CD192: ROUTSTR - NOSTR, AI, AND BITCOIN

Happy Bitcoin Wednesday, freaks. It's your host, Odell, here for another Citadel Dispatch. The show focused on actual Bitcoin and freedom tech discussion. It's a very happy Bitcoin Wednesday. Seems like, I'm gonna say the quiet part out loud. Hopefully not jinx it. It seems like the bottom is in in a big way. Bitcoin is currently at $73,000. That's 1,368 sats per dollar. Current block height is nine three nine two eight three. In Fiat land, that means today is Wednesday, March 4, and

we're at sixteen hundred UTC. You guys will probably be listening to this in a few hours. I like to upload them right away. I thought maybe I'd start including the price of Bitcoin and gold and silver because I think we're gonna make a generational run against them. So it's just a fun thing to track. Current price of Bitcoin in gold is 13.27 ounces, and one Bitcoin can also buy you 819

ounces of silver. We're currently outperforming them both on the one day and one week while they're crushing us on the one month and one year. So we'll keep an eye on that. Freaks as always, dispatch is audience funded. We have no ads or sponsors. Just me and you guys. So thank you for continuing to support the show. Our highest zaps of last episode. Was, we got a 21,000 sat zap from Andy said, I didn't know what a lot of the words meant, but I love the intro.

And then Rider Die Freak map twenty one sent 10,000 sats, and he said, great rip. I know I said that the next show was gonna be Bitcoin, but Leah from Vexel was sick. So we rescheduled her show. It'll be in the future. It should be in the next week or two. I have, like, four weeks of shows planned for you guys. But we'll be talking about AI again. We're gonna be talking about AI a lot, I think. It's an exciting rabbit hole. Anyway, freaks.

Easiest way to support the show is sharing it with friends and family. Civil Dispatch is available on every major podcast platform by just searching Silo Dispatch. All relevant links are at silodispatch.com. Thank you for your support, whether that's sharing it or supporting with Bitcoin.

Today, we will be talking about a project called Routester, which is this beautiful intersection with Nostril Bitcoin and AI. And I really do think it's all kind of compounding together. And this project is a perfect example of it. And we have a party rep, which we haven't had in a while with dispatch. We have four, the four main members of the routester team here.

I'm gonna have them go around the horn and introduce themselves. We'll start with red. How's it going? Red shift, I think is your name. How's it going, Red? Yes. Yeah. I'm Red Shift. Hey. I I mean, thanks for having us here. First of all, glad to be here.

I'm Red. I've been working on Lowser for almost, like, nine months now, one of the founders. And I've been a developer my whole life. I think, like, AI has just made it so much more fun. Now we can just, like, build some projects on the go, you know. So that's what I've been doing. That's how I came across Routesda.

I I as a developer, I wanted to also as a developer, I understood on them. I wanted a solution like Routes to that's why I just wanted to try and build it myself instead of hoping hoping that someone would build it. Yeah. That's a brief intro. Obviously, I can't talk about about my past because I'm more suited on them. So yeah. What's I guess, let's introduce everyone. We have Abdul here. How's it going, sir? Hey. Thanks for having us first.

I'm Abdul. I'm also a software developer since I don't know what account. Also started working in Nostril and Bitcoin couple of years ago. Also had the fun to start the $4.00 2 LLM thing. I saw your paper request and get your refund a couple of years ago for a friend. It was really funny. And then after I listed, I joined the roster team to collect ideas and work on the Freedom Tash of What did you say did you say LNVPS? Is that what you said? No. No. No. It's not me.

With what project did you say? I missed it then. It's called the OTRTA. It's like one to rule them all. It's basically you pay you send the link cashew token with the LLM request, and then you get refunds Gotcha. Back from the providers. Gotcha. Gotcha. Thank you, Abdul. We have Evan here. How's it going, Evan? Hey. Thanks for having us here. So I've been building with AI for the past few years. And last year, I started working at Rouser, I think, since March 2025.

So it's been exactly a year now. So yeah. Love it. Thanks, Evan. And we got, last but not least, we have Shroom Inc here. Shroom in it? Shroom in it? How do I pronounce that? Shroom in it. Yeah. Yeah. Shroom in First, I wanna I wanna say support the show and leave a big fat zap. I I copied color here.

But, yeah, I I was working a lot with, like, AI agents and doing, like, different open source projects, like, building, like with, like, LangChain, like, these, like, agent orchestration frameworks. I was always, like, really interested in Bitcoin. So yeah. Then I I started working on Rasa with Evan and RedJoint and OptoJoint, and, yeah, we've we've been pushing this now and and, yeah, really, really excited times.

Love it. Before we jump in, I just also realized there was a fat sap that I missed from Phaedrus, 21,000 sats. He said incredibly inspiring. Love this narrative AI empowering the plebs. In the end, it can go either way. So inspiring more of us to work towards this path is so important. Thank you, guys. And, mean, I think that was incredibly well said setup for this conversation today. So let's get started. What is Routester? Why should people care?

Yes. So Routester's basically, like, if you think about, like, the whole Nostre ecosystem, like, Nostre is just like node based database system in a really abstract way, but just like a way to build applications on top of Nostr to make decentralized applications. And Routes is kind of the layer to build AI applications if you wanna create decentralized AI applications. So it's kind of like the same philosophy of Doster, but, like, going like, trying to, like, decentralize AI in a way.

But I thought maybe we can have different explanations for from everyone. Yeah. So I mean, like, also the diff like, what it is is also decentralized, you know, the definitions.

Yeah. I mean, like, I see it also as this one one gateway or one protocol through which you can access any model out there, and you can just pay with pay with Bitcoin anonymously as well. So I think that's actually the best part. It's when the first of all decentralized information list, anybody without without an email can just use any LLM out there by just paying with paying with Bitcoin and Cashew tokens. And it's built on Nostril because

someone there's there should be a way for us to discover, okay, who which are the nodes that are available in the market right now and which are the models that are available in the market right now. So Nostril helps us with the discovery part where every node announces an event saying that, hey, I'm available right now and these are the nodes that I offer at this price. Let's say thousand stats per million tokens. And as users, you just query Nostril relays and find the best router provider

that you can find for your requirements. Just like start using it. And the best part is obviously there are 15 active router providers right now. And if one of them goes goes down, you can always switch to the next one and the next one. Yeah. It's pretty wild. Yeah. I mean, so one of the things we've been focusing on the show is you have this dichotomy with

AI tools. Right? You have these large big tech models and people using them in, you know, fully cucked ways. They're fully KYC. They're tracking everything you do. They're controlled and permissioned. And then you have these open source AI models, people self hosting them, people trying to use them in more freedom oriented ways. And then you have like this middle ground where you have things like PPQ and OpenRouter and Venice where people are basically paying per token and using proxy services.

And there's all these different varying ways that people are using AI and how they are and how freedom oriented they are and how privacy oriented they are. And you guys are in that middle range

somewhere there. The proxy services are also pretty hard to use privately besides routester. So with routester, what's kind of fascinating to me is you have these AI providers. And then these providers, like, it's not like it's Claude providing it directly. It's like a routester note. I'm reading them like non KYC AI is one of the providers. The Fox is another provider. And then they have they're also providing models.

How is that working behind the scene? Because, I mean, I'm looking at like, you're able to hit the awesome proprietor, like the open source models kind of make sense to me. It's like, okay, like, anyone can really I mean, it's expensive, but anyone can technically host these open source models. But like for instance, the Cypherpunk provider is offering GPT 5.3 Codex, which is a very proprietary walled garden open AI model.

Like, how are they providing that? Are they just sitting in the middle and being a prox they're being a proxy to OpenAI. Right? Exactly. Yeah. I would say it's They're just being a proxy. Alright. Yeah. You wanna go? Yeah. I'll go. Yeah. They're just being a proxy. Like, you know, they are paying for their I mean, paying for inference using their credit cards in USD. So their KYC was OpenAI. Right? And then they're providing it at a higher cost to users of Routester?

Yes. They would they would have a markup. Yeah. They would have a markup, and they're just reselling access to it. And that that's the only way we can access this state of the art models. Right? Because, like, we still need these models. Like, these are obviously the best models. And Right. Cyberpunks, Bitcoiners need the best models to build the next tech I mean, build masterfully.

So if one of the nodes go down, there can be, like, 10 more nodes that can just, like, offer the same models from the same proprietary providers. I just thought you can maybe also think about, like, the master analogy that you have, like, random people running relays and they just relay messages. In Rust, they relay AI messages, but AI is, like, in a way centralized that, like, you need big data centers. So the only source of, like,

intelligent AI messages is kind of like these big data centers. So just random people relay it and resell AI messages kind of. But also if you have an data center at home, you can relay your own messages. So you can basically resell your own AI if you if you wanna do that. Or maybe you have like some strands who has a data center so you could resell his AI.

That's fascinating. And so, basically, what the result is is you're using Noster for identity and comms and reputation. You're using Bitcoin for payments. Obviously, the AI models are hosted in a variety of different ways. And there's effectively like a free market for these models, right? Like there's users can choose which provider they want, they can switch on demand.

The providers need to be competitive with each other. Otherwise, people won't use them. They'll use the cheaper version or whatever. Right? Exactly. And also like Nostra helps with coordinating among the users to figure out, okay, if there's actually a provider that like, you know, was down or they overcharged,

like, than what they announced. Right? So master helps us discover all the all of these problems and, like, make sure you're using the best AI provider that is in the price, quality, everything. Because some providers can just, like, sorry. Go on. Continue.

So there's actually one problem with AI. I mean, LLMs that we cannot easily measure the quality of LLMs. Like, for example, any open source model, let's say, like, Minimax m 2.5, which is the most popular one right now. Like, anyone could just run the same model in a quantized

way. They can just quantize it and just run it at one I mean, half the price or, like, you know, one fourth the price. And It's like compression. Right? It's like how something think about it? Yeah. It's like video compression. Right? Seven twenty p to two forty p. Right? So, of course, if you're if you're compressing it, you get a lower quality LLM, but you it's still not bad, but still, they announce it as though, you know, they offer it as though they're offering the full model. Right?

So that that's actually something we are also solving with Nostril where people can say, okay. This model is this model from this provider is quantized and outside is of low quality lower quality. So Nostril helps with that as well. Yeah. So there's basically reviews. Right? Yes. How does that work? I don't, like, see reviews on the site. So we don't have you can go to into a model, then you can leave a review inside that model. Got it. Just the bottom.

There there is a Nostre event for it, but it's not yet fully implemented. Log in Yeah. With Nostre to Oh, yeah. Exactly. Not just that. Like, we'll also we also have a way where, you know, if there's actually, like, the clients, they check how how much a provider is charging, and they automatically emit emit events as soon as, you know, a provider overcharges. So our chat app does this. There's also an SDK that I'm working on that also does this.

But, yeah, it's not fully So the the system is not yet perfect.

Or you're this is kind of the ideal vision where you have this, like, fully agnostic ecosystem where you have this social graph and you have rankings, and you can downvote one provider, you can upvote the other one. People block the one provider or, like, it automatically switches to the cheaper one, or the one goes down and you switch to the next one, like, kind of like this full algorithmic ecosystem where everything based on Nosta as a source of truth.

That's awesome. So I'm I'm also, like, so are you guys as route Routester as a Routester project? Or do you guys all are you guys also taking a cut of the usage? Like, I'm paying if I'm using Routester, I'm paying the provider. Am I also are you getting a fee out of that as well? We only lost money so far because cashiers are complicated to work with. But long term, is the goal to take a fee as well? Yes. So every provider can take a fee. And even if not a

Hey. Go on. Sorry. Yeah. Go on. Sorry. Sorry. Yeah. No. Every provider can add a fee to whatever whatever they're offering. Of course. But we also have a way for them to add a fee for us, like, us a percentage of the the cut to us in the code. But, obviously, since it's open source, they can ask their agent agent to just, like, remove this part. So it's an optional fee. Yeah. Yes. I mean, because I I don't think Yeah. Go on. It helps, though. Are we funding you through OpenSets?

No. Not yet. Okay. Have you applied? We applied them multiple times, actually. And we turned you down? No. So right now, we I just got an email back, like, last week asking for a asking a few questions. Yeah. I think you're the review process right now. Right? Yes. Yes. Yeah. I think I reviewed your application. I I'm not supposed to talk about existing applications, but also I was just

mean, we have so many. I mean, people are like, I think we've gotten like 3,000 applications in the last year and a half, which is how we end up in a situation where I have people on and then I ask them if we're funding them or not. But the reason I the reason I bring up the my dream, and you can't do this with every open source project. But I think the Holy Grail for open source projects is when possible is some kind of sustainable

revenue stream. But like, yeah, you bootstrap with grants, but like the ideal situation is, I mean, anyone who's played with any of this AI tech knows that we're spending a ton of money on models like people I've spent. I spent so much money

trying to learn how these things work. And so when you have that much money going through the system from user to provider to LLM host, then even if you guys take a small fee and even if that's voluntary, there I think there's a situation where you guys actually have, like, a nice sustainable stats flow, like a sustainable stats revenue stream that you can fund the project and improve it over time. No.

No. Absolutely. Robo sets, I would say. That's like one example of they're doing it in in such a way. It's also like, that's our goal as well. You know, we wanna be self sustainable. We wanna be profitable as well, and we wanna pay ourselves. Right? You know? We don't wanna be reliant on grants. I think, like, Noster and open source ecosystem as a whole, like, Bitcoin ecosystem,

we should all obviously, you guys are doing great work. Right? But we should also look for ways to monetize ourselves so that there's actually more money coming into the ecosystem and we we can find more people and more projects. We gotta access it. Right? So that's definitely our main goal. We wanna sustain ourselves.

We also think about, like, building more projects on top of routes too. Like, the chat interface is gonna be something that we run, and, like, we we add a fee there. And, of course, if you're using the API, then it'll be an optional fee. But, like, the chat interface, something we we run, and we can charge a fee there. And we wanna build more agents and more applications on top to

be profitable too. Yeah. Yeah. I mean, look, I think I think, you know, the the mission of OpenSets is to accelerate the freedom tech movement because there's there's a significant lack of funding on things freedom related. But at the same time, like charity is just not sustainable. Like I have to. I'm a volunteer and and like our team, we're like outgoing out there, like trying to raise money constantly.

The sustainable path is actual open source projects that are able to monetize in some way ethically. So I'm hoping we see more and more of that. I think Bitcoin makes it more possible than ever. Mean, historically, like, imagine, like, trying to integrate, like, credit cards into an open source project ethically is very difficult. So that should be interesting. Okay. I want to talk about the latest the latest craze has been OpenClaw.

And you guys are really quick on releasing guides on how to easily set up OpenClaw with Routester. And I thought it was particularly cool because there was this also another Bitcoin project, LNVPS, which is briefly mentioned earlier, that allows you to buy a server anonymously with Bitcoin. And so, basically, the setup was in,

I guess, in one line in one script, you're able to provision a server, pay with Bitcoin, and then use routes, then install OpenClaw on it and then use Routester as the back end. How do you guys think about the OpenClaw movement? I mean, it's like one of the fastest growing open source projects of all time. Is that where we're seeing things do where you think things are going? Do you think people are gonna use more integrated stuff? How does that affect how you're building out routester?

What are your thoughts? No. I mean yeah. I mean, yeah, I I work on the as soon as I came across OpenCloud, okay, this this it was, like, clicked in my mind, you know, we just wanted to work on LNBPS and make sure it just becomes, like, you scan one lightning invoice, and it pays for both your host, the BPS,

and tops up your LNB balance. That's actually gotten us so much more volume too. Like, there is, I think, in the last one and a half months, we've been doing like so much more volume than we used to do before. We're we're almost at, zero. And the last one and a one and half months, we've gotten to, I don't know, 100 x, I think. Yeah. I I think OpenTor as a I I don't think it's a hype. I think this is actually here to stay.

We have been thinking about such systems for a while now. One thing that we were discussing two, three months back was that agents should just be able to, like, you know, find PRs on Nostril using, like, Engit and just fill bounties and sustain themselves.

That's something that we've discussed in the past even before OpenFlow. We used to call it like a task marketplace where humans or bots can just, spin up tasks and LLMs can just, fill bots fill sorry, reach the tasks, earn SaaS, and pay for themselves. And that's where the Sovereign stack comes in comes into play here, where LLMs can pay for their own post and pay for AI and be profitable themselves. Of course, it sounds

too good to be true, but you were also running a board. I mean, I came came across. You were running OpenClaw, and he was making money using LN Markets. It was pretty cool. Well, he ended up losing money. Now he's just staying humble and stacking stats, but he was up for a little bit. But it was fun though, it was like, yeah.

Oh, yeah. I I I think also if you think about if you look into the code or like what what made OpenClose special, like, it's kind of like integrating with all these like permission systems. So it's like integrating with with WhatsApp, with like Google, so it can access to your Google Workspace. It gets access to whatever, like, APIs and all these, like, different permission integrations. But if you think about the the whole Nosta ecosystem,

like, is permissionless. I think the optimal AI bot would be even, like, something that is, like, fully integrated with Nosta. So I think like also if you think about like what is the future of like similar types of agents, I think there will be like a bunch of like variations of OpenCloud.

We we already maybe saw like ZeroClaw or, like, spec down versions, but there might be also, like, spec up versions. But I think, like, really interesting will be these, like, really Nostra integrated versions where you integrate all these, like, different Nostra tools into, like, one really, really powerful agent.

Yeah. I mean, I I have literally and the freaks know I've been a massive Nostrad bull for a while. I've never been more bullish on Nostrad because of the AI stuff. And all you have to do is you play with one of these bots, try and hook them into Twitter. You will have a very bad time. And if you're gonna have a decent time, it's probably because you paid for their expensive permission API. Try and hook it into GitHub.

Your account will probably get banned and get was mentioned earlier. That's like the Noster GitHub alternative. There's a couple of them. If you if you ask your bot to just generate an NSEC, it just does it. You don't even have to explain what Noster is. It just it just knows what it's doing. And then all the pain points that we've talked about with Noster, whether it's webs of trust, or running relays, or key management, the bots handle that perfectly.

They they no longer become friction points. So it really does seem ideal. And all you have to do is play around with these things for an hour, two hours, and it becomes very obvious that all the most of the pain points are the closed systems, all these proprietary walled gardens. And so we open them up and provide frictionless alternatives powered by Nostrand Bitcoin, then the sky's the limit. I mean, like Not just that. Reporting

Stop. We we we bought it for some of the exact same time. Latency and okay. Wait. I I wanna say, like, the the first thing that was built in Austin was kind of the NAK CLI. And, like, the big achievement also from from OpenCLI was, like, working with CLIs. So I think like OpenCLI is actually like perfect to work with Nostril because it can just use NAK to like publish Nostril and do all the things on Nostril. Knack is basically a Nostril CLI client. It does everything. Yeah.

Also, I wanted to mention more about, like, you know, how easy it is for AIs to just, like, fetch events from the relays and go through them and, like, see what's actually something that they can contribute or they can build. Right? Let's say we we just discuss a problem with a pro project or protocol on Nostril. And you should just like create an issue on GitHub, and like say, hey, I can fill this in like 3,000 with 3,000 stats, or I can finish this in 3,000 stats and

pay for it. Because its cost is also gonna be denominated in stats, and it can say 3,000 exactly 3,000 stats, I'll do this. I think that'll be, like, a perfect best case scenario where agents can just, like, create and build protocols that are already being built on master and just, like, pay for themselves.

And, like, since we were discussing about open source funding, I think that's where, like, I see there's, like, an overlap here because, like, if there's actually an agent that is, like, so well set up with a harness, let's say, a factory droid or codex. Right? OpenSats, like,

I don't know. I I feel like OpenSats should just be able to, like, just dump, you know, just put 10,000, let's say, on a point $0.01 Bitcoin. Give it this to an agent, and let it let's see how how far this agent with the help of a few humans or master can go in building up a project or building a protocol. Maybe this also helps with open source development and accelerating Nostril development as well. Have you guys thought something like this at OpenSaaS? I mean, it's crazy. Look.

First of all, the way we've designed open sites is we have a nine person volunteer board for people that are not aware, and all funding decisions need to be approved by five of nine board members. So simple majority of board members, and that's designed to reduce corruption of any individual board member, right? So like, you basically have to corrupt five board members to corrupt the funding process. It's not perfect, but it's, you know, it's still a centralized org. But we try our best to

avoid the pitfalls of other orgs. But as a result, we move slower. Right? We we move we prioritize reducing the potential corruption, but it results in us moving slower. That said, Gigi is our fearless leader on operations. And

I worked very closely with GG on big picture stuff on where the org is moving. And me and him are neck deep in the AI stuff. And it's clearly going to have a massive impact on our operations and the open source movement as a whole. But there's so many open questions on, you know, the the simplest question, which was the first question we hit, which is, do we fund vibe coded projects? And I think the answer is obviously yes, because almost every project is gonna have some element of

AI development built into it. It's just unavoidable. Of course, it's going to, it's a superpower. It it a lot of focus has been on its ability to empower people that are not already devs to ship projects. What it really does is it supercharges existing devs significantly. Like, if you're already a effective dev and you use some of these tools, it's gonna 10 x your output, 50 x your output. So it's a balancing act. We use it internally.

We are obviously not excluding projects that are getting funded. But also, like, if we get an application that's just like pure AI slop, then it's like, okay. Like, if you're not gonna put an effort to submit the application, why should we put an effort into funding you? So there's all these different aspects there. But, yeah, I do think in the future, will we be funding agents? Probably. Will agents be funding open stats? Also, probably. Right? I mean, we accept

we're probably one of the few charities in the world that you can just stream stats to our lightning node. So Yeah. With no with no you give no personal information. Like, you can just send us lightning payments at any at any point. So will there be agents that are doing that? A 100%. Are their agents already doing it? Maybe. There's no way for us to tell.

Yeah. No. I mean, I think, like, I understand your perspective there. Of course, you guys are decentralized in terms of, like, you know, five out of nine. So maybe I think, like, the way way forward could also be crowdfunding since we we also want to move away from charity funding, we should also be crowdfunding. Like like, I think zapping is something that's super popular. Right? Like, the I mean, that's so common right now in Nastra. If as soon as someone comes with a comes up with a new idea,

we should just be able to crowdfund, crowdsapp, and just fund an agent. Let's see how it go how far it goes. Obviously, there has to be competition. Like, obviously, not any not every agent will be able to, like, do with the same quality. Let's say, a high 10x developer builds an 100 x agent with a very nice prompt or, like, very nice harness. Maybe it can do a lot better, and it can just pay pay and, like, earn SaaS and, like, the crowd fund pay for itself.

I see that as a future. I think, like, we should move in the direction as to sort of, like, you know, we we build the public goods in this way.

No. Yeah. I mean, my dream is look, I'm a Bitcoiner. Right? So I think Open Sets itself is a middleman. Right? We try our best to do it in ethical, transparent, efficient way, but we're a middleman. The ideal the ideal future is people funding open source projects, whether that's an agent or a human directly. And Zaps make that very

frictionless and also fun, right? Because you get the social signal. People like the social signal. People like supporting something like Routester and then being able to brag to their friends that they're one of the biggest supporters. And I'm hoping that we seem to be trending in that direction, and I'm hoping that's gonna be the case. One thing I will mention is before we launched OpenSats, I actually

was like, this is too much paperwork. I don't want to deal with like all the legal bullshit of like, creating a five zero three nonprofit organization having lawyers involved and accountants and all this shit. We should just fund things directly using Bitcoin. And we created a website called Bitcoin Dev List, where you can just fund devs directly.

And one issue that did pop up significantly, and we'll probably continue it in the future, at least in the short to medium term, is people want to support Freedom Tech, but there's a lot of friction in choosing what to support and how much to support. They just want to give up that responsibility. You know? They wanna they wanna say, okay. I'm gonna give $10,000 to

the overall movement, but I don't wanna pick which devs get it. And so I think there'll always be a place for something like OpenSats for those types of people, particularly the larger donations. Right? I mean, we got Bitwise ETF, and they're giving 10% of their profits to open source. They don't wanna pick, you know, they're running a publicly traded business. They don't I mean, I guess Bitwise isn't public yet, but their ETF is public. They're running a massive public ETF.

They don't wanna deal with the intricacies of deciding which project to support. They just wanna give to an organization that does the hard work of vetting projects. And so there'll probably be a balance in between. But I would encourage all the freaks that if you are enjoying an open source project, support it directly. I think one of the coolest aspects of Noster is when someone like Craig Raw releases a new Sparrow wallet release, it is top of trending. It is one of the most reposted

things on Noster. It is one of the most zapped things on Noster and people are really supporting it in a value to value way. And then we're seeing the early signs of that. And I think it's it's gonna compound beautifully. Like, I think people are really sleeping on that aspect of the acceleration of the open source movement because of Bitcoin and because of Noster combined. Yeah. No. Of course. Yeah. And, I mean, I think yeah. There should be a middle ground. Yeah. Yeah. I I recently Go on.

So have you come across this project called cadillacs.network Yes. For master? Yes. Right? It's a Cool project. I post right? Yeah. I posted a good pounty on Cadillacs to just fix a bug on browser. It's like, maybe one case ads. And I think that's when it kinda clicked clicked as well because, like, Sean

Sean I mean, someone someone I called on Rowser, he just, like, tagged this agent. Hey. Hey. Forget it. Hey, agent. Just can you fix this build this bounty and, fix this bug? And all that's all you had to do. Like, you know, you just went to GitHub, like, created a PR, did a PR to our project. Like, of course, he applied for the grant, but it was not good enough. So I didn't accept it, but a human finished fill.

But it's pretty interesting, though, that this thing can be built on Oster. And this this is an Arbiter. I think Cadillacs is is a shout out to Cadillacs. I think that'll be something that that's that'll be big. I feel like people will zap. There will be more boundaries. Asians will form with boundaries. But choosing yeah. Choosing is something that's gonna be hard because I also had to choose between two submissions.

And obviously, that's also more work because I have to review multiple people's submissions right now. So maybe agents can also, like, review in the future. Maybe Cadillacs figure something out. I I I don't know. Let's see how that goes. Yeah. I mean, that's where it is. Right? Like, a lot of the friction points that make these things seem kind of out there and untenable agents will will smooth it out, whether that's on both sides. Like you said, like, I can I I would totally see a world

in the not so distant future where the agent is actually reviewing the bounty submissions and deciding which one's the best? And then some people say like, oh, well, then why doesn't your agent just fix the project directly? And it's because we'll probably see agent specialization. There'll be an agent that maybe is lower cost, that's just reviewing bounties. And then there's an agent that is very specialized and and maybe more expensive in in certain ways to handle certain tasks. Right?

No. No. Absolutely. Yeah. So that's something that that we are thinking about a lot internally here as well because, like, all of these Chinese agents, Chinese models, they can do the exact same tasks, exact exact same things at a fraction of the cost. One tenth, one twentieth, you name it. Right? So, like Yeah. I mean, it's significant. It's a significant cost difference. Yeah.

It's what yeah. It's just one tenth is amazing. So, of course, like, you don't want your agent just to to always be running OPUS or 5.3 codecs. Obviously, it's way more expensive. I think that's where, like, this actually is still middle ground here where humans with smaller models that who I mean, that know, okay, a particular area, a particular domain that they are an expert in. I think that that will shine because it will

solve an issue of warranty, but a fraction of the cost. And I think yeah. Yeah. You're right. Yeah. We'll we'll see more and more innovation on the specialization. And I don't know when the this whole thing will plateau. Sorry, Evan. You wanted to go? Yeah. I think there'll there'll probably be, like, roster before, like, agents. So instead of, like, LLMs provided by providers, you can probably provide agents.

And you just so you have your own skills or whatever that you own, and then you just take the input and then just give people output for sets. Yeah. I mean, it's I think anyone who pretends they know where we're gonna be in five years is full of shit. Like, the crazy part of all of this to me, and it just blows my mind, is that the potential outcomes are just so wide. Like I have no idea.

I just feel like it's just a it's a constant rabbit hole with no bottom. I have no idea what comes next. There's so many different possibilities. Yeah. I know. So it should make got him. Yeah. Go. Go for it. Party rep. Oh, no. I I thought I I told you to go on. It's funny. Yeah. No. I mean, like, yeah, there's no way we anyone can predict what's gonna happen. You know, like, Anthropic guy, Dario said, we were gonna be out of jobs in, like, 2025.

Like, they had and, of course, 2027 is when next year is gonna be the year of AGI, and we're gonna go to UBI and all of these things. It's it's like I I feel like there's a lot of fear mongering. Like, to to be honest, like, we work in AI, all of us here, and, like, especially I mean, we talk to all of the AI developer users, developers. And I think I don't know why this this in Silicon Valley, they want to create this fear that it's gonna be dystopian. But I only see utopian

future because AI is gonna just like let us build Noster to the fullest. It's gonna let us build Freedom Tech to the fullest as well. And there's no way there's no way it's gonna be dystopian because we're all working hard, make sure it's it's gonna be utopian. So I don't think human people should be worried in any way.

I feel like sometimes we maybe need to think a bit dystopian because when you see, like, all these AI things human are building humans are building, like, for OpenClub, like, they basically rebuild, like, some social media thing for for AI bots. But then you, as a Bitcoiner, you think, fuck, there's there's Nosta. Like, why are these people, like, rebuilding something that already exists?

So I feel like maybe we we need AI agents to, like, just realize it themselves because they read the entire Internet so they know, okay, NOSTA exists. We should just use NOSTA instead of, like, rebuilding everything from scratch. I think we'll get there. I think there's needs to be patience. I don't know. First of all, I don't trust Dario for shit. The Anthropic CEO, I think I don't know. There's a bunch of red flags there. I will say that he

Anthropic clearly wants like heavy regulation in the space. They want their moat, which which is something similar we've seen in the Bitcoin land, like Coinbase was one of the first exchanges in Bitcoin. And they lobbied hard for the bit license and other regulation that made it harder for competitors to come in. And I think that's clearly a strategy. And I think that's why he does a lot of the dystopian stuff.

He he he reminds me a lot of SPF of FTX. He's like the SPF of AI. More and more people are reliant on Claude, which is their their LLMs, their LLM models. And I it's it's it's concerning. I it's concerning to a degree, and none of this is static. Right? I I think if we want the utopian future, we need to make sure that the freedom alternatives are powerful, easy to use, relatively accessible.

I think that's why routes are so cool to me. Right? And so it makes sense that you have a more utopian view because you're actually in the trenches building the utopian view because if no one does it, then we very much will end up in a dystopian era situation. And I I think regardless,

even in the bull case, even in the optimistic case, this tech is incredibly disruptive. So if you're not the type of person that's willing to get ahead of it, yeah, I think there's gonna be a lot of pain for people like people will lose their jobs, there are a bunch of jobs that probably will not be there in five years. And if you're just sitting there flat footed, you're gonna have a bad time now.

No. Absolutely. Yeah. Yeah. We are we are gonna be on our toes at all times. Like, we're we're moving moving very fast. People should learn how to use agents. Just set it up. Open cloud, open code, not code, any of those things. Right? But, of course, like, you know, in in Anthropy is, of course, dystopian kinda. They would do it's feel it feels like they want the dystopian future. I don't know why. Like, it's yeah. I I see the parallel is too. But that's the that's why the Rouser is decentralized. I would say, like, even right now,

it's better than OpenAuto or CloudCore, any of these things. Why? Because, like, I don't know if you saw where Cloud was down, I think, like, 10 times in the last week. Yeah. You can see you can go to status.cloud.ai. You can see cloud is down like two hours, three hours a day. Whereas if you were if you

were just using cloud code the whole time and you're paying for the max subscription, obviously, you're just stuck without an AI. Right? But that's why, like, using route still in the background is like the best case scenario because if cloud is down, switch to Minimax, switch to Kimi k two. Like, this can't be down because this it's not like there's one server, one data center that is running.

This open source models, there are like 10 that are competing for the price. Right? So if one is down, there's gonna be another one. There's gonna be another one. It's like Bitcoin mining. Right? So everyone's sites kinda. And That's awesome. This can't

yeah. There can't be a single system that goes down. And OpenAuto was also down twice in the last two weeks. So, yeah, that was always never down because we are 10 we have 10 nodes. And will there be more? Right? It's just because it's early days. Let a thousand flowers bloom. Yes. With a thousand hours. We we just saw, like, eight notes, the spin up that came came online in the last one month. And I think, like, out of eight, four

were just, like, copies, I guess, because they're they're down, but at least the four new ones are live. So, yeah, slowly and then suddenly, it goes. Yeah. And, I mean, the agents can just dynamically do this in the background. Right? If prices change, if there's a different model that maybe is better bang for buck for his particular use case, if there's things down, I mean, it is kind of crazy that at least in in normie AI land, everything's like, centralized around three big tech companies.

Those things go down, and everyone just stalls out. They there's they can't do anything. Yeah. No. Exactly. Like, that's why, you know, we want more solutions like LMVPS. And I'm also, like, talking to someone someone else who's building something new called solving hybrid compute. I don't know if you know Dimi. A shout out to him as well. I'm using his system too right now.

And, like, he's also gonna offer BPS by the day, you know. And if you if you're just trying this system out, you should just be able to like if you if you see one server go down, just switch to a new one. Switch to a new one. That's why like pay per request is another highly I mean, very important important feature of Routeser along with privacy and all the other other things.

Because you you never have credit with Routeser notes. You always get to get your money back. You're only paying for what you are using right now. Right? So the switching cost is like way way less than any any other system out there. And the chat interface we have, and also the SDK that we're gonna release in the next couple of weeks. Next next, I think, week. Like, it it automatically switches.

If a node just like charges you higher, switch to the next one. If a node is down, switch to the next one. Switch to the next one. You never keep any balance with any of the nodes. That's awesome. That's that's the thing. We need such systems. Like, it should be like VPS. That should be the same for relays. That should be the same for any applications you're you use. No subscriptions. No lock in effects. No modes. Fully decentralized. Commission less. And, yeah, built for you.

What do you what is your opinion on so, like, a new user is using these tools more and more, and they're thinking about spending a lot of money on hardware to try and self host some open source models. I mean, I I've been thinking about this myself. Obviously, a freedom perspective, being able to self host your own open source models locally

seems like the holy grail. But there's also a balancing act there where I feel way more empowered being able to hit a bunch of different hosted models on something like Routester

than spending a ton of money and basically maybe having one model or two models that are local. How do you guys think about that balance there's actually a a synergy in a way that, like, maybe you have, like, really specific questions where you wanna, like, ask them locally where you really wanna don't let them leave your house. Like, you don't wanna have them anywhere else than, like, in your house. So then you ask them, like, your own local models. Then because you're not using your machine

the entire day, you maybe use it for, two hours a day or maybe four hours a day. You can like basically sell all the other twenty hours on Rauter, so you can basically make better use of your hardware because Rauter allows you to like sell that compute to other people. And then maybe you can use these sets that you earned to buy some more expensive models.

Because you maybe wanna ask some like scientific questions or whatever like that where you don't feel like you need to keep them in your house, but you're just curious about things. You need more power. Yeah. But you earn the stats yourself by, like, selling AI, and you buy it out again. It's just like refilling battery in a way. That's so fucking cool. It's a new Bitcoin mining. Right? It's exactly like Bitcoin mining. Like, you just buy the GPUs, like, you buy the my mining rigs.

And I think more more and more of the miners will do this in the future where they can just, like, sell the compute or their their energy over by mining Bitcoin, earn Bitcoin, or they can sell compute and, like, LLMs and earn SATs throughout system like. Right? And as if the demand for LLM is low, switch to bitcoin mining, and you get you keep earning bitcoin. I mean, that'll be the future for sure. There's actually this guy, Leonardo asked for I forget his exact last name.

So he he's he's he's the guy who predicted AGI in 2027. He I've I've heard that he's buying Bitcoin, like, mining farms, and, like, he's converting them into LLM data centers. You know? I think it should be the other way around. I mean, I think they should be together kinda, like, where you do you have both. Right? You do you keep your mining rigs and you buy GPUs too. And, like, you

another thing that that people have to realize that, like, computers can be super cheap. Like, NVIDIA is not gonna be the as big as as it is right now because chips are gonna be cheaper. So as that happens, like, there will be the only thing that matters is energy and power. Right? So, like, you just you would have to do both. And as Bitcoin becomes more more popular, of course, as the price goes up, the people will take this for sure.

Yeah. I mean, I think it's a time frame thing. Right? It's all like, I think, first of all, I think AGI, this idea of, I don't know, like, conscious computers, the robot slaves finding consciousness and self replicating and stuff is probably overhyped significantly, but also the capability and the acceleration of AI stuff is also underhyped.

And the truth is, like, somewhere in the middle. I mean, we also live in a bubble. Most people aren't thinking about any of this shit yet. There's a very small group that are thinking of this. I think the chip stuff, I think it probably gets more expensive. Compute gets more expensive first. I think there's very real physical shortages that we're hitting. I think energy production, Bitcoiners have been on the forefront of saying we need more energy production.

That's still a limiting factor in a big way. Probably it all gets commoditized on a longer timeframe though. I think I really like the idea of people self hosting some things and then hitting distributed marketplaces for hosted things. For other things is probably a balancing act. I think most people that are going out there and buying really expensive hardware right now will probably regret it unless they know what they're doing.

Because it is so early and because the variables are so high on on what the different outcomes are and everything's moving so fast. But it's just wild times. It's just it's it's it's really cool to be living in this moment. Yeah. I know. Of course. Yeah. I I yeah. I I think you're I agree with you on that that the chips people are buying this hard people buying this hardware are definitely gonna regret it because NVIDIA keeps coming up with, like, more efficient, much cheaper hardware

every year. Right? Like, I also read a post recently that, like, the hardware that was sold for, like, $40,000 two years ago is being resold for $8,000. Yep. That's that's like 75%. Right? That's insane. So, yeah, there'll be a lot of, like, Microsoft spending billions on, like, these NVIDIA chips are gonna I don't know how it's gonna go. Yeah. It will commoditize maybe slowly. But yeah.

And I will say on the AI compute on Bitcoin miners and the convergence of AI hosting and stuff, That has been fascinating to watch. I mean, my investment fund ten thirty one was one of the few large firms that was investing in proof of work mining. Because if people remember, Bitcoiners remember, even just two years ago, most of the crypto industry was saying that proof of work and energy use was the problem that needed to be solved. Everyone was moving to proof of stake.

Now, of course, all the big tech guys are all like, we need more energy infrastructure, using energy is good for humanity. They're all like repeating the Bitcoin or talking points. So the Bitcoin miners were already there, right? The Bitcoin miners were there before everyone else realized it. And a bunch of them have at least soft pivoted into AI like they're doing a combination play. And my partner at ten thirty one Marty and also my co host and my other podcasts, he calls it mullet mining.

Do have Bitcoin mining in the front? You have Bitcoin mining, AI compute in the front, Bitcoin mining in the back because they have different uptime requirements. And it probably is a combination. Right? The most successful operations will probably have some kind of combo play, which is kind of fascinating to watch. Yeah. No. It it has to be like yeah. Yeah. I don't know. It feels like the intelligence, how it's gonna be priced, it's gonna be very interest it's gonna be very interesting.

Yeah. And then you have all the big tech companies that their CapEx spend, they're spending a shit ton of money building out things except for Apple, by the way. Apple's like the only big tech company that's not spending a shit ton of money right now, which I think will probably pay off for them. Yeah. They just caught this hit the hit the jackpot with the m one chips Yeah. And they're all Mac minis, you know. Somehow, like, Mac minis are one of the best machines to run LLMs locally.

Well, you know why people are buy like, people are buying the Mac minis because they want the Apple ecosystem. They want their robot slave to be able to use iMessage.

That's basically what it is. And they're obviously not self hosting. And then people are paying like $10,000 for these Mac studios, which once again, I would say to people that are thinking about doing that, just know what you're doing. Like, you're not saving money. You're not doing it to save money. You're doing it because you have disposable income and you wanna learn these things and you wanna self host, you know, maybe a model or two.

But you're not don't think you're doing it to save money and don't think you're doing it to have an advantage over other people. The person that is using a hosted version of Opus is gonna have way more capability than whatever you're if you're like, self hosting Kimmy k two $5,000 studio, like, you're probably still gonna hit hosted models for a lot of things anyway.

I would recommend if anyone thought about buying like a Mac mini, just look on like, some like, used goods platform, like Facebook Marketplace or something, and get, like, a really old, like, office PC. I I bought, a $50, like, computer, and it was good enough to, like, run OpenCloud on it. So I just have, a $50 machine. Like, I would pay more, like, renting a good server for a few months. It's like Well, the cool part about OpenCloud is you

cool part about OpenCloud is you can run it on a piece of shit computer. Like, you can run it on a RazPie. Probably not that great, but it technically would run.

But once again, the reason to be clear, like the reason people are buying Mac minis is because they're stuck in the Apple ecosystem. They want access to iCloud. They want access to iMessage. And obviously you can't get that on old IBM computer. Yeah. I mean, if your cost is if, if cost and freedom is your, is your goal, then yeah. Buy and just buy an old PC, maybe add a little bit of Ram to it. Don't go out and buy a ton of Ram. You probably aren't going to be self hosting much if anything, and just experiment. But I also I really like the VP. I think for most people right now,

just spin up a cheap VPS. Then also, don't have to worry about the security concerns of having this thing on your internal network either. I think most average people will have trouble isolating it in their internal network, and you don't you don't really want a robot slave at this at this point. Everything's so early, and you have this thing in your home network and it's fucking around in your home network. There's a lot of damage it can do. So just be wary of that as well.

No. Yeah. No. I agree. Also, like, I think a Linux VPS is probably more suitable for these agents than, I would say, Mac. Because, like, on Linux, since all of Linux recommendation is open source, everything has gone into these LLMs in the training stage. So they pretty much know all commands that they need for Linux.

I think if I were up Apple, I would actually be worried because Linux is actually gonna be ease use sorry, usable for a lot of people because agents can use Linux the best among all other systems. So Yeah. I think you know, Linux. Yeah. It just goes with the the trend that it seems like

these agents, these bots are removing a lot of the friction points that Freedom Tech has had up until this point. Because if you're just prompting, if you're just sending a chat message, if the UX for the average person becomes a chat message, then all those pain points become obsolete. They're no longer pain points. The bot's handling the pain. Yeah. No. Absolutely. Yeah. And and like, I think Apple, like Apple

keeping this iMessage closed source and like WhatsApp blocking, rate limiting IP addresses, saying, okay, there's some bot is using this. I think it's all bullshit. It's also they're all gonna have to shut down, shut shop in the future because, like, agents will be the main users for most APIs. Right? And like Yeah. You either switch to a paper request model or just like go obsolete.

That's where like, cashew is like the best case scenario for this. Like, it's just pay for that particular request, and there's no dadocing. There's no I mean, that's like a solution for dadocing. That's a that's a solution for rate limiting. Just let them pay for it. Pay per request. Right?

I love it. Yeah. I mean, I my last show, I had Justin Moon on, and he's been working with a bunch of guys on a chat called Pika that uses Marmot as its back end, which is the protocol that white noise the white noise developers have made, which is using Nostril Relays for encrypted DMs. And if once again, if you use one of these bots with

and you try and use it with any of these closed chat systems, whether it's iMessage or Signal or WhatsApp, Telegram is probably the best of the closed ones because they've kind of embraced the bots. But even if you try and use that with a burner phone number, they often ban your account. It's very difficult. These open systems will be the ones that

the agents use to communicate with each other, but also with humans. Being able to communicate with your agent should be as simple as pasting an MPUB. It shouldn't be connecting into a proprietary walled garden. And that friction will result in people moving to the open systems and agents moving to the open systems. And I think there's a lot of hype around this concept of agent agentic

payments or agentic comms. Well, what people miss is it's not just agents to agents. It's it's humans to agents to agents to agents to agents. It's it's all of it. And all the proprietary walled gardens is completely break down when you have those types of interactions. So open open protocols will win that. It just will take a little bit of time. But I don't use I I installed Pico as well. Shout out to Justin for building this. This great. I think we we need interoperable

systems like Marmode to just work. I wanna move away from signal. It's it's been too long. Yeah. I mean, I love signal. Signal is one of my favorite projects, but try setting that up with your bot, and they're actively rate limiting you. They're actively throttling you, trying to stop you from doing it, which is kinda wild.

Yeah. And you need a phone number. Had to like make a burner phone number and burner signal account, and I still was having issues. I mean, another perfect example is, you know, Facebook, who's the juggernaut in the room meta, like they came out with their own open call competitor. And by default, it uses Telegram because WhatsApp is such a piece of shit when it comes to bots being able to communicate with each other.

And they have billions of dollars to spend to try and make their systems more set up for it. And they're using the competition because they're not ready for it. Yeah. I think, like, it's it's time we give Facebook the run for their money. You know? It's like Nostril

should I feel like Nostril, it's about time. And so I feel like Nostril is gonna blow up in the next couple of months. It has to. I agree. Like, Facebook is just too complacent. Like, all of these big all of these, like, big tech companies are, like, too complacent. And we we will definitely do this here. I'm very bullish. They move too slow. Like, I had Justin Moon on the show two weeks ago, and I'm in the development group chat at Pika.

They're using LLMs to to build it out. They keep shipping new changes on on, like, a six hour cadence. They just ship, ship, ship, ship, ship. There's nothing blocking them from shipping. And meanwhile, it's, like, seven humans working on it, but it was like a 100 agents are working on it at any given time. It's fucking wild. So I think this stuff accelerates really quick. Anyway, guys, we've passed the one hour point. I thought this was a fascinating conversation.

I think routesters an awesome project. I would love for us to do maybe like six month

check ins and just constantly let the freaks know where we stand and how you guys are looking about the future. But before we wrap, it'd be great to just go around the horn and have you guys each give your final thoughts to the freaks. And I guess we'll start we'll start with I mean, Evan, you've been quiet. Why don't we start with you?

Yeah. Sure. So, yeah, thanks for the thanks for the show. And I think we're gonna see more and more agents in the future. And I think, Rouse is the final piece of the entire ecosystem. So we have, cache as a payment protocol, and we have, like, like messaging layer or, social layer or whatever. And we have, and now we have Rousers, the AI infrastructure. Yeah. I think we're gonna keep growing Rousers. Longest. Abdu, final thoughts. Yeah. I was quite because I Shuram and Red

Does already a great job explaining what we're doing and what we are going, but I think we have to make more effort to bring it to the public, to bring more regular people that do not know how to program, to code, agent and stuff. The onboarding of those people is more important also. So we have the more public space. And yeah, that's a lot of money. Thanks, Abdur. Shermanek, final thoughts? Yeah. I think Nosto is the perfect playground.

I think it will be really interesting to watch agents grow and, live inside Nosto. And yeah, I think it will be really, really, really interesting next few months. Love it. Radshift, final thoughts. Yeah. I know. Mean, like, this has been a fascinating conversation. Thanks a lot for inviting us. Yeah. No. I think, like, Rowser is is the solution for for developers, agents, anyone using LLMs, you know.

It's I I would say, check out our house to check out our products. Please give us feedback. We just want feedback. We want we wanna improve our products. The experience on experience of using Freedom Tech should just be perfect, you know. Because, like, this is the tech that needs to exist, and we we are building this, and we want to we wanna make sure it's, like, the best you get the best experience because I should get the best experience.

So, yeah, I just request all of you guys to check out browser.com. We also have an OpenClub script that you can run. We're gonna have more options for VPS hosting too soon. And I'm also working on an SDK that will help you just instantly switch on the go, local cache wallet, and more social features. I think not we're gonna lean in the coming months, we're gonna lean so much more into Nostril.

Yeah. You will wonder whether it's actually an AI product or a Nostril product. Yeah. Thanks for having us. And, yeah, master to the moon. You guys down to come back on in six months for a check-in?

Oh, for sure. Absolutely. Glad to. Awesome. Yeah. I'll put all relevant links in the show notes, freaks. Play with these things. Give them feedback. You guys are crushing it. You're doing great. Keep it going. I'm a big fan. Very excited. If I can be helpful in any way, you have my contact now. So feedback. Don't forget to leave a big sap. Thank you, Shermeneg. Freaks, as always, all relevant links are still in this citadeldispatch.com.

Share with your friends and family. I hope you appreciate this conversation. The next rip is on Friday. So stay tuned for that. We're gonna talk about how Noster's revolutionizing networking. It's going to be a really fun conversation. It's with the Phipps project. So I'm very excited about that. And like I said, I have a bunch of shows lined up for us as we dive into this deep rabbit hole of freedom tech that just seems to be accelerating. Anyway, thank you, guys. Thank you all.

Thank you. Thank you. There we go. You're good. Thanks. Stay on the stacks sides. Peace.