Mosh, Caddy & Tailscale: A Remote Dev Deep Dive

Everybody JS building tools for agents.

Nobody is building actual products. Once I got TMUX, and I was like, oh, I get it that I need these long running sessions. Then Herder for me was kind of like a,

oh, this just works. Let me tell you the amount of times I've had to reboot my computer in the last, like, six months. Unbelievable.

I I also don't know why I'm still in the terminal.

I I hate the terminal, but it's I I can't get out of it.

Are you going to be running your entire dev process on a remote server somewhere? Well, we're gonna be talking all about the flows for that, the tools for that, the pros and cons, and if you should be doing it locally or in the cloud, all of that good stuff today on Syntax.

Still talk about this. I think the whole, like, remote workflow as well as, like, not just having a thousand terminal tabs open, there's so many products in this space of how do you get a better UI Node that it's not, like, just everything in an IDE, but, like, you need some sort of nice UI for managing all of your agents and all of your folders and whatnot. And there's a bazillion products, and they all slightly overlap each other. So I'm excited to dig in and talk about all these. Yeah. Me too. I've been running this, setup now for quite a bit, and, I'll talk about my setup. We can talk about some of the the tools, the flows,

why you might even want to do this. And then, again, we're gonna be getting into one of the, maybe some of the pitfalls. And, maybe if you have solutions to these pitfalls, comments below or tell us how you are using these types of tools. So let's talk about, first and foremost, like, what this means, why you would do it. And and, really, what it is is that you'd be running your dev process on a machine that is maybe not your main machine or your laptop.

And this is something that is gonna become more popular. It's already becoming popular.

You see memes about people not shutting their laptops. You see software that is written to specifically prevent your computer from going to sleep when you close it because we end up having these work processes that are now detached from when we are interacting with it. Like, you prompt an agent to do something. You wanna be able to close your laptop and have that continue working. Right? So one, you can now close your laptop.

Process can be running the whole time while you are not necessarily at your machine, so that's a pro here. Two, better battery life because you're not running 8,000 different things on your laptop at all times.

Let me tell you the amount of times I've had to reboot my computer in the last, like, six months.

Unbelievable.

It's just all these little tattered ends installing endless new apps to try. Things are running in the background.

I'm I'm ready just to start over, but a lot of it is due to to having having all these agents running on my computer, and they have just child processes that go rogue.

Yeah. And and, actually, what's so funny here is that, you think Apple isn't paying attention? The latest version of macOS Wes, the beta, has an a new checkbox in settings that is reboot after a power fault. So if, like, there's a power fault and you're away from your remote machine, it can can reboot itself, which it can do before.

I think it could maybe back in the day, but they took it out at some point. So, specifically, I would be so annoyed if I was on a trip or something. And even before I ran my dev process on this this other computer, I've ran my Plex server. And so we'd be, like, out of the country and trying to access the Plex server, and it would be down. You'd be like, oh, come on. You know?

So now you're in it. A thing before? So if you had, like, a a power fault, then it would just not reboot. You had to physically press the button.

Yes. Very ambitious. Yes. Unbelievable indeed.

So better battery life. You can close your laptop, actually. Another one is you can access these processes from any machine. And in this case, we'll be talking about using Tailscale to do so, but, it depends on how you host your stuff. But it it really is awesome to be able to throw up your dev process on your phone, your laptop, your iPad, any computer on the network and just, like, be able to access that stuff for testing or whatever. It just it's it's just nice to have. Another thing you get out of this is, in this setup in particular, is HTTPS Node dev. Now that's not necessarily something that is unique to running things remotely if you were using Caddy or anything like that before.

I know, Wes, you've done HTTPS in dev this way for a long time. So Yeah.

That is a there's a number of great things that'll, comes from being able to run things in a secure context, including browser APIs that only work in a secure context. Right? Yes. Exactly. The the biggest thing that people hit is there's so many browser APIs, like being able to access webcam or microphone or,

I believe on Safari, you can't access local storage.

And And, like, HTTP only cookies too in Safari. Yeah. Yeah. Exactly. And those things work on local hosts because local hosts, even though there's no, like, icon and HTTPS, it's still a secure origin. It's it's considered a secure origin. However, as soon as you wanna bring it up on your phone, even if you're on the same network or as soon as you wanna share it to to Scott or something like that. I do this a lot with CloudFlare Tunnels where I have something running on my local machine. I wanna be able to have other people play with it. Then then you're out of luck, and you gotta go, now I gotta figure out how to how to expose this thing. So it's it's kinda nice just to do it from the get go. Totally. Yeah. There's a lot of pros here. And, like I said, I've been running it this way for a little bit, and it's great. And another thing here too is that, like, you can access your agents.

You can access your AI tools from your mobile devices. You can access them, from your tablet. You can access them away from your computer when you shouldn't be working. I'm in the gym, and now I gotta prompt my agents because I am addicted to it.

It's all possible with this.

Well, I'm excited to hear what your your thing is because I've had sort of a half arsed stack where I've got an old laptop here beside me, and then I also have, like, a like, a home server, home NAS that's running, like, Docker. Right? And I have I have processes running on both of those as well as most of my stuff is still just on my computer, and I I'm hitting all of the pain points that you you just said. And I have, like, kind of a I've got CloudFlare tunnels and whatever, but I'm not happy with any of it. And you've spent a lot more time kinda going into the what tools you need to get this to, like it's something they're actually going to use day in and day out. So I know, like, we were at a conference, and you were doing your slides from Amsterdam, and your slides were living in Denver on your Mac mini in your in your house.

Yeah. It's I I trusted my setup enough to to load my slides remotely, from my Mac mini and just, like that was dicey. That was almost as dicey as installing the, first developer beta on my machine the a couple days before getting on stage for Yeah. How many people were there? The audience.

A lot of people. At least a thousand. Yeah. Yep.

And it all worked. So, that means that it was a valid approach. So, typically, what I'm doing is I'm just creating projects, like, normally on the Mac mini, whether that is I'll tell you in a minute. Here, I use a tool called Mosh, but you're essentially just SSH ing into the computer, creating the tools with the CLIs that you know and love. And then I assign a domain to them. So I give each project a unique port, and I basically assign a domain to it. And then I connect remotely to run my AI. I connect remotely to code via zed and those types of things. So, that is the general flow. It feels very similar to working on your local machine.

After you get things set up, you're just visiting, you know, actual domains instead of, just a local host port whatever. Mhmm. And you could do IP port whatever.

Again, opening Zed and just saying connect remote, all the files show up. It's, like, working locally. And if you're using a TUI, whatever for your AI,

again, it just feels like exactly Yeah. The same. Right? And for you, a project is that like, that's a folder. Right? Like, that's kinda something I'm thinking about lately is that, like, most of my stuff is is one project, one folder. But every now and then, I I have things that are just, like, global or or things that span multiple projects.

And, I think about, like, how would I best link these two things together, rather than just give it a a path to where it lives on my machine.

Yeah. I mean, it it all just lives in folders for me. Everything is in a site's projects folder. Yeah. And then if I do have the AI agent go outside and look at another project, it's usually just doing that, since I'm not running these sandboxed in that kind of way, which

is nice sometimes. Okay. So let's talk about the actual tech that you use to get this this done. What's your stack?

My stack. Yeah. This is the new stack, folks. This is, I don't have a a fun name for this because these are all kind of, constant letters. You're talking about it. I'll come up with a cool name for it. Okay. So, one, the connection protocol I use is MOSH. I believe it is a is it a protocol? I don't know. Either way, I connect to my remote servers with something called MOSH. It is basically a drop in replacement for SSH.

The benefit there is that when you mosh instead of SSH into something, well, what happens? It stays connected. If I close my laptop and reopen it, it reconnects automatically with, like, no lag time at all. So I don't have to reassess agent every single time I'm doing anything. And, actually, if I have, like, a long running TMX or or or JS I'll talk about a herder session going, I can close my laptop, open it, and it's gonna stay connected to that all the time. MOSH reps. Now you do need MOSH installed on all of the machines you are moshing into, but, a lot of the tools these days, support MOSH just fine. It's great. Very nice. Another one is Caddy.

So Caddy, Wes, you've used Caddy for a long time. Do you want to explain what Caddy is to the the folks here? Caddy is essentially a web server that you can use to

it's it's like Nginx or or Apache. Yeah. Yeah. It's what most people use it for is for proxying through to other applications. So you would have Caddy running on your local machine or on whatever machine that it is, and then it will it will spin up a web server and take in requests, and then it will then be able to route those requests. It can, of course, like, serve up files and whatnot, but what most people use it for is they say, alright. I'm going to like, I have tips.local host on on my website. And then what that does is it's a a caddy server that proxies tips.local host, gives an SSL certificate. You Node, you got full HTTPS, and then it proxies it through to, like, local host colon four nine two two. Right? And that's really nice because you can have multiple rules. You can get a a really extreme caddy server where you can do, like, tips. Local host forward slash Scott. That would may maybe in that case, I wanna route it to a different server. Maybe in that case, I want to proxy WebSockets.

It's a fantastic little web server for that. Yeah. And,

AI is very good at Caddy files, because Caddy has been around for a long time. So you can always just tell Caddy, this is what I need to do. In fact, I did exactly that because I have a the next tool in here is Tailscale, and Tailscale, uses WireGuard protocol to connect your machines together. And what's great about Tailscale is that to do any of this, it is freaking free. It is free, folks. And in fact, so far, all of these tools except for your AI inference are free. Free 99.

Okay? So it's all set here. So the, Tailscale, you install Tailscale on the devices in your network. They all run Tailscale, and then they can all talk to each other with, either IP addresses that are local that you could connect to directly. Or in my instance, everything just has a name. And people don't reassign their names in Tailscale because it, like, by default, gives you some long ass terrible name. Yeah. But in the Tailscale admin, you can change the name of all of your devices.

So my NAS is just NAS. My mini is just mini. MacBook Pro is MVP.

So now when I mosh into something, it's mosh, Scott Tolinski at mini enter. That's it. Boom. And that's so dope. I don't have to remember, IPs. I don't have to, you know, have that stuff stored somewhere. It's just easy. Yeah. And with tail Tailscale, I I then am able to connect, essentially, these addresses with Caddy.

So my domain that I'm using, which JS, this is actually a sick domain, Wes. It's robo.online.

So I can have, like, a i.robo.online.

That's my and these are only routed locally through my network. So Wes couldn't access a i.robo.online.

It's not gonna do anything because all that's doing is connecting to a local tail scale address. Yes. So I have these. I know robo.online are all going to be my tail scale network,

apps. That's that is different than a lot of the other approaches, which is I have all of these applications, and I am going to expose them to the wide Internet, and then you have to put some sort of authentication or access control in front of it, which I think that this is really nice because, like, even if I wanted to like, I'm sure there there is a way with Tailscale that you could. Like, let's say you have an app running on your Mac mini and you wanted to give me access to it. There is a way to, like, poke through that and expose it to the the global web. Right? Yes. Yeah. It's called a funnel. Okay.

So, CloudFlare has tunnels.

I think it's it's called funnels, yes, in inside of, in in Tailscale. Yes. It's very possible. I keep everything on my local network, though, just as a a Vercel. And, again well, not even as a serve. I I, again, I just have it all through this Sanity reverse proxy. So do you mind if I,

like, rattle off a couple alternatives for each of these pieces in the stack that you're talking about? Because, otherwise, the comments are gonna be why not. So one Caddy alternative that everybody talks about a lot is it's it's not necessarily like a a full Caddy replacement, but it's it serves similar purpose that for what we're talking about, and that is portless. It's from Yes. From Vercel Wes you can just assign each of your apps one single port, and then you have a single portless instance running it. You can assign different subdomains.

I'm not I'm not a user of it. I prefer my my caddy setup, but it's, I know people are crazy about it. Portless is fine, but, like, hold but, I I I think that's something to think about is the Vercel Bos stuff

is oftentimes just kind of thrown up there.

It's a like, it's a lab.

Like, they're testing things.

Yeah. Wes Caddy has been around for a long time, and it is battle tested. And, again, it's more full featured, and AI is really good at writing Caddy files.

And then you look at something like Vercel did that zero lang. It's like their their, you know, AI language, and it hasn't been touched in several weeks. Like, it it's like they did that initial post and then forgot about it real quick. So I I think there was a case to be made about go.

I, To what? To portless? No. To zero lang.

What? Yes.

Okay. Well, before that, the last commit was

three weeks two weeks ago or whatever. So know if it's if it's serious or if it's just a Labs project, which, I think it is. So I wouldn't give them too much guff about that. They're certainly exploring, and I think the portless one is pretty

stable.

The point being is that if you're picking a tool for this, picking this this battle tested tool that has more features and is AI JS, like it's in the training data and all this stuff, like, just go with that. There's no reason to pick up the labs thing. Also,

alternative to Tailscale is Cloudflare Tunnels and Cloudflare One. Cloudflare Tunnels allows you to poke it through out to the global Internet, but then they also have Cloudflare I think it's Cloudflare Node. Yeah. Here, I've got the app on my phone here. So when I have the I can't get it to focus.

Just imagine seeing a Cloudflare application. There we go.

So when I have that turned on on any of my machines, pnpm it running on my NAS.

Then I'm, like, sort of in the the Teleskill network. It is I would say it's very complex, and Tailscale is significantly easier to to get set up. I, I only went the, like, Cloudflare route because Scott was all in on Tailscale. I was like, we should have, like, a we should both do a different one and balance take on it, but it was it was not easy to get set up. And and even then, there was, like, some weird issues with accessing some local host on my phone. So I wouldn't, I would probably say check out Tailscale first. Yeah. And the nerds in the chat are gonna say just use freaking WireGuard.

WireGuard being the, I believe, the protocol at which Tailscale works. Let me tell you about Tailscale. Tailscale has user management.

It has device management and white listing where, like, it pops up in a nice UI, and it's like, this device is trying to join the net. You don't wanna accept them. You wanna give them a new name. There's a whole host of cool magic DNS stuff. I will tell you, Tailscale rocks, and the free tier is excellent. So The free tier, you get six users.

So Yeah. That's that's your whole your whole family. That's my whole family. I got we got six people in our family.

I could even get for me, it's like a good access to my docs. So I don't have I don't have a a I don't have two additional kids, but I do have two dogs, so they can get access to my Tailscale network.

Tailscale.

Tail dog tail? Okay.

Let's talk about the next tool on my list is Zed. It's a text editor.

Wes JS. Not happy with the the tail. No.

Continue.

Not happy with it. Hey. Yeah. It's you know what? You gotta gotta, you gotta throw some balls out there sometimes and see if they get some hits.

Baseball.

So Zed. Zed is a, text editor that I like to use, and the reason I like to use it JS the remote, setup JS really stinking nice on it. I just wanna do a new remote connection. Yeah. It has SSH built in. And then, again, I just pop my details in there, and I can browse any of my projects. I can quickly and easily connect to them as if they were local projects.

And then bingo, bingo. I got a text editor right there to connect remotely, and it acts like it's freaking local. So, shout out to the Zed team for making that remote process nice. I need to try that because I

had been using the, like, a Versus Node remote or the cursor remote, where the editor lives on on your machine. Like, there is a way to, like, put Versus Code on your machine, and then you simply just visit, like, a web browser and use Versus Code. But the way that I was using it, I was using it on my my machine. I was SSH ing into another machine and working on it, and it was like it kept disconnecting and, like, it wasn't a great experience. And then I found that the, like, the hurdle to getting in there. So I need to try so you're saying that the Zed one works with MOSH because MOSH is just Deno. I think the Zed one works with just SSH.

Okay.

So what are you using MOSH for? They're just just for terminals?

Yeah. Okay. I use MOSH to connect via terminals where I run all my agents. So we'll talk about that in just a minute. But the, like, the re I need JS Yeah. Automatic. Anytime I need to term into my remote machine, I'm using MOSH. Okay. Okay. Makes sense. Yes. Yes. Yes. Yes. Yes. So, again, Zed for editor. I like that a lot. So let's talk about those remote sesh. So I use, number one, I I use Mosh to mosh in. The terminal app I do use right now is Ghosty. I do like Ghosty.

And so I'm Moshin' in there. And then I'm using something called Herder to manage all of my individual terminal tabs and sessions and all kinds of stuff. So you create one or as many herder sessions as you want, and herder is basically TMUX.

And so TMUX, herder, these types of things, the reason why you would use that is because you can create, different panes, different tabs, different sessions, all that stuff.

And then if you were to leave that session and come back, all your stuff is still there. We know that, like, if you are ESLint into something and maybe running an open code session from SSH, you close that session. That open code session is gone. So with something like Herder or TMux, you get those long running sessions that just stay there, and all your stuff is exactly where you want it.

The benefits for Herder over Tmux is that Herder has this unique little sidebar that reports the status of any of your agents in any coding agent platform, whether that is even, like, Hermes or OpenCode or or Pie or anything.

It it shows you the status. So whether that is blocked, so if it needs your attention, maybe it needs approval for something or it's running or it's done or it's just, you know, just sitting there, you always can look at the sidebar and instantly see, is there something that needs my attention? Before Herder, I Wes, ding, tabbing into different Oh, man. Just tabbing into different browsers and being like, oh, so does it need my oh, this one didn't start because it needs my attention. This thing has made that so much stinking better.

It's just the best. I love Herder.

Warp terminal has had, like, the agent status, for a while. So I I I quite like that, but I've been using Herder for only about a day now since Scott was telling me about it. I've been using Tmox for about a month since we've had Ben on, and I absolutely love it because it makes, having multiple, like, terminals open and having your agent be able to do it. Because often, like, a lot of people's process might be this. Like, you say, okay.

I want you to do something. For example, I have this little NFC reader right here, and I wanna be able to, like, read and write from this NFC reader. Right? And and to do that, it's a it's a terminal process called Npm three.

And I don't like being like, can you just do this? And then you don't get you don't get insights into it. What I like to do is have one terminal open with the actual process and then another one open with my agent, and I can tell the agent to do things, and it will run it in that other one. So watch here. I can say say, like, scan the card there. And what it will do is it will just tell the other terminal here we go. H f search. And now it's it's running that search. It's running against the hardware on my desk right now and and bringing back the information.

I absolutely love that about TMUX, and I just didn't the thing I didn't like about TMUX is that you can't just, like like, click and drag. You know? Like, here, I can just, like, resize things.

Yeah.

Absolutely beautiful. And People got really mad about that when I said that that you can't click and drag or any of this stuff. They're like, g mox can do all that. It's like, yeah. But you gotta configure it. You gotta go back and forth with it. I gotta turn on mouse support.

Just give me a freaking working under the box. And You know? I I also don't know why I'm still in the terminal.

I I hate the terminal, but it's I I can't get out of it. So, like, I've tried, like, solo, which is, Aaron Aaron Francis's, like, application.

I've been trying to get into, Conductor.

It's a little bit different. Like, it's a kinda Gitwork tree Bos and projects, but you can have multiple projects going at once. Like, there are lots of very good, well polished UIs for running multiple coding agents, and I really like that. But there's just something about my brain that JS whenever I wanna do something, I go straight for the terminal, and that's where I like to live, and everything else needs to kinda go over from there. I don't like going to an app and opening up a project first. You know? I wanna go to the folder

and and do everything from there. Yeah. I I there is something great about, look. Because I I'm the same way. I'm a GUI guy. I've said it many times. Yeah. But, like, I there JS something nice about moshing in and then having access to all of my tools just there, whether that is Cloud Code or Open Code or whatever. That's all just there, and I don't have to have a separate Open Code server running, and I don't have to have a separate, Cloud Code server running and these things that I have to then open up on my desktop remotely. Like, it's all just there when I need it. And so I will sacrifice some of the niceties of a GUI to just to have the ease of having all my stuff just here inside of Herder. If there was a truly like, if I could get if I could just get OpenCode that had a my Cloud Code subscription in it like I used to, I would just use that for everything, honestly, but

I can't. So I gotta run Cloud Code. I gotta run OpenCode. I gotta do all this stuff. Yeah. I also really like the, like, whole idea of, like, workspaces, which JS, like, you could have, like, a personal workspace and a work workspace, and then you can have multiple tabs inside of that, which is kind of just different layouts of it. So you might have, like, logging up top, and you might have your agent on the left and your actual terminal on the right. I really like that. And and taking it a step further of not just running it on your, like, your computer or, like, being able to, like, close the entire thing and then just reopen it and use everything is still there because it's all on, their machines. The amount of times I don't wanna reboot my computer because I have, like, 18 different Warp terminal tabs open at once, I'm just like, I don't feel like setting this all up again or figuring out where don't have to. One of the biggest problems I have is Wes on my computer is this this the code that I wrote? I can't I I know I wrote this at one point, but I simply cannot find

where this thing is on my computer. Yeah. And brand new, hot off the the presses in the past couple of weeks is that Herder has launched plug ins, and there's a marketplace for plug ins now. So I don't even have any plug ins installed.

But, in the age of AI, of course, there's, like, a thousand plug ins already automatically that pnpm existence for this thing. So, man, I love it. And there's just so many little niceties like the automatic Wes you select text, it automatically copies it to your clipboard again. That.

I love it. Tmux folks. Yes. You can do all this shit with Tmux. Who cares? It's just out of the box and works. I got tabs. I got all my stuff in here. Node I got plugged in. It's just lovely.

The the, like, copy to clipboard drives me crazy because I am just Really? Incessantly highlighting text. I can't Yeah. I can't sit still. Wes, I just sit there Clipboard is free, brother. No. It's Scott. Because then my clipboard history has, like, 80,000 things that I don't care about. And then I use my clipboard. I I no. No. No good.

I use my clipboard history as, like, a I I search it all the time. You know? I know that I copied it at some point in the last week. It's gotta be in my clipboard. Yes. I don't know. I like it a lot. So for mobile, I do Termius.

This is an app I've been calling Terminus for a long time for some reason. But Termius, you can see I have my herder loaded up inside of here. And the cool thing about Termeus is that they have this bar just above the keyboard that you can customize with your macros and stuff. So I have mine set up to make it easy for me to get it from workspace to workspace or next agent to next agent. And then, again, you just type in this bad boy, and all of your sessions are here. Now I the one thing I haven't solved about this ESLint case anybody wants to solve this is, like, touch support. Like, I would love to be able to just touch on the agent that I need to access at any given ESLint, because then they're all just right there just staring at you. Like, can you touch them? So I can't do that. I have to run the stupid, you know, control b. They haven't hooked up because you can click them.

But they haven't figured out how to, like, transform a touch to a click.

See, this this is why why are we doing this in the portal? So Like, what? Come on. That's what I was telling ESLint desks is that the GUI is so much better. And I don't. And I say this as, like, a heavy terminal user that I just I'm like, why? And I've tried all of the, like, googies, and they just have they're just not as sticky. I don't know what it is. Yeah. So, folks, if you

if you have a solution to this, I'm gonna love to hear it because this sucks to do. And in fact, I have a list on here. It's like annoyances. So, number one annoyance, screen tearing in Hurter.

Have you experienced screen tearing in Hurter?

What what is that? Like like a rendering thing? So,

yeah, sometimes what I get is that, like, the text, is get, like, from some some tabs bleed into other tabs, so then I have to, like, resize the window, and then it reshuffles it. It's, like, just too kind of annoying stuff.

Yeah.

So I I do get some of that sometimes, especially when I am going from mobile to desktop because that's, like, reshuffling the the session and and things like that. So that that's an annoyance. Again, the mobile sucks for for TUI. I I still you can't be convince me that Termius is really, like, a great experience. It's like a a an evil that I have to do if I want to prompt away from my desktop. So I had asked on Twitter what everybody is using in this sort of space. Right? There's lots of TMUX, lots of CMUX.

Andres, who's the guy behind COOLIFY, is building gene.build.

You know? There's a conductor JS a really good one. Soloterm, we talked about.

Crazy space right now, I think. And they all, like, overlap different parts. You know? Like like, whereas, like, Herder is a bit more terminal SSH based where where it's just, like it connects to remote things. A lot other ones are I'll I'm gonna run them all locally.

This Herder one is, like, the most sticky that I I did. Like, I I read the docs in, like, in, like, like, fifteen minutes. I was like, I get it. You know? I I totally get this, and this

this feels really good for me. Yeah. For me, again, like, I was one of the things I didn't resist it necessarily, but I didn't get it. And then once I got Teamux, and I was like, oh, I get it that I need these long running sessions, then Herder for me Wes kind of like a, oh, this just works. And once I got that flow, it's like, alright. I guess I'm just living in the two e for now just because it all all feels very nice.

Another annoyance that I have with this flow and setup is that if you need, like, browser testing, which we often do, Chrome DevTools MCP and agent browser, those types of things. I like Chrome DevTools MCP more than agent browser specifically because I like to see the little window open and see the stuff do the stuff in front of me, and, like, that's just the flow that I've established.

But if you're running that can be head headful? Can it? Yeah. It's just not by default. Yeah. I I just haven't yeah.

You know, Chrome Dev Tools MCP has been so effective for me that I just use it that way. But, like, now it's running on this machine, and you can kinda even see maybe some some ones in the background here if you're watching on video of it working on some stuff. It's working on Better Sanity for me. And, Better Randy's gotten really sick, Wes. I'll I'll, I don't know if, I'll I'll show you some time here. But, it's not running on your local machine. So there are some things that and, like, even, like, the auth story there with these things. Now granted, I'm not doing auth in the right way with these headless browsers. And so, folks, how are you handling auth for headless browsers in this type of AI agent world? Because outside of giving it some, you know, fake local credentials that it can fill in, I think most people are doing it via just straight up, the API is validating with real tokens.

Yeah. How are you doing that? Because I'm not doing it the right way JS really what it comes down to. Yeah. I I keep dancing between,

like, an agent browser. I tell it to fire one up, and then I will just manually log in versus, like, I know you can get this is actually something that's disabled with the Sanity Enterprise cloud code, is the cloud code Chrome extension.

Mhmm. I'm not sure why that's disabled, but, like, it can, like, control your browser. Because, like, part of me is like, yeah. I wanted to have access to my main browser so it can do have access to my, like, logged in things. And then the other part of me is just like, I'm not sure if I trust that just yet. I I prefer to just simply spin up a headless or headful, agentic browser and then have it be able to to puppeteer it. But, also, like

the I do prefer head fault to headless. The,

what, Cloudflare turnstile is getting pretty smart about having those things open. Right? I I was looking into how these CAPTCHAs detect if you're running a, like, puppeteered browser, and it's crazy. They throw they throw a tiny little debugger statement in a file that does not exist. It's crazy. If you watch it, you can't even pause it to, like, disable that

because it's they have some pretty nifty things that they do around that, which is good because you don't want you want websites to be able to stop bots as well. Yeah. I think a big thing here too is that these are all, like, free services. I know there's things out there that can run browsers in the cloud and stuff and and do all that. But, again, I want something that just runs on my sheen machine and is nice and easy Yeah. And all that good stuff, which is a big part of this whole workflow JS that it all runs on my stuff. I'm kind of sick of the effing cloud, man.

Bring on the cloud. We need more cloud.

Somebody did send me this Muxy, muxy. App. I've not tried this yet, but it's a terminal multi multi

Oh my gosh. But it

has an iOS app.

Interesting. Okay. Node I'm listening. Yeah. Yes. Okay. Lightweight. And it is efficient. Libghosty.

Yeah.

Meaning that, like, it it uses the underpinning of the ghosty terminal to render.

Right? So, you know, it's it's probably performant.

How does this work? Control your terminals and monitor sessions remotely with the iOS and Android companion

app. Does it give me my agent status? Like, I'm seeing, like, projects. I'm seeing, like, an open code style thing where your projects are on the left, and then you have tabs. Yeah. Like, I want a big old listing of all of my projects all at once, all the stuff that's going on. So that way, I can not focus on one singular thing at a time, and I can focus on 20 things at a time. Yeah. Just where where does this fit in? I mean, it seems like kind of like it fits in kind of like Like, it's it's like a Tmux alternative. Right? Yeah. And that's the one thing is that, like, Herder is a Tmux alternative as well. Yeah.

And so when people suggested what was it? Sea mux? Yeah. Man, when I tried to find CMUCs we talked about CMUCs on this show before. When I tried to find this recently, I found GMUCs. I found DMUCs. I found I think you could just take any letter.

Here's GMUCs, Steer every agent from any device. Same thing, man. There's just so many of these. CMux, though, to me, is just like another it's a terminal app built on Ghosty for vertical tabs. I just this to me No. It has a built in browser.

Okay.

But it's not a multiplexer.

There's an iOS app. There's SSH, but then it oh, no. There's skills,

workspaces. This to me seems like if you were running all your stuff on one machine, yeah, that's fine. But in terms of needing a multiplexer, I don't see this being it. So people who would say, why not CMux? It just to me, it's not the same thing.

GMux, who who knows? Muxy, who knows?

There's so many of these. This seems like the same thing as Muxy, but also has a a browser built in. Right? You can have multiple terminals. You can have workspaces on your left. You also have a browser built in, and there's a companion iOS app. No. But Muxy is a multiplexer.

What does that mean? Oh.

So you're having these multiple terminals.

Okay. Yeah. Yes. You're having the sessions that you can connect to like TMUX.

Yeah. So, again, there's a lot of stuff out here, folks. What are you using? Let us know. Another consideration, Wes, that we haven't gotten into at all that I know people are gonna talk about is work trees.

Do you use work trees at all? And if so, do you do anything to manage work trees? Yeah. I've been getting them into into them a little bit more.

I mostly use them with, like, remote agents. So if you have cloud code in the cloud or you've used CursorCloud or anything like that, I much prefer to just, like, fire it off, have it run-in that cloud, and then and come back with the the pull request.

When I'm working locally, I I don't find as much of a use case for, like, firing off two or three things at once unless I'm specifically, like, I'm gonna work on these two totally different features that, like, won't overlap each other in any way.

Maybe that's a bit of a skill issue, but I just have haven't found the need for that just yet. Yeah.

I same thing, man. I, like, I find myself being like, I I guess I I just I I it's not that I wanna manage things that that tightly, but, like, I don't wanna work on that much different stuff at once that is not touching each other. Like, to me, that feels like I've lost control over it. So I I I don't know. To me, I just I have not really gotten into work trees in a way. And if I'm working on one singular feature, I don't see the need for work trees. I'm just gonna use a regular old branch. So I'll tell you one one place I have used them, and that is different implementations

of features. So if you say, okay. I Sanity to explore the best UI for doing x, y, or z, you can, like, go down one route or you can take one model and say, how how would you implement this? But in some cases, if you've got tokens to burn, it may make sense to, like, have three different models, approach it three different ways, and then all of a sudden, you have nine different approaches to a specific implementation.

Because so often, you you implement something, and you go, that's not at all what I wanted.

And you have to sort of, like, wind things back. And due to the cost of tokens, I don't think a lot of people are doing that, but it is an interesting use case. If if there was no cost, could you run three or four implementations and then choose which one is you you actually wanna go forward with? And in that case, it makes a lot of sense to have three or four work trees running at once, and you can you can fire them all up at once.

Well, Wes, let me introduce you to DMUX, dmux.ai, the other, letter mux. Multiple manage multiple AI coding agents and isolated Git work trees, branch develop, and merge all in parallel.

And I do believe that this is doing a lot of that specifically because you can have different models attaching the same project on different work trees and stuff like that. So Yeah.

Yep.

That's good. I've I've mostly used this feature in in cursor. Yeah. What or, yeah, Syntax Mux. I'm in. No. There's already a smux. It is a smux?

Is a stream multiplexing library for Golang with minimal memory usage. Okay. The next JS or NAS that we're doing on the live stream mux. Needs to be just a a blank mux.

Okay. So we so we found there is an AMux, an open source cloud code agent, BMux.

Oh, there's no BMux. JS such a cool word, though, so props to people using that. Props to the actual mux, the video guys.

Yeah. Shout out to mux. There is an x mux. My god. There's one of them.

Is there a w mux? Hold on. Let's see.

The terminal built for cloud code, WMux.

The window application built on electron, vertical tabs, notifications when agents need attention.

Everybody is building the same thing.

Everyone's building tools for agents.

Nobody is building actual products.

Everyone just is making tooling.

There's gotta be, Wes, there's gotta be you gotta start moving into double a's. So, like, double a mux, double x mux. What about

m because m m u x.

Mux.

Mux.

No.

Mux.

Do you want would you buy a domain name where if you search it, Google doesn't think you spelled the thing right?

Mm-mm. Mutt. Did you mean mutt? Did I? I get so annoyed with Google doing that sometimes.

It's like, obviously, I meant this the thing that I'm searching for. But yeah. So Indian Vercel.

Mutt.

I mean VMux. Neo VIM session handler for TMux.

I think that's the previous part of the call. Manager for AI assisted development.

Then they're all doing things slightly different. Right? One feature I we didn't even talk about is this, like, the Scratchpad feature is really cool in, solo term Wes you can have multiple agents running on something, and they could be done in different harnesses, but they all have access to a single scratch pad, which I think could just be a text file on your computer, with with some instructions. But it's kind of neat to give it a spot where it can store stuff and communicate with the other agents.

Yeah.

Okay. Sick. I'm compiling a list of all of the different muxes right now. So we talked a lot about hosting these on on your local machines or anything like that. Like, I host mine on a refurbished Mac mini with an m two processor that was fairly cheap, and it runs everything just fine. There's also services like we just talked to DAX, and DAX, shared with the exe.dev, which I had seen before. But, this is something if you're looking to host in the cloud.

You can host this stuff on the cloud too if you want to.

Now granted, I don't like, I don't I'm just, like, tired of paying for subscriptions for stuff. So for me, throwing something else on the cloud and managing another server that needs to be powerful enough to run this stuff, is not super interesting to me. But if you're hosting this on the cloud, that's another possibility. Maybe you don't have the ability to have an another machine

in your workspace or whatever. So the Internet connection is is another one as Wes. JS a lot of this stuff works so much better when your your speeds and your pings are fast,

and it it sometimes just makes more sense just to SSH into some rented machine that lives somewhere else. Yes. Sick. So let us know what you're using, all the tools, and all that stuff. Now is the part of the show where we're getting into the sick picks and shameless plugs, or just sick picks in that regard. So, Wes, do you have a sick pick? Anything that you've been feeling? Anything that you've been liking?

You picked this probably three or four years ago, and my wife just told me to get into them. And that is packing cubes.

We're about to call vacation. And we've got four kids, two adults. We've got all kinds of stuff, and it's just chaos trying to pack it all.

And simply just giving everybody a suitcase doesn't work because it's just it's it's too much room. We can't fit it all into the car. So now we give everybody several packing cubes. You put your socks and your underwear in one. You put your T shirts in another one, and it's just so nice to have everything. It it made no sense to me when you sick picked it. I was like, why why would you want, like, more smaller suitcases inside of a larger suitcase? But the organization, absolutely on key.

On key. On key.

On key is fine.

If you're a music head, yeah, it's on key.

I I will say that, like, yeah, I, I I the packing cubes were so essential to me, and I was gone for, like, two weeks and some days. I've never been gone for that long in my entire life. And, like, without packing cubes, I would have been absolutely host.

So shout out to packing cubes. I think they're very effective.

I am going to sick pick a foot pedal.

I got a foot pedal. I mentioned this in the last episode with Dax, but let me show you.

It's made by Elgato, and it has three pedals on it, which is nice. All the other foot pedals look like garbage, and they're also expensive. So, Elgato made this one, and it has the same software that the Stream Deck has on it so you can set up macros to your heart's content on this thing. So the way I have mine set up is that I push to hold to start dictation so I can talk. I have this one that is just enter, so so it's the enter key. And then this one hits a macro that I use for all kinds of things, but the thing I really like to use it for is cycling through my active agents in Herder.

So I'll just click this, click this, talk, hit enter. Click this, click this, talk, enter. And my Node.

It's so goofy.

It's so it's so absurd, but it's really pretty dope. I I'm like, I'm I'm absolutely shocked at how much I've grown to really like that workflow. So shout out to Elgato for making a nice hefty foot pedal that doesn't look like it was made in the nineteen seventies.

That's sick. Man, I I think we're thinking about inputs quite a bit, and I think we've got a video on this, coming up down down the pnpm. But funny inputs are just, like, nonstandard inputs. Everything is changing right now. You know? We're using everything called mux.

Your keyboard is maybe not the best way to input anymore.

Everything is absolutely changing, and it's kinda fun to experiment with these different inputs.

Yes. Alright. That's it for today. Thanks for tuning in. Peace.

Peace.