Hands-on Introduction to Swamp Club

0:08 From the first container to a thousand nodes, breaking down barriers, rewriting the code. From dock at the rancher, the journey's been wild, Making the complex feel simple in Hello, and welcome back to the Rawkode Academy. This is Rawkode Live, the show where we take a look at the latest and greatest tools in the cloud native space and beyond. I'm your host, Rawkode. And today, we're taking a look at Swamp Club. If you've not heard of that yet, stay tuned because this is gonna be interesting. Joining me today is Nick. Hey, man. How's it going? Hey. I mean, what an incredible intro that was. Like,

2:01 whoever says AI produces slop or whatever, just listen to that song. That was that was fucking awesome. So, yeah. Thanks thanks for having me. I appreciate it. It's amazing how much we can just magic up now from a prompt and be able to just get, you know, cool little tunes like that. I mean, they don't have to be Metallica grade tunes. They just have to be good tunes. And, yeah, it's a great way to open a show. I love it every single time. Absolutely. Yeah. That was great. Anyway, yeah, we were taking a look at Swamp Club today. And I'll be honest, I tried to

2:33 look at this before and I got confused and I reached out and I was like, hey, man. You gotta help me. So let's take one minute first just for you to say hello, tell people who you are, what you're up to, and then we'll cover what is Swamp Club. Yeah. Okay. So, yeah, I'm Nick, known on the Internet as keeb forever. You know, I run go to market at Swamp Club. And you know, I'm not surprised that you were a little confused about what Swamp does, because the aperture of what it does is so wide. It's hard to actually come up with

3:08 a message to, you know, that resonates with everyone kind of at all times. And, you know, what we've really distilled it down to is, you know, Swamp is the right way to build with AI. And what we mean by that is is, you know, the same work that you're already doing. If you install Swamp alongside it, then what it's gonna do is it's gonna make that better. And the way that it does that is it encodes the way that you interact with the systems that you use to do your work as behavior. It builds extensions for itself to go reach into those systems

3:45 so that if you want to get information from your Kubernetes cluster, it can go pull logs. If you want to go get tickets in something like Jira or Linear or whatever or Asana, it knows how to reach into the system and get that information. If you want to do product analytics, you teach it how to reach into your data lake and get that information out. You encode how you interact with those systems as behaviors, and then what you do is you teach the AI how to do those things. And so swamp is the framework for doing that, for building the system of that interaction. And

4:21 why it's the right way to do that is because what that means is you teach the AI to work with your systems as opposed to generic distillation of how you would work with a Kubernetes system. You teach it to work with your system in the way that you want it to. Therefore, you reduce the amount of tokens that you spend, and you speed up that interaction. And it requires less intelligence to do so. So you can use less intelligent, faster models order to drive the harness to make that happen. And so that's it in a nutshell. We're gonna, you know, bring that into practical purposes as a

4:55 part of, you know, the session that we're doing today. Nice. So automations, AI, very cool, but two questions set front and center for me. One, why is it called Swamp Club? And two, the first thing you see on the website is a leaderboard. Can you maybe share a bit more about that as well? Yeah. So Swamp is an anagram for the five founders of the company. So I'm Nick Stinemates. I'm the s. John Watson is the w. Adam Jacob, our CEO, is the a. Mihir Lupinacci, our COO, is the m. And Paul Stack, builder extraordinaire,

5:35 rounds us up with the p. And, you know, the reason it's swamp club is because that's, you know, the first rule of swamp club is to tell all your friends about the swamp club. It's the community experience. It's the last thing that's left. You know? There's a there's a belief in AI that software is dead. The thing that's alive is all of us building. And so whatever form that takes, that's what we're trying to encapsulate in the club. And so, yeah, what we've done is turn all of that into maximum maximum fun and maximum weird. So the leaderboard is, we

6:14 truly believe that we're building at the frontier of AI, really pushing the envelope on what can be done and how to think about it. And the leaderboard is a great encapsulation for who is doing that building, how they're communicating around that. So you can't be at the top of the leaderboard without talking about how you're building it. So there's blog posts related to that. You're publishing extensions, so ability to go in and build this behavior or build this ability to go pull information from systems. And then you have a profile that reflects all of your accomplishments

6:53 inside of the club so that you can stand out and gain credibility among your peers. And so it's a real social aspect to a technology that we believe is is transformative. Awesome. Alright. I'm looking at the website now, and I should pull it up so other people can see it as well. Right? But are these numbers legit and live? Like, are these hooked up with something? Those numbers are legit and growing every second. Yeah. Almost 200,000 builders and 355,000,000 automations of moments of greatness. Like, that is big numbers, man. Fucking hell. Yeah. Yeah. And we just started in February. And

7:34 so, you know, the the project went live. We announced it in Ghent at Config Management Camp in February and we're just completely floored, you know, by the uptick in the number of builders who are coming in and building and then the usage that correlates along with that. And obviously, all of the great customers and case studies that have come behind that, like real builders actually delivering real AI outcomes for their businesses and talking about it as a part of our community. It's a it's an incredible achievement, but we're just, you know, we're just getting started. Sweet. So

8:13 with regards to the use case then, if people are watching us and they're like, okay, it's AI automations that brings chaos to what can be quite a turbulent experience of, you know, slob generating code sometimes. Is the use case here for Swamp that it's coding software focused? Can I use it to power my or to automate my house, my home lab? Like, is there any bound really to what it can be done? And what weird things have you seen people doing with Swamp that you don't expect? Yeah. It's such a great question. And it's one of the challenges back to what I mentioned, you

8:45 know, when when you said it was a little confusing. It's like, when I first went to make art, our first iteration of the use case page for Swamp, There's a few ways to look at it, you're like, okay, well people are building software factories with Swamp, people are automating their home lab with Swamp, people are doing some ITSM or SRE work or some business analysis or whatever with Swamp. And so how do you define a tool that touches so many different parts of the industry is a real question. And like, the the most broad conclusion that

9:27 I've been able to draw is, like, if you're you know, Swamp is as as applicable to is is as applicable to anyone that is using AI to accomplish a task today. Like, it it just makes that task better if you build if you build for, you know, if you build with Swamp in mind. And so what we find is that, you know, we're we're AI maximalists ourselves, and we use Swamp for everything across our business. So from financial modeling to forecasting to customer CRM to, reconciliation of actuals between Stripe, which is how you pay for Swamp,

10:10 to what our fundamental business plan is, to projecting out in the future. We have Swamp workflows for all of that. But I think that that's the frontier of the frontier. How many finance people are actually using and driving a business model through something like Swamp, I think we might be one of the few that have taken it to that extreme. But it's crushing it in our use case. And our COO has now become a builder himself. He literally lives in the code and is building out financial models all with Swamp. So yeah. Long way to answer your question, but it's very broadly applicable across the whole show, which

10:59 makes messaging a real challenge. I I mean, I I love it. I think AI is such an enabler for everyone to become a builder no matter what they wanna build. Alright? There's no task too small or too big, it would seem. You can just dream it and I I can do it. So I I love that aspect about AI. Now, let's focus on this getting started experience. Let's break down the confusion that I've had. So I wanna I wanna swamp. Do I click start my quest? I need to log in. I've got the CLI in my terminal. Like, what is that onboarding experience? How do I do

11:32 my first thing? Yeah. We've got a few ways to get started. So the first one is there's like a little manual button in the in the top right next to the GitHub icon, like in the the top bar. Yeah. You can click on that and you can, you know there there's there's three things that you can do from here. Right? There's a button up top that says start with swamp start. There's build your first automation by hand, or if you're, you know, wanna understand more about how the system works before you jump in and start to build something, you know, you can do that. And so,

12:06 yeah. Click on one of those. That sounds interesting. And what it should do is take you to the page for how that will happen. Well, I mean, I guess this one is the one that I'm drawn to. Right? Because you've colored in a little bit. So I guess we should start with swamp start and, like, see what happens. Yeah. You got it. Alright. So I do have this installed. I normally don't install stuff in advance, but because I'm back on my next source machine, this was just safer. So we're gonna do a swamp start. And

12:40 I don't know what's going on. Alright. Here. Yeah. Yeah. Wonderful. So, I mean, I got a browser. Yep. It And wants me to name my works. Alright. Okay. Alright. Okay. This is an we onboard them for them. So let's call this Rawkode Academy. Yeah. And you want it in that directory that it created in? I I mean, can I I can change it? Like Yeah. You can I believe you can go back and start the workspace? Alright. Alright. Alright. Yeah. I mean, let's just I I I can see where it is. I don't mind. Okay. So Yep. Let's do that. And at least until

13:26 we understand what it does, and then I'll start being a bit more particular over where all this shit lives on the drive. But right now, I'm just like, show me show me the power. So after right now, give it a brief okay. Yeah. So we I just tell it about it. Yeah. Just say just tell it a task that you wanna automate. Ah, okay. Let's triage open neglected issues on academy GitHub. I'm gonna pick my agent. I'm gonna go with Codex. I tend to have a bit more success. Opus 5.5 is an amazing coding model, but Codex is still, like, my main

14:11 orchestrator, the one, like, I I can talk to. If that's I don't know if people talk to their agents, but, you know. And go. And now, I need to do this, which is just yeah, I don't know what that's doing. Codex. Oh, it's running Codex with that thing. Right. Okay. I can just ignore this now. Right? Yep. You can control c out of here and and yep. Alright. So c d into my workspace. It's told it to read the task, which is probably just that one sentence that I stuck into the web form, and it's just

14:52 gonna go do the thing. Yeah. You got it exactly right. I mean, is that is that it? Have I swamped? Yeah. You're you're now swamping. You got it exactly right. And so, you know, what this is gonna go do is it's gonna go find out, like, hey, are there are there extensions for you to, you know, pull in for GitHub. So it's gonna go out to the Swamp Club extension store, find any GitHub extensions that are relevant, like this one. You know, webframp has a GitHub extension. It's gonna say, hey. Do you wanna use that? Can I trust that? Yeah. You can go to

15:38 Swamp Club and look it up if you'd like, you know, and and you'll see that it has an a rating, and you can find, like, all of the behaviors that it has. You can ask it if you should trust it, and it can introspect it. Like, yeah, you should just operate as as you would, you know. Don't take my word for it. Do the investigation that you would normally do. Well, yeah. I mean, just out of curiosity and showing the fill picture. Right? And there's multiple GitHub. So we got this one. And you can see it's

16:12 got some stuff. It's got some downloads. So, yeah. I mean, can I just go and see the code? And this is gonna use the GH CLI for all of the work. So Alright. Cool. So my question then is, I I got this web interface, but really all it did was write a markdown file and then run Codex for me. Yep. So it it doesn't feel like I'm swamping, but, I mean, it's obviously got access to a whole bunch of, like, swamp skills that tell it what Swamp can do and how the extensions work. Like, how does this

16:50 thing work under the hood? It's a great question. But you are swamping, and the reason why you know you're swamping is because you can see, like, at the top of your skill, you know, it said, like, swamp behavior type search GitHub. Yeah. And so what what it's doing is is the AI agent is interacting with Swamp in order to fulfill what it what you've asked the task that you've asked to give it. Right? And so what it's looking for is the ability to reach into GitHub to find those issues that have been marked neglected or, you know, whatever, open and neglected issues. And what it's

17:28 finding is it doesn't have the agency to do that yet. Like, there's no extension. There's no ability for it to do that. And so it's saying, hey. I I can either build that for you. I can build that behavior for you directly, or you can use this existing one that exists in swamp club and pull that in. And so that's the decision in front of you. Asked me to confirm my repo, and then I'll say is web ramp. GitHub. Alright. It's kicking. Yeah. And so now you can see it's doing swamp extension pull. So it's gonna learn how to to pull that extension in. It then it's

18:21 gonna create to the behavior to go out and, you know, reach into your to your repo to go pull those issues. Alright. So we are installing the GitHub one here. Yep. So this is a Deno module. Yep. Now, it's doing this just some code that knows how to pull repos. Yeah. You got it. Yeah. It's So when I ask Swamp to do something. Mhmm. It's just writing TypeScript code. Right? Yeah. The type the the the TypeScript code is the deterministic layer to reach into these systems. And that's the way to think about that is in basically object oriented programming terms. What you have is some schema, which is called ZodSchema.

19:07 It defines the object model. And then there's there's methods that hang off the object. So you may have a an object that represents a GitHub repository, as an example. And then you have methods in there like fetch issues or whatever. And the entire point of this part of Swamp is to map those methods as tool calls into the LLM. Nice. Alright. Let's see what it comes up with then. Yeah. So right now, what it's doing is it's it's understanding that schema of what it is to have a, web web frame GitHub object so that it knows the values to populate so that it can

19:55 ultimately go out and pull these issues, like accomplish the task that you've given it. So are these behaviors that I see as a this subcommand on swamp? Are these the deterministic steps that we enable swamp to do? You got it. Yeah. Alright. So, I mean, I'm assuming I can just run this command myself. Right? Like Absolutely. Yeah. Oh, wrong one. Oh, it's not gonna go to the question. Oh, nice. Yeah. And this this is pretty interesting because, you know, part of the part of the savings that you ultimately get when you encode things in this way is you can run Swamp directly, any of these commands directly. It doesn't

20:47 have to use AI to drive the harness at all. And so you can stitch this into existing scripts. You can stitch it into CI. You can you know, call it over a WebSocket. You can do whatever you want. And you can add you know, it gets real meta when what you have is a whole bunch of determinism, and maybe what you want to do is, in the middle, inject some a l some AI summarization or, you know, some LLM goodness, some intelligence. You know? And then then what you have is basically the deterministic layer and then just in time intelligence where you want it to. And

21:25 the way to stitch all of that together is, one way to think about a swamp workflow, as an example. And so you basically get AI when you need it and how you need it. I know I could turn my permissions to auto mode, but I'm I'm actually enjoying seeing the commands. So maybe Absolutely. I'll switch that one in a minute so I don't have to keep sitting here being Yeah. Enter What's fun is okay. So so there's a there's a an important underpinning here. And you maybe you can run this Swamp data query commands also in the terminal. Right.

22:14 So swamp data query. I mean, just that command has questions on my head. Right? So we have this behavior that has the ability to go and fetch issues. Is Swamp maintaining state then? Like, when it pulls those issues in, they become locally available and queryable, and then that's what this is doing. It's like working with data that the behavior is exposed. Yeah. Managing state is the right is is the wrong frame. Capturing state is the right way to think about it. So, like, the issue may mutate right now. You know, you may go in and, like, click on a button, and obviously, that's not gonna get reflected

22:49 here. But a snapshot at the moment of time that you called get issues is is stored in in what we call colloquially the swamp. And that that has a lot of amazing properties that we'll talk about later once you understand the core concepts. But yeah. So so now what you've done is you've built a system. Okay? So so what's happened here at the bottom is now there's five open issues, that you've pulled using this behavior that you you just taught your LLM how to to bring in. And now what you have is the ability to ask your LLM at

23:31 any time inside of this project, hey, get me those issues. And what it knows when you say, hey, get me those issues is what it, you know Oh, we're offline. Hello? What it knows, I think for those that could still hear me, if you are still there, is that how to go and get those issues. It knows that what you mean by those issues, like go into Rawkode Academy, get the issues that have been abandoned or that are new, and then bring those back. It doesn't have to rediscover which repo, how to go get that information, waste tokens on, exploring the different issue types that

24:20 exist inside of there. You've already encoded exactly what you want to get, And so the LLM can be much more precise, and it doesn't have to waste tokens munging all of that data and then storing it in a way that's queryable after. Cool. Sorry. My camera decided it's it's finished with me. It's I see. It's turned off. Yeah. It was it was saying offline on my side. So I didn't on both, like, the screen and the yeah. Okay. Thank you, man. Alright. Let's can still see oh, yeah. Sorry. I've been you can hear me now. Right? I can hear

25:06 you just fine. And I can see the screen just fine too. Yeah. Yeah. Yeah. Yeah. Two things. No. I'm not gonna mess with that. People don't need to see my face. It's fine. Okay. Let's focus on the Swampin'. So I did miss a little bit there, but we have these I'm gonna call them projections of the the GitHub issues into something that Swamp is able to enumerate and search and query and other stuff. So Yeah. And and now, basically What do we do next time? Like, I have the Yeah. The important part to understand is now what you've taught your LLM is a behavior of how to reach into your

26:02 specific repo to get your particular issues in this particular shape. You've encoded that as a behavior, and so now the next time that you want to do it, the LLM has been taught how to do it. So it's gonna be faster, it's gonna use less tokens, You can use less intelligence in order to do that. So you can use a much faster model like Astra or whatever one of the faster ones is on the Codex side. I think we lost you again. And anyway, you're gonna get the same exact results every single time. That's a that's a great win, like not having to use frontier intelligence

26:38 for every task after building. I think I'm gonna on the side. I'm gonna I'm gonna build up a similar thing so that so that if we need to, we can continue the demo. On my cloud session. Yeah. Thanks, Matt. I can hear you again. Can't don't know what's going on with world is against me. It's in that swamp. And then my computer is not happy about it. Alright. We will attempt to continue. If this continues to be detrimental, we can always come back and look at this another day. But let's hope that that is the end of my

28:23 networking problems. Fingers crossed. Touch wood, all that stuff. Alright. Sorry about that, everyone. Okay. So sorry. What were what were you just saying? Do you mind repeating the last twenty seconds for me? Yeah. So what I was saying is that your what you've just done is teach the LLM how to reach into your repo to pull in your issues in that particular shape. And so, you know, back to the beginning of why Swamp, you know, building Swamp is the right way to build with AI. You know, because you can teach your LLM these behaviors, what that means is that you get a more you get the same output

29:09 every time that you ask with this with, you know, with a little bit of a variable input. So if you say, hey, go get my issues from my repo, it knows what my repo is Rawkode dash academy. And the issues that you want back are the, you know, the neglected issues because that's what you've taught the system to do. Yeah? And then it stores all of this data, you know, in the swamp. And so then you can say, you know, once you've done this a bunch of times, you can say you can you can ask questions like, hey, how how have my neglected issues changed over

29:44 time? As an example. And all of that data is available as memory to the LLM for it to go and, you know, get all of that information and build the right reports and do the right analysis for you, if that's if that's a direction you wanna kind of take the question. Okay. So assuming I'm happy with this behavior, it's doing something useful to me, and I want to run this once every morning at 4AM to get something prepared for me. Like, what does that kind of setup look like? Because so far, what we've done is, you know, kinda put in a

30:18 a a simple task and then copy and paste it into my terminal. But Yeah. Is that is that the standard operating mode, or is there something else? How about you just type that into Codex? Say, I wanna run this at 4AM every morning. How do I do that? Alright. Yeah. And do you want the output to look exactly the same? Where do you want it sent to? Do you want it as a report? Do you want it saved as a markdown file? Like, what do you how do you do you want it to show up?

30:55 I hadn't thought that far ahead, but, you know, maybe I want a little HTML one pager that I could just open every morning from my downloads folder or something like that. I mean, I don't can it email me? Can it can it like It can do whatever it can literally do whatever you want. So what it's telling you is it's gonna add it's gonna create a workflow around finding these these issues, and it's gonna add to that workflow a cron trigger that says, hey, every day at 4AM, run this. And so what that will start up is a persistent daemon. We call it a swamp serve

31:31 daemon, like a swamp server That is gonna wake up at 4AM. It's gonna trigger that to run, and then, you know, the workflow will run. And so the thing that you would add to that is the output format that you want. Like, you could say, hey, make me make me a swamp report as an HTML file that gets stored in location x, you know, and it will build that for you. Or send me an email or, you know, whatever you wanna do. Okay. Let's keep it simple for now, but we can start connecting more things. But I think

32:11 just a simple web page, put it into my downloads folder. That that's fine because I'm gonna ask it obviously to run it now. I wanna see what this gonna look like. Yeah. Of course. The fact that I'm just asking Swamp to do this stuff, it's like it be is it becoming like an agentic interface to my machine? I mean, I'm assuming that is that cron served by Swamp server, or is it literally putting a cron onto my operating system? It can do if you want it to be a systemd unit file or something like that and and be a cron managed that way, you can absolutely

32:42 do that. Or, you know, in this case, what it's suggesting is to run swamp serve with a trigger for your workflow that runs at 4AM every day. So it's your prerogative. That's that's an important part about how to think about Swamp is, you know, it should adapt to whatever workflow you want it to. And, you know, like and so it's it's up to you. And there we go. Yeah. So this is explaining to you how the process that you just that you just went through. Yeah. And so now just go back to the LLM and say, no. Actually, you know, I want

33:32 I want the issues as an HTML file every time this runs. Alright. It took me too literally there, I guess. I wasn't communicated very well with my robot. Yeah. Alright. So, obviously, we're at a point now where it's doing basic things for me. Mhmm. We've got all these extensions that we found here. So, like, the more things I want it to do for me, the more I'm gonna have to rely on these extensions or have it build an extension for me. So I'm curious to me, is there PostHog? There's not. Good. I store stuff in PostHog. Let's see if I can

34:17 have my swamp overlord magic that up for me. So Absolutely. I'm assuming I could just have multiple Codex's running and Yeah. Yeah. Swamping all over the place. Yeah. You got it. So Yeah. You just wanna have it running in that, yeah, in that same directory. You got it exactly right. Yeah. Alright. So let's integrate with PostHog. I want another daily report on the videos people watch from the website. So I'm assuming it's going to use the swamp behavior search extension search. It's not going to find anything, and it's magically just going to write me a behavior. But, you know, we'll

35:02 we'll see what happens. You've got it exactly right. That's exactly what's gonna happen. So now we're juggling two swamps. I mean, I'm assuming that's pretty normal. We're all juggling many agents now. I mean, how many are you normally rocking at any given time? Is that a It's it's a question. It's it's it's a I'm I'm I'm a little bit insane. By the end of the day, I usually have between 90 and a and a 120 Alacritty instances open. That's my terminal of choice. I'm I'm very much a spawn one agent per one task sort of person. And

35:43 so what you see on this very wide screen that I have is, you know, a 100 terminals open at any given time as I'm bifurcating my mental load, you know, across all of that and trying to keep up with the Swamp Club community and chat and all of that type of stuff. So, yeah, I think all of us at Swamp Club, we I think we're actively, you know, spinning probably eight to 10 agents all day, every day. And I just happen to be an extreme outlier in terms of the number of sessions that I load. I think it's like I think the last time

36:21 I looked at my usage stats, it was like 13,000 sessions in the last thirty days or something insane like that. But, yeah, I do most things that I do on a computer these days are all swamp workflows, as it turns out. So that helps. Alright. Well, it's detected that my Codex has a PostHog MCP, which is mildly annoying, and it's telling me that it can't reach the extension, which I suspect is my fault or at least my Internet's fault rather than something else. Yeah. You just tell it to build a swamp extension for PostHog. Right. I

37:10 guess sometimes you just gotta be quite bossy with the robots, don't you? Just do what I tell you. Then over here, this is just setting up that workflow with the actual report. So I'm trying to let that one finish too. It's it's it keeps trying to go into my okay. Do not use Codex entities. Even though it can't get to registry.npm. Is this just the sandbox? Is this the Codex sandbox being a PNAS? Let's we'll see. So you said you've got tons of swamps running all the time. Do you have, like, any you know, when you speak to people about Swamp Club and why they should be testing

38:06 it out, what's, like, your go to example that you use for people? Like, your workflow that you just think everyone else needs to be doing something like this? You know, typically, what I do is I turn that on its head because because of what's so simple, like, what did we do in five minutes? You had an issue that you you you had a a task in your mind that you wanted to go accomplish, which was, hey, go to my GitHub and pull, you know, neglected issues or whatever. Like, what's amazing is I don't have to have a canned demo that I go and I show

38:39 everyone that I get really good at practicing. What Swamp and AI allow us to do is pick a problem that matters to you and let's go solve that together. And so that's the best way to do it too because what leaves behind is something that is meaningful to you at the end of the day, not meaningful to me or that shows some, you know, some aspect of the product that I want to show you or whatever. You know? Like, you get to literally drive and I get to literally help. Does that make sense? It does make sense. Yeah. I'm just gonna keep trying to give this. It

39:13 seems like it's about me and extensions. I'm gonna give it a 1Password secret reference in hopes that it doesn't print it on the screen at any point. No. No. No. It's I it absolutely should not. The the word for this is a vault inside of Swamp. And so you can say, you know, use the 1Password vault implementation. That itself is also an extension. And so Swamp is built to be extensible at every layer, and you can see that that's what's happening here. It's creating a vault to store your encryption into or sorry, your your credentials into. Okay.

39:47 So, I mean, there's obviously some similarities here between Swamp Club and things like Hermes and OpenClaw, etcetera. Like, do you see them all as the same kind of automation frameworks? Or do you see them as completely different use cases? But people still have OpenClaw and Swamp, or is Swamp would it be enough to just do everything that they wanna do? Yeah. A ton of people are using, you know, the Swamp under Hermes, basically, in the same way that you're using Swamp under under Codex here. Swamp is agent agnostic. Like, how how you want the agency to happen matters less.

40:28 Sorry. The the input into the agency matters less than than us taking over, you know, to to drive all of the value that you're getting from this thing. Yeah. I think I think instead of doing this, David, what I would do is I would just tell it to use the 1Password extension. Yeah. Yeah. Yeah. Yeah. Oh, so it was trying to actually create a local vault and then put the password in. Copy your that's gonna copy your the that script is gonna copy the value from 1Password into a local encrypted vault. But you wouldn't wanna do that. All you'd wanna do is just use the extension to,

41:13 you know, go into 1Password directly. That way, there's only one place to update instead of two. Yeah. That makes sense. Okay. Yeah. I didn't even notice it was doing that, but you're right. When you look at the actual command, you can see that it it did not. So it should have definitely caught that. So, what's There's a person in the comments that said their Hermes is only allowed to use swamp. And if it's missing something, then I build it. I build it in and with Swamp. Alright. So, like yeah. So, yes, you set up your Swamp workspaces, and then your other agent

41:59 just understands those skills and how to work with them and then builds extensions and does all the pouting there for you. So, basically, your OpenClaw and your Hermes just become that kind of agentic chat interface, I guess. People are Yeah. You got it exactly right. Yeah. Alright. So the sandbox is blocking my swamp install. So this yeah. I should've used Claude. Sometimes the Codex sandbox is a huge pain in the ass. Alright. So, I don't know. Like, what is the one video that was watched most in the last Monday. That's just gonna be embarrassing if there's no views, but then

42:54 most people use most people use YouTube rather than the web site. So it's it's expected. Oh, it can't do it. Right. Okay. Oh, it's trying. No. Oh, it's a sandbox. Now, the extensions that we have put in place, if I drop out here and I say, fuck your Codex, we're going to Claude. Does it still have access to all that? Yeah. Yeah. So I can say, use our, PostHog Swamp extension to find the most watched video. That's seventy two hours. Alright. Swamp scale is loaded. Find behavior or searching for behaviors. Alright. Okay. Let's see if Claude can do a bit better. I think Codex is tripping us up here. What

43:50 do you use when you're swamping? I, you know, I use them all, but I mostly use Claude Code. I'm an Opus 5.5 kind of from a development standpoint, and then I use Haiku for the, you know, if I'm driving swamp, basically. And I love Haiku for that because it's a great router to the right behaviors to accomplish the task that I ultimately wanted to accomplish. Alright. Well, let's see if this manages to pull something to go. It's standard behavior, but there's no instance of it and that's because it needs a password and it's saying that I'm not logged in. What?

44:58 What am I doing with my life? I've broken 1Password. Oh, there we go. Alright. That was my fault. There we go. Try again. Much better. There we go. Yeah. It was just the checkbox to enable the 1Password CLI, which wasn't enabled. Well, it just goes to show that the human is always a problem, not the model. Yeah. And what's what's great about just, you know, to talk a little bit about Swamp and the framework here. So, vaults are themselves extensions. So, you can reach into things like 1Password if that's where your information is stored. What

45:56 what that means is that the raw values of the secrets are never stored with the LLM. It's not stored in the configuration of how this behavior is actually modeled. What it is is a is a reference to a value that exists inside of 1Password that gets injected just in time when the call is actually made. And so, you know, there's there's no ability to kind of leak sensitive information as a part of your as a part of your session. And, you know, as this scales into teams, what that allows you to also do is, you know, give people permission to run workflows, not necessarily

46:36 have access to grants. Or sorry, not necessarily have access to vaults with the credentials inside of them, but still able to do work using those keys that, you know, exist there as an example. That's a common enterprise pattern that we have kind of in the wild. Nice. Well, it looks like I found some sort of weird schema thing in PostHog, I identified it and then patched it. So it's getting there. Yeah. That's the one thing that I hate about 1Password is those dialogue boxes that show up Yeah. As you're doing these things. It's like, oh, look at that. Two videos are

47:22 tied. Congratulations. Yeah. Probably with two views each, but we'll see. I'm curious which ones though. But I never gave it any information about the PostHog project, what the metrics look like. It's it's nice to just kinda dive in there and find it. Yeah. You got it. And look, there they Yeah. Two plays. There we go. Alright. I don't like CUIDs. They mean nothing. Always use the GraphQL API to resolve video titles. Let's see if I can confuse it. Yeah. This is gonna do this is gonna do exactly what you want it to do, by the way. Then this becomes a behavior that it it does that just create behaviors on

48:32 the fly depending on the things that I asked? Yeah. You you got it exactly right. Yeah. And what's with this memory? What like, how did it determine oh, look at that. It fucking worked already. Yeah. Yeah. And so I hope what you're starting to see, right, is you're teaching the LLM how to reach into these systems. And, you know, now the next time you load this project and you have this question, it's gonna come back to you that fast. Like, you don't have to build anything. Don't have to teach it anything. You don't have to give it

49:06 any information about how to access these systems. All of that's been stored into the model. All of the last time like, all of the previous queries that you've done are all there in memory, and so you can ask questions about what happened last week. You can get information about what's happening now. You can decide what you wanna do with the data like, hey, I don't like CUIDs. Give me the actual titles. It'll go and resolve that for you using GraphQL. Yeah. Like, the world's your oyster. And so, yeah, now you're swamping. You're like, you're a bonafide swamp user now. I am I am starting to see the

49:39 power here. Like, the fact that it's able to say, hey, I can put together a small behavior for this, like, the small piece of functionality that I'm expecting it to know how to do. And Yeah. I'm assuming I mean, these are gonna compound over time. The more it learns about the way that I wanna work and the way that I want things delivered to me, it's always gonna get better. That to me is almost magic at a certain point. Yes. Yes. Yes. You fucking got it. You fucking got it. That's exactly it, man. You got it. One of the things I do after every livestream because I

50:07 stream to YouTube, as in I have to go and download the video from YouTube and get the thumbnail and re encode it and stick it into Cloudflare r two and add it to the web And now I'm like, hold on a minute. This is this is a swamp behave this is a collection of swamp behaviors and a workflow. You have got it exactly right. And that is when I say I have you know, when I code everything that I do as a swamp workflow, it's exactly that type of work. You know? And also, just bringing it back to the beginning about why you were confused in the

50:39 messaging and all of that. Like, how do you put that workflow on a website so that it resonates with you? I have no fucking clue. If you have an idea, let me know. You know? But, like, that's the power of swamp. You know? What you just discovered there and how you thought about, you know, bringing that into your daily work, that's exactly it. Alright. One more thing for me, and I want you to check again because we're Yeah. Running out of time. But this is all being very reactive right now. I tell it to run at a certain time. I ask it to do something. If I

51:12 want GitHub webhooks to come to my swamp, so whenever there's a new issue, I want Swamp to run a workflow for me. Like, is that something that's possible? Absolutely. Yeah. Tell it to do that. It will do it. You're you're just gonna say it's as swamp, aren't you? Like Yeah. Yeah. Of course. So what like, the way that that works, you know, just to give a bit more detail under the hood, what will happen is there's a swamp web server that accepts that accepts webhooks. You give on the GitHub side whatever the the webhook caller is. You get a you give it

51:46 a token that gives it the ability to, you know, make that call in. And so that gets, you know, the authorization and authentication layer done. And then what it does ultimately is call a workflow or call a behavior for you to go and execute. And so, yeah, if you if you tell it to build up build a webhook to do that, then what it will do is set up a swamp serve with the webhook configured in the appropriate way so that when you trigger it from GitHub, it kicks off a workflow automatically. It just works. Awesome. Alright.

52:24 That's definitely my next task. But Paul is in the comments saying that I'm now a fan keeper. So Yeah. Let's focus on that leaderboard, the social aspect of it. Like, all I've been doing is asking the agent to do some simple tasks for me. Yeah. How is this looking back to, like, statuses and points and, like, mean, that Yeah. So go back to go back to Swamp Club, you know, and, yeah, you can log in here. I'll in because I wanna I wanna see my new my new bling. Perfect. Look at you. So you're now a fen creeper. You know, what you have here

52:59 is, you know, a profile that you can start to be proud of. You know, you're you've reached tier six out of 20. You have 22,000 points. If you wanna know how those points got accumulated, you can go to your combat records, there right below activity and click on that. And so, you know, you're on a two day streak. Like, all of the commands that you're running are giving you points. You see you have, like, a history. You have a GitHub style, you know, view. You can see you signed up, like, five weeks ago or something like that in the heat map down below. You can see all

53:34 the swamp commands that you ran today. You know, you can see your trajectory. You basically were idle forever, and then today you popped off. And you can see the the breakdown of your points. And so, you know, this is this is just as much for you as it is for everyone else. You know, what what's amazing about what happens when you transact in points, on the leaderboard and see yourself is, you know, people inside of the community are interested in who you are, what it is that you're building, how it is that you're building it, how active you are. You know, they're talking about you in Discord

54:09 right now, because, you know, it's like the first rule of Swamp Club is to tell all your friends about Swamp Club and, like, to join in the fun and talk about what you're building. And so, you know, points accumulate by activity. They also, accumulate by you submitting, writing. So writing a blog post or posting a video or whatever about your usage with Swamp. You know, if you click on the lab there in the upper right, what we built into Swamp this is right next to account in the upper right. What we built into Swamp is if you ever find a bug in the

54:48 process of using it, Swamp itself knows how to file bugs in the Swamp Club lab. This is where we pull all of our work from and how Swamp builds Swamp. So you could think of this kind of like as a GitHub issue list if you wanna draw an analogy and a full, you know, a full life cycle of taking an issue from files from a user all the way to, you know, delivery back to that user. And so, you know, from a metric standpoint in terms of our velocity, if you file an issue, if you find an an

55:26 issue with Swamp right now, chances are we'll triage it within three hours. And then after triaging it, we'll get a fix out to you within an hour. So you find a bug today, you get a bug fix same day on any issue that you file in the product. Uh-huh. Yeah. You can see that if you click on the insights button there to the right of the drop down that you just, clicked. So, you know, in the last seven days, we had 211 issues open. We shipped a 169 of them. Time to triage was one point four hours, and triage

56:03 to pick was one point four hours. Very cool. I like this. Alright. Last question for me then. There's a pricing page. Yeah. What's the business model? Yeah. So if you're just an individual user building shit in your home lab or whatever, it's free. You just go to the billing page, claim a personal license. We have a large community of people who, you know, have decided that that's the route for them. For everyone else, you pay. And the idea is we make it a nominal payment, but what you're doing is attaching to the product, you're attaching to the community, you're attaching to the

56:45 club. And so, yeah, most people start off on a team, and what we build is proficiency building at the frontier, and what that grows into is enterprise deals. And so, you know, from a customer mix standpoint, we have customers at all at all tiers of the business or all tiers of the pricing that you see right here across basically every industry and technology today. Nice. Alright. This has been awesome. I I really I think, you know, when this stream finishes, I need to get onto my website. And I think what I'm gonna do is put together

57:24 those swamp behaviors and see if I can just automate what is typically the worst part of my day. Because I just hate the whole process of doing it. Yeah. And I I should be automated. But it's one of those things that's difficult to automate with code because I don't know. It's just it it feels like it's such a manual task, but with Swamp, it kinda removes that feeling. Like, I feel like I could just give it out, like, just describe what I wanted to do, and I don't have to care anymore. Like, the code is not important as long as it achieves what I want it to

57:52 achieve. And that is pretty cool. And I guarantee you it will work. Like, the workflow that you just described to me, I guarantee you that it will work because I've seen hundreds of workflows that work very similarly to that. And so, yeah, I know it will work. And then you're gonna tell me, you know, in twenty minutes after you built it that it absolutely works for you. Nice. And the fact that you can take the worst part of your day, but I'm just gonna I'm gonna tie the two together. The fact that you can take the worst part of your day automated in

58:27 twenty minutes and have it be repeatable with Swamp, you know, that's why it's $9 a month for you to pay for it. It's like, how do you not how do you not draw that solving the worst part of your day for a $108 a year or whatever, you know? Anyway, that's the pitch around the business model. I mean, it works for me. Like, no questions asked. I don't know. Thanks, Paul's in the comments saying that we have a Helm chart and we can swamp in my Kubernetes cluster. I mean, I think that's a whole another session where I may have to grab you grab

59:01 you back on and say, let's do something a bit more extravagant. But I think for today, just seeing you know, the the really standout things for me where the behaviors are, like, these little not little isolated tasks. I get that. The fact that there's then the data layer and the ability to query that data and expose that to more behaviors is it just makes so much sense when you think about it. It's like, yeah, of course, it should work this way. And then the whole concept of the workflows and the webhooks and stuff like that. There's just so much here that I can go and spend, hopefully,

59:33 a few hours and take all of these really worst parts of my days, all those annoying manual toil and just and plain English, say, Swamp, help me. Like, I'm drowning. Like, there's a thing and that to me is awesome. Alright. Before we wrap up, is there would you like to share any of your workflows with the audience, or will we save that for another session? Let's save it for another session. You know, a lot of my workflows are related to building and running Swamp, the company. And so, you know, they're probably not as interesting as things that you're building. So but, yeah, let's save it for

1:00:07 another day. Alright. Awesome. Well, I'll tell you what, I will keep swamping. Actually, I've got the bug there. I'm gonna share that with people and just talk about it and of course with you. I'll be in that just for I am in this. I need to start. Thank I you so much for your time. You've gotten really quiet for me. No. You're back. I I heard, oh, no. So I was just saying thank you so much for your time and for helping me get over the the hard bit of the swamp and seeing how I could utilize this every day to Yeah. Make my life You're welcome

1:00:53 and congratulations and welcome to the club, you know? Like, what what we just did was pretty incredible in such a short amount of time. And so, yeah, good job. Alright. Well, thank you for watching. Thank you for joining me. Until next time, I'll see you all then. Have a great evening. We'll see you next time. There's more to do.