuff. Now, I've got a really cool feature to show you. It's to do with agent orchestration.
So I mentioned earlier that Herder has a Unix socket. And what that allows us to do is to be
able to modify Herder directly inside of our CLI. So let me show you what I mean by that. So if you
run Herder workspace, we have list, create, get, rename and close. And if we do Herder workspace
list, it dumps a JSON of all the workspaces that we already have in our default workspace. So it's
able to create more workspaces, read workspaces. And there's also a command for Herder pain. Now,
pain is super important because now you're able to use the Herder CLI to create more pains. And
inside of those pains, you can send keys and you can also wait for specific conditions to
be reached inside of those pains as well. So that allows you to use Herder as your agent
orchestration layer so that your cloud code or codex or open code or what have you is able to
use Herder CLI to spin up more pains and instances of the coding agents directly inside of that
session. So let me show you how that works. Now, I'm just cleaning up my pains here. If you are
using this for the first time, you'll probably be wondering, how on earth do I close a pain?
You can always use that helper. So control B question mark, we'll open up the keyboard bindings
and I'll show you this already. But to kill a pain, you're going to do control B and X. Anyway,
let's open up a codex again. At this point, codex doesn't really know what Herder is. So I'm going
to tell it to go read the help on that command so that it can orient itself and learn how it works.
And honestly, this is seriously cool to be able to, you know, take a new CLI and just like feed
it to an AI and it just like learns it and kind of figures it out. Super cool. And I'm just going
to preemptively prompt it. Use the CLI to spin up two codex instances. And inside of those,
look up for the capital of South Korea and also France. Make sure to use this workspace.
Now, I don't know if that last command is going to work. Last time I tried this command,
it actually spun up a different workspace. So I do want to see if it is able to use the current
workspace. Now I do want to take this time to tell you that, you know, this is a relatively
new software. I think they've been in active development for the past couple of months.
They're gaining a lot of traction, but this does mean that the software can be a little bit buggy
here and there. So what I personally noticed that made it a little bit challenging for me to record
this video was that when I ran some of these commands inside of cloud code, the statuses were
not actually matching up perfectly. So there was that issue, but the codex was working fine. So I'm
just kind of showing you all of this in codex. But as I was talking, you can see that, you know,
codex, it was using the pain commands to split and it's creating two versions here. Obviously,
this one that we already had before. So I'm just going to close this. So this is a fun experiment.
It's using codex CLI directly. So it's not really going into the interactive mode. I did a previous
run with this where I use cloud code and I was doing it flawlessly. So maybe I'll try again with
cloud code. But this is hilarious. Obviously, when I'm recording things go wrong. I'm probably not
going to edit this out, but it's screwed it up enough times that it's spun up a third instance
here. That's that's hilarious. Let's try out a different angle. Spin up cloud code instead.
Use interactive mode. And I do apologize, guys. Cloud code did work for me in the past,
but we had that kind of like state error. So I chose to do it on codex this time. I'm sure your
experience will be a little bit better once you kind of iterate with this and kind of refine your
prompts. I would honestly recommend that you get your coding agent to read through the documentation,
read through the CLI commands and make a skill out of it so that you have a repeatable process to be
able to spin up these panes directly. Once you have that skill set up, this should be so much
easier going forward. But yeah, this looks like it's done the job correctly. It's spun up two
different cloud code instances. Look up the capital of South Korea. Seoul is the correct place.
And there's another one here. It's got France and Paris. Exactly what we're looking for. So it's
perfect. So at this point, you might be wondering, hey, what's really the point of having all these
abilities to be able to create panes? What's the point of agent orchestration? I'm sure there's
tons of different videos out there. But just to kind of answer that question, the idea here is
that you have one main orchestrating AI agent and this AI agent is able to delegate some of
these tasks to different agents so that you're able to do multiple coding sessions at the same
time. Or if you go the other way, which is to use multiple agents to improve the quality of a single
coding session, you're able to use maybe like a planning agent, a coding agent, a reviewing agent,
a QA agent, all these kind of different agents so that they kind of collaborate on a single instance.
Agent orchestration is a bit of a can of worms when it comes to like what it can do. I would
suggest that you just kind of play around with it. A lot of different coding agents are kind of
coming up with their implementation of agent orchestration, but it's just really nice having
a tool that is able to kind of support it out of the box. Now, the one downside of Herder is that
compared to CMUX, it doesn't have a built-in browser. And because of that, it doesn't really
have eyes to kind of see what it's doing. Obviously, if you have skills inside of your agent
harness that is able to use stuff like Playwright or Chrome DevTools, MCP, that will do just fine
as well. If you're running Herder inside of CMUX, you also have access to the CMUX CLI. So you can
use a CMUX browser like that. But the downside there is that you're now not able to kind of use
that detach attach functionality to be able to kind of go in and out as easily. That persistence
feature is kind of the biggest selling point for me. So that to me is a bit of a letdown, but
still, there's a work around it. The other kind of pitfall here is that my biggest complaint of
TMUX originally has been the fact that you have to learn a lot of keyboard shortcuts. Now, with
Herder, that's not exactly the case, but there's still cases you're going to want to reach for some
of these keyboard shortcuts. Like right-clicking it and splitting it is functional, not ideal.
CMUX has that button that you can press. So if you go back to CMUX here, you're able to create
splits with these buttons here. And I think that's just a nicer improvement. There's a possibility
that as this product matures, that the developers are going to add more features to make the UX
nicer around some of these kind of edge cases. But yeah, for the meanwhile, you are kind of stuck to
what it is for now. But I will say it is a nice software and it's completely free. So, you know,
beggars can't be choosers. But yeah, that's the end of this video. I do like to bring these kind
of lesser known new pieces of software to light as soon as possible. If that's the kind of thing
you're looking for, please make sure to subscribe and follow. I make a lot of content that tries to
cover these kind of content. But yeah, hope you have a good rest of your week and I hope to catch
you in the next one. Bye-bye.