Transcription: Episode 221 - Introducing Pylon

Christopher David (CD): And we're back. This podcast may contain content inappropriate for children. Ignore this part. Car, welcome to episode 2. Thanks so much.

Car (C): Thank you.

CD: I'm going to try not to talk three times louder than you. Where do I sit? Okay. All right, so here we got some Zerglings.

C: They look very Zerg.

CD: You know, Zerg was probably my favorite race to play because I'd make like 255 of the fully maxed out Zerglings and then just go into your base. All right, we got some Protoss dudes. Okay, why am I showing this? Because we're launching a product today called Pylon and the headline is that Pylon is our compute miner so you can sell your compute for Bitcoin. And we'll get into it in a second, but I just want you to get it, okay? So here you got Protoss guys. In the center you got a Probe. Okay, Probe is what we're calling our coding agent because it's like the dude who goes and builds stuff for you, right? Well, what's he going to build here? He's going to build a Pylon in a second. Okay, so what else do we have here? So, so here's the Probe. Nice. Mew. Now it's going to take a little minute, it's going to take a minute to build.

C: There it goes.

CD: Now here you got some little suicide bomber dudes and the Zealot will hold them off. Okay, so what else do we have here? The Protoss have this like psionic power that they shoot that's like their mental telepathy thing. So what's our ML framework that's written in Rust and going to absorb all of the PyTorch and other crap from the Python ecosystem? That's called Psionic. So we got Protoss, or we got Pylon and Nexus. Nexus is like the central Nostr relay super server. Ultralisk, uh oh. We got Pylon, we got Nexus. Now these guys are High Templar. You can always merge two Templar to make an Archon. Oh, look at that. So we gotta have a product at some point called Archon. That's gotta be like our, maybe that'll be our version of Mythos, lol lol lol. Okay, so we got Archon, he's got shields. Man, it'd be nice if that Pylon was done yet. We need a Pylon here guys.

C: It takes time.

CD: Yeah, Archon versus Ultralisk, you know. So the Archons have a ton of shields, but once you get through the shields they're pretty weak.

C: So the Probe is just right now, he's building out the Pylon.

CD: Probe, Probe is floating around. Once you start a Pylon you can just kind of walk away from it and it'll load in. So we don't know what the Probe is, Probe is probably hiding right now. And I guess he only had the minerals to make one Probe, one Pylon. Archon, that's not even really an ability. Okay, so. And I asked ChatGPT, I was like "Who is this Zealot? Is this like a named dude?". This is just like an unnamed Zealot. Man, that Avengers scene probably stole from this scene. Everyone steals from StarCraft without acknowledging it. Guess what? Pylon boom!

C: And then everybody shows up.

CD: They kind of like teleport in. So when you have a Pylon you can teleport in. Just imagine these are open agents, okay? Open agents have come to the rescue and we're going to fight off the bad Anthropic Claude army, okay?

C: Look at that.

CD: And so I just, I just want to tell you Microsoft who owns Blizzard's IP, if you ever send me a single legal document, here's what we're going to do. If I ever get something that says Microsoft doesn't like you using Blizzard IP, I will say "We will destroy you, knock you out of the Mag 7, hostile takeover all Blizzard IP and create StarCraft 3 and World of StarCraft and we're going to do it right, okay?".

C: For the record, Microsoft is known for cease and desist in the Bitcoin space.

CD: Do not f with us. Car, do you know what my favorite number in the whole world is?

C: Sats?

CD: Well, it's a specific number and I can't tell you off the top of my head. But I can tell you that my favorite number ever is right here. It's 4,826.

C: Dude, it even went up more than just right now.

CD: This is my favorite number. Look at those Pylons. We got the Pylon picture up here. 4,000 okay my favorite number is no longer 4,826. My favorite number is now 4,842.

C: Wow dude.

CD: Note I didn't even have to refresh the page. This is just live hooked into our Nexus, okay? So here's how it goes. Pylon 60. Okay, so um we've done two previous videos on Pylon with kind of like different versions of it, but the common thread between all of them is it's a piece of software, it's a node, and it sits on your computer. It's a Nostr client, it's a NIP-90 service provider, and you can Google Nostr NIP-90 data vending machine to learn about like the actual protocol that is used. Um but basically it's a way for people to buy data from you for Bitcoin or something else. Um and then we did a proof of concept a couple months ago with a Pylon and a Nexus. So Nexus is basically a glorified Nostr relay. It'll have some other um things specific to Psionic to allow, you know, hook it into like different AI workloads. Um but the Nexus, we're running a Nexus. You can see in the source code that like we have a nexus.openagents.com. Oh, there's even something there. I don't even know what any of that is, but just some vibe-coded crap. But anyways we have a Nexus that's operating here. Um the code for Nexus, so for example let's say you want to run your own Nexus and you want to buy compute from people and you don't want to use our network for some reason, you can go to our OpenAgents repo and all the Nexus and Pylon codes all in here in the crates. Um and you can run your own Nexus. Uh but for now you shouldn't care about that. All you should probably care about is the fact that you now have a single thing that you can go copy your agent instructions, paste this to your Claude code, paste this to your Codex, paste this to your Hermes thingy, your open Claude. Here's what it is. This may evolve over time but who cares? You're just going to paste this to your agent and um I'm not sharing that screen you can't see but like Car, does this work?

C: Dude, so I ran it yesterday just threw it inside of...

CD: Online now just ticked up from 6 to 7.

C: Yeah I just threw it in yesterday into Cursor and then it just started, just started compiling and then it just did its thing and then it launched on my Air and now it's running and making Sats. It's pretty cool dude. Like and I don't have to do anything. That's just um that's just out there turning spare compute into Bitcoin.

CD: Yeah it should be the simplest like "How are you going to get product market fit? We give you a button that you press and it makes you money!". Is that the simplest fucking thing ever?

C: But I even like the Substack article that I wrote today. When you think about it more long-term, I think one of the most interesting things is around this kind of like global kind of like economy layer that's being built right now.

CD: Yeah take us through this article.

C: Click like that right there, yeah. Click that one. Yeah, click that.

CD: What percentage did you write versus GPT?

C: No, a lot of it me and Ovi wrote it together. Oh good. Yeah me and Ovi wrote it together but if you scroll down, like one of the biggest things and it's in that first line right there. So like we definitely do think that some of the models alone aren't going to be able to solve a lot of the um how would you say, like a lot of the economic layer. Like it'll be able to do enough but at some point you're going to need to create like a global economic layer that's going to live beyond that. It's um and you're already kind of seeing it because Visa's kind of doing it, Coinbase is kind of doing it, Cloudflare's kind of doing it, even Stripe had it in its annual letter. But the more you look into it even BPI just released some research um last month I think it was or in March. But the more you look into it the more you realize that the future where everything is headed it's kind of inside of this um this interface that's going to kind of just start forming around all these other things and it's I would even say it's beyond agent to agent machine stuff. It's a bigger global pie than people realize.

CD: Software is everywhere and as software becomes more intelligent...

C: People are underestimating...

CD: We're going to be surrounded by these agents and you'd better hope that they are open and neutral.

C: So yes I agree but it's even beyond that Christopher because like what you're talking about is like the entire economy of the world running on this economic layer. And when you start thinking in terms of like five, ten years out, just at the pace of everything that's going, this could um if you're not in this area, if you're not building infrastructure in this area you're probably going to be um left out of that economy in a lot of ways. So this is kind of like the bigger thesis around kind of like what OpenAgents is doing is just like this is going to exist and if you don't start building this infrastructure now um for Bitcoiners, for open source, we're going to lose way for all these other closed systems.

CD: Yeah we've got to definitely build out the Bitcoin version of what the big guys are building. So you've got Stripe, you've got Coinbase. I mean Cloudflare, I had hope for Cloudflare but they're just basically taking bids between Stripe and Coinbase for what's going to power their like internet dollar shitcoin stable. Um we've got to build the Bitcoin version of all of this.

C: And in my personal opinion, and I don't have, I haven't really looked into it as much we have one guy working on Quantum inside the lab but I think those are the two fronts. I would even consider this a way bigger problem for Bitcoiners in the future than even Quantum because if we're not participating in that economic layer and if we're not dispersing capital into that ecosystem then what's going to happen? It's going to lead to the same situation that we had in the fiat era where everything is controlled by SWIFT and it's going to be but it's going to be a different sort of group, right? And that's the biggest problem that I think that's facing Bitcoiners right now that we're actually not paying attention to.

CD: I think um it's serious, it's really I think it's serious.

C: I think a bunch of people are starting to get the importance of like Bitcoin and Bitcoin versions of AI and we've got um a bunch of people in, I mean heck they, one of the reasons they stated for like Austin BitDevs taking a pause is you got a bunch of people in our community here who are like super excited by AI and that's good. Like I think that's a good thing that we're like shaping new frontiers. Here's a little shill for our panel in Vegas. Oh yeah!

C: Dude you're going to be up there with Roland? Snap that's going to be badass.

CD: Why AI agents want Bitcoin. Do agents want Bitcoin?

C: Or it's not even if they want Bitcoin, like if we don't build out this infrastructure layer we're not going to be able to participate.

CD: It's a funny misnomer. Like agents don't want anything but anyways it'll be a good topic of conversation. So bunch of people are starting to build on this and then my argument for like we just need, we have the benefit in the Bitcoin space of everyone already speaking the same language. We speak Bitcoin at the base layer, we speak Lightning and other related L2s that all use Lightning for interop. Like there's an increasing consensus around Nostr for this sort of like data layer that's adjacent but not like spraying data onto the chains. So we've got all the makings of like decentralized applications, all the substrate for what should be like sort of the agentic AI layers for that. And one thing that I think is going to help the many people who are building little infrastructure pieces in isolation is to have some sort of unifying context here that we're trying to get at with Pylon. Which is like if you, if everyone, like literally everybody can use these Bitcoin, Lightning, Nostr and little like tech based on top of it, can do something really cool like plug in any spare compute that you have and make that available into this open network. We are going to be using it. We are buyer number one. We want your compute. But guess what? All of this is on an open network. It's Nostr NIP-90. If someone else wants to come in and outbid us for your compute and pay more than we do, we want that to happen because we want this sort of like thriving open marketplace to happen.

C: I think one of the cool things that I think that just from talking to you these past month on all this stuff, especially around the Rodeo when you were kind of telling me about this, like the more I realize is like it's thought out really carefully and intentionally because you have to have this part first, right? Before you get to all the stuff that like Alby's doing, right? Like the actual sandboxing of like being able to pay for things without like a digital identity and stuff like that. And I think everything is kind of intersecting right now where you're kind of seeing even the Nostr stuff is kind of like maybe they'll get to a DID at some point and then like this infrastructure has to be built and then at some point all that just works open source. Then that's when we get to that bigger pie, that global pie.

CD: So this is a little bit what it looks like. Um we're evolving the UI but it should just be like a very lightweight um terminal UI. Uh here's how you download it. Well I mean this is all you do really. You paste that in to your agent. Um we'll have some other ways for you to install it other than that. Uh you should be able to install this from NPM. Um but the stats will keep ticking up. We got the number of Pylons here.

C: That's crazy.

CD: We note that like you know the biggest training run ever from BitTensor Subnet 300 billion, I don't know what it is, but like they had 70 people contribute compute to a training run. Um Pylon which right now we're doing very basic inference. Your Pylon will load a Gemma 4 model. We're kind of like collecting telemetry to see what your device can handle. Can we reliably pass you um inference requests for a lightweight Gemma 4 model, one of the bigger Gemma 4 models, something else from Hugging Face? When we launched a version of this a couple years ago we had support for inference, fine-tuning, image generation, embedding. So your machine, your, the purpose of the Pylon is basically you're saying "Yes OpenAgents, like here's the subset of my computer that you can use for stuff. You're selling that, we're paying to your built-in Bitcoin wallet." Um in the next few days, hopefully by Friday but probably more next week we're going to be adding in the sort of decentralized training infrastructure. So we're going to be doing actual training runs. Um aside from inference, like we're definitely going to be doing inference but we're mostly excited about doing inference not just for open models generally but also for open models that we train ourselves using this same system. So the kinds, like I was saying on the last episode like the multi-billion dollar training runs, the multi-multi-hundreds of billions of dollars that are allocated by the massive AI companies for compute. A lot of that goes to training. It's like instead of that going out to an Nvidia and these big cloud people, like why don't we pay that to actual people? There's still a bunch of kind of like ML research that we need to do because no one's really optimized for what massive decentralized training runs on the edge can actually do. There's been great work and even though I make fun of them for doing shitcoin stuff I have to like give a lot of credit to Templar in the BitTensor network and Prime Intellect, Nous has got some stuff on this where they've taken the ideas come out of DeepMind for DiLoCo and some of these algorithms for decentralized training and they've put them to use. Now they've generally attached these shitcoins that I think are completely unnecessary, but hey if you have a you know 70 people that will contribute compute to be paid in your shitcoin, if we make it easy for Bitcoin, the massive biggest most secure coolest awesomest network, why don't we get 700 people or 7,000 or 70,000 people contributing their compute to this network? Can we build an entire vertically integrated AI stack on top of retail compute and get you paid and get you paid and get you paid? Um this is what we're building.

C: Yeah I think you put together a good team man. There's um I think one of the interesting things that you have is like some people actually doing research behind the scenes on all this stuff so it's not so much on your plate to get it there. So it's great that we have people who have contributed in the past to open source bounties, developers um starting to come back. Um we've got people outside of like our explicit community in the open source community like Sarah I mentioned last video. Um there's a bunch of kind of people in the open source and local AI community that we're excited to work more with. Um I'd like to get those people hooked up to revenue streams where they can be contributing their compute into our network. Um why not get paid Bitcoin for all of this stuff? So yeah we think that one thing realistically this will look like is that we're going to be like running slash leading the largest AI lab, decentralized AI lab um this year.

C: Yeah I would say get to the GitHub and like start contributing. There's room there to experiment.

CD: Let's touch on a couple other topics. So today big news. Meta is back! Muse Spark scores 52 on some benchmarks blah blah blah. New model from Meta. We were given early access. It's the first frontier class model from Meta since Llama 4... oh wait. Llama 3 was released in April 2024. This says July 4th, 2025. This post is a prediction for the future. "The model is not open source and is not yet accessible via an API but Meta has shared they expect this to come soon." Trust us bro, it's really good.

C: So he's not, so they're no longer doing that anymore?

CD: No they're done. They're off, they're off team open. Uh yeah unsubscribe. I don't care what your closed thing is. Losers! We forecasted this, we foreshadowed this back in July of 2025. New York Times said that Meta was reconsidering their open approach to AI and we trumpeted this loudly because it was very clear that they were just ditching team open and they're going to keep being mealy-mouthed and maybe we'll... no they and then we say like "If you leave team open you're going to lose." Now here is a nice post from Nathan Lambert. So I mentioned, I have a slide in our pitch deck now where I talk about Allen AI and how they've been talking a whole bunch about let's build American DeepSeek. Yeah we think there should be an American DeepSeek. And but guess what? Uh this team Allen AI just got their core team raided by Microsoft uh few weeks ago and they had all this funding and stuff in motion to train their own models and they no longer have the resources to train their own models. They got raided and eviscerated and this is what helped me realize that like oh you know what? In the past when I saw this I was like yeah we need that, but I don't have the credibility to lead that because I don't have an ML background. And I've now come to realize that the relevant part here isn't the ML background of who runs the thing, it's probably the ability to tell the big labs "Fuck you, you come try to acquire me you're going to get a big middle finger in your face. You come after our people and try to acquihire our people, we will destroy you." So you know we're going to fight you, we want to fight all these people simultaneously.

C: And you're kind of seeing it in Austin already. I mean you posted to Slack earlier about like the Austin's going to be more of the open, how'd you say that? You were like...

CD: I think I think SF will be the sort of closed, you're going to have these sort of nerds who want to get their tendrils around DC through regulatory capture uh because they can't actually compete. So here's, we'll close with this, here's my prediction. This is in response to Anthropic releasing some managed agents platform, everything you need to build and deploy agents at scale. It's an agent harness, like you would think that as OpenAgents building platforms for people to use agents uh that I would really care about this, but no. Anthropic is going to keep releasing vibe-coded slop every day that sounds cool but no one actually uses. Claude Code and related products have done well enough to let them bullshit their way through enterprise procurements. Their revenue will keep increasing for a while. They spray a few more months of bullshit then they IPO, exit their early backers, step up lobbying and regulatory efforts loosely coordinated with OpenAI's trying to hold off open source competitors like us. They get a few lefties excited about commie UBI programs in elections '26 and '28 but nothing substantive happens aside from performative regulatory efforts that do nothing but get lobbyists paid. Then what's going to happen? Team open meaningfully catches up on models and products and we begin our own march through the enterprise. Everyone realizes the regulatory capturors were primarily bullshitting. They oversold benefits and risks to cultivate an aura that helped with sales but not outcomes. Team open delivers the real goods. The benefits of productivity, deflation and dividends finally start materializing and benefiting normies. Team America merges with team open led by OpenAgents based here in Austin. Anthropic and other closed AI companies become a laughingstock and historical footnote. We build floating data centers in international waters aka Atlantis and go to space and begin a millennium of human-led AI-assisted galactic empire. Remind me of this in three years.

C: That tracks, yeah.

CD: That tracks. That tracks. Uh we'll end it there. You can help us in this effort by selling us your compute. We want to turn it into cool stuff. Uh go to openagents.com, click the button and uh watch the numbers go up. Help us make these numbers get bigger, okay?

C: Dude that's crazy.

CD: See you soon.