NanoGPT | Closing the loop on crypto AI Agents - Milan de Reede | ETHDam III - 2025

Welcome to East to E to E to E to E. Uh, okay. Hello everyone. Thank you for coming to listen to this talk. I'm going to be talking about nano GPT, about AI, about AI agents, which I realize especially combined with crypto is a lot of buzzwords altogether.

Um so I'll try and keep it practical and not go too deep into just making this sound good. Uh so first of all one of the issues that I think Marco also talked about at the very first talk of today is AI is quite centralized right there's a few big parties open AAI Google Elon Musk with his XAI and this kind of makes sense because in AI the bigger the better if you have ah perfect if you have a big scale you can train bigger models there are better models so you want a lot of money to pour into creating these AIs. But that kind of means that these big companies have the best AI models. Nope, that doesn't work. Does it work?

No. Um, and so the issue with that is these big companies then have a lot of power and what they do is they quite restrict the access to these models. So, for example, for OpenAI, for Grock, all of these different models, you have to pay usually about 20 bucks a month, for some of them more to have access to the top models. And I think all of us here, we can pay pay $20. That's fine.

But there's a lot of people in the world for whom that's just way too much. Um, so you have to be rich enough to be able to use it. You also need to have a credit card. And I think we're all quite enthusiastic about crypto. I think Joelle asked uh Joel from Dash and almost everyone here had done crypto payments.

Uh so we like crypto but for the majority of the world um it's just the case they don't even have a credit card to do payments with. So in many ways you have to have money you have to have a credit card but also you have to give up a lot of privacy when using these models. You have to trust your government. You have to trust OpenAI. Again Marco already mentioned this.

You log in with your name. You pay with your credit card, which has your name, and then you start talking about all these weird kinks that you have, uh, all these weird fantasies. You start asking for relationship advice with your girlfriend. Uh, there's some stuff in there I wouldn't want anyone to see, and it's kind of scary if that's all linked to your name and all on there. Um, so you kind of give up privacy pretty completely.

Now, of course, oh man, forgot that this doesn't work. Now, of course, we are here because we think part of this can be solved through crypto, obviously not all of it, but what we can solve is instead of having a subscription to use these top models, you can just deposit a few cents in crypto, only pay for what you use, and just use the top models when you need to. We have people that deposit $1, and then that lasts them quite a while. Uh, this is of course only possible with crypto because low amounts actually work with crypto. Second of all, crypto is actually accessible to everyone, which I'll get to later in the demo as well.

But we have many users from Kenya, Nigeria. They have like no chance of getting a credit card, but they do have some crypto or they can earn some crypto. We have an earn page on the website as well. So, we're trying to make it so that everyone can access these top models um through crypto in a sense. But then third of all, and I guess most importantly, it gives you actual privacy.

So, when you use NanoGBT, and there's a few other services that probably do something similar, you don't even need to make an account. We don't care who you are. We prefer payments in crypto rather than credit card. And we don't see anyone's name when you pay with crypto. So, we also don't log anything.

So, if you do a prompt, we don't know who it's who it's coming from. We don't store the IP. You hopefully paid with crypto. So, even if we did store it, we would have no clue who to link it to. And in that way you can have at least a layer of privacy when using these models.

Um I always want to say for the purists using local models is even better. You can run them on your own computer but it's usually harder to set up and that way you will never have access to like the latest chat GPT and Gemini and whatever because there's no way to run those locally. Um so what we try to do is if you want to use these top models at least have a protective layer of privacy. Now this is popular with our users obviously and we have sort of three classes of users. We have individuals like just people from poorer countries who cannot afford a subscription.

We also just have people who want to use all the different models in one place. We offer literally every single model you can think of. There's a list in a few slides, but like literally if you can name a model that we don't have, I'll be very impressed. Um, so some people just like playing around with all of them, all the text models, image, whatever. Some businesses like using it because for different use cases, for different employees, you want different models, right?

Someone in marketing will need a different model than someone working as a developer. And it's easy to have all of that in one place. Uh, but what I'm here to talk about mostly today is the AI agents. So we also have people who use our platform to run AI agents and specifically crypto AI agents which I think usually crypto AI agents is just a bunch of buzzwords like okay we're trying to increase the value of our token by like appending AI at the end. Um but we have people who very proudly share with us how they can actually do this without ever issuing a token but have AI agents running that make them a lot of profit.

Um, so I created a very simple demo. Um, let's see. Oh, yeah. First of all, AI agents, for those that don't know, I hope most people know, uh, you can call pretty much anything an AI agent if you're creative enough. A thermostat takes in information from the outside world and takes actions, changing the temperature.

Um, but I think the ones that we actually find cool are the generalized agents that we have nowadays, right? chat GPT Gemini agents where or AIS where you can speak to them in natural language and actually get them to understand you and take actions. Uh so what I'll be demonstrating is not a thermostat but actually a generalized agent. Um let me see if I timed this roughly correctly with this. Yeah.

Okay. So what we have running here um if I refresh Yeah. what we have running here. No, it's not showing up on the screen. I'll move it to the right screen.

There we go. In a world of AI, we still have these issues. Uh there we go. Okay. So, what we have is essentially just trying to show off the different stuff that you can do with the models on our website.

It's a very simple demo, but like let me walk you through it and you'll understand why it's actually kind of cool. So, first of all, on the left we have an AI model like chat GPT create prompts. So, essentially it's been told we want to see some cool images. Can you generate some prompts to make a cool image? Which, you know, is a very simple prompt.

Sometimes it's like a steampunk hummingbird sipping from a neon flower. And then you get this cute little picture generated by an image model. That's kind of fun. Uh, but it's mostly just for shits and giggles. Um, what is a lot more useful already is you can then have a different model like Gemini in this case analyze what it sees on the image.

Now, for these images, that's not very useful. But in general, there are a lot of use cases where you would like a model to describe what it's seeing and to understand what it's actually seeing because that way the interaction with the real world comes through camera like our eyes which we're used to rather than only being able to send text and read text. Um, so what it does is it describes what it sees in the image. I don't know if you can read the text or whether it's too small, but we then feed it into yet another image model um which creates an image based on the description. Um I see yeah it doesn't always go with the exact same thing but essentially this is just a few like regular text models, image models all doing their thing.

What we have at the top however is a balance. So I didn't write this program myself. Uh an AI did. I literally did not write a single line of code for this demo. But what it built in is that there is one manager AI that's constantly checking the balance on our service.

So it checks how much it has left in the wallet on the service. And it's slowly running out, right? Because every prompt that it does, every time it creates an image or creates some text, the balance goes down because, you know, you have to pay us. Um, so it's slowly going down over time and it realized that if the balance runs out, the program stops and my task was to make a good demo. So we have to make sure the demo keeps going.

So what we did was we gave it access to a tool and with the tool it can not only call our service to check what its balance is, it can also check the other API stuff that we have on our website. And one of those is generate an Ethereum deposit address. And another tool that it has in its tool library that that it can use is send an Ethereum transaction. So it wrote some code which essentially says if my balance goes down below I think it's one. I hope it's one otherwise I timed this very badly.

If my balance goes down below one then I need to send a few dollars into uh this wallet to top up again. And what it then does is it calls on Etherscan. It uses the Ether Scan API to check the current price of Ethereum to dollars and then it will convert like $5 or $3, I don't know what it is exactly, into an Ethereum amount and then actually use a tool to send in that amount to our wallet which then tops it up again and so the program can keep going. Now, I'm really hoping this happens at one nano because I'll be honest, like I literally didn't write this program myself. This was created like two hours ago by one of the AI models on our website.

I think a few people here saw me building this and sort of trying it out. But yeah, um it should then like show on screen that it's topping it up and then top up the balance. Um in the meantime, ah there you go. Okay, so it's calling a tool. Uh, it just called like get uh deposit address, get the price, send Ethereum transaction, and then it's sending like $5.

Hi, Milan. I've sent 0.002155 E to NanoGPT. Transaction hash 027C. Okay.

Waiting for confirmation. I don't know how long it keeps talking, so I'll just shut it off here. Um, but so it just sent an actual onchain transaction, and I can go to Ether Scan. uh which should be here I guess. Yeah.

So if I refresh here we should see that it's just h verifying that I'm human is becoming increasingly important I think. Uh let's see. There you go. Right. So we see that it just did a transaction.

It actually did two because the balance is still too low. Uh it actually did a lot. Okay. I hope it's not draining my wallet. Uh, luckily it's paying my own service, so it should be okay.

But essentially, it's trying to top up its balance because it saw that the balance wasn't increased yet. So, it sees, okay, my balance is still low. Um, I'll try and go back to the demo page. Um, so it sent in some ETH. Unfortunately, ETH is not always very quick because I don't know.

Ah, okay. Now, our balance updated in the top left. Um, so essentially this is a very simple demo, right? But people are using this, like I just mocked it this up in like an hour or two. People are using these AI agents with tools that allow them to send onchain transactions to build actually really cool stuff.

Like this one guy, he told me very proudly about this and allowed me to share this luckily. He built um I guess you all know Poly Market, right? Like yeah, okay, I see some people nodding. He built this whole team of AI agents. It has one manager agent at the top that checks poly market and gets all the current markets that it sees.

Then for every market that it sees, like for example, what's the next Romanian president, it will query a bunch of other models, sort of its employees to start gathering information. Um, you can give web access to our models, you can equip them with tools. So these models go out and search and they report back to this top model and they'll say, "Okay, what we found is actually the polls say something slightly different, but also the Twitter sentiment says this, whatever." The same thing any of us would do if we were trying to figure out what the odds are on a thing happening. Um, this manager model can then make a decision for itself.

Is this a thing I want to invest in or not? like maybe it thinks a different candidate has a higher uh probability of being chosen, right? Like we sometimes think, okay, maybe maybe Camala will win and let's play some money on that on poly market. This manager model that he created can then like propose to another model to invest in this uh which is the same way it would work in a human investment firm, right? You write a proposal, you get someone to say go no-go.

If that model says go, let's do it. it can do the transaction itself like it will actually invest in the like on poly market itself and then afterwards it will sort of just check like okay are we making a profit are we not making a profit um all of this without any human oversight without any fiat in the chain and without any human necessary um the cool part for us is he's using our service he can pay an ETH so he's making ETH this model on poly market and then when he sees that his balance is running low from all of these calls that he's doing, from all of this research that he's doing. He just tops up on our service again to put in more ETH because he's like, "Wow, I'm being profitable. You know, I want to keep this going." This program can run completely autonomously without a human, but this is kind of why I called this closing the loop on crypto AI agents.

He also doesn't ever touch fiat. like there is no need for any credit card for anything that he uses in this service. Everything is paid in crypto. Um so yeah, that's uh the very short version of of what he's built. It's actually a a massive program.

He's told me he it's running for like weeks at a time without him even looking at it. But so when we talk about AI and about crypto, I think the cool thing about it is that we can actually close the loop. we can build something that can do stuff online without needing a human, without needing needing any fiat. And I don't think that's been possible before. And that's just not possible unless you can use crypto, which these agents can now do.

Um, so I'll shut down the demo. Um, yeah, I think my balance is still okay, so we'll probably just keep going indefinitely if I don't shut it down. Um, let me see. So I can switch back. Yeah, there we go.

Okay, so what we can do now, which is of course the last point on the slide, is get an agent a wallet. Um, which is a sort of a way in which you can make these AI models far more useful than they ever would be otherwise. Like it's all very nice that they can generate pictures for you, but I actually want my AI to make me money. Uh I want it to trade on my behalf if it's actually smarter than I am. And also just in a more practical sense, I wanted to be able to buy stuff for me.

If we get to a future where these AIs can take stuff out of your hands, like going shopping, I wanted to be able to pay. And I think building on crypto rails is very important for that because it's just natively intuitive to these agents. Far more intuitive than trying to get it to understand how to open a bank account using a traditional system. Click. Yeah.

Uh okay. I'll largely skip this because I don't know how much longer I have. But um for those like that use our service, you can use every type of model. Uh, I think in this example, the the guy with the poly market bot, he only uses text models, but he uses like 30 different of them because some have different specialties than others. Um, we also have other AI agents running on our service that are mostly Twitter interaction based.

So, they create images to go with posts that they do. They also use the the vision a lot because they can like analyze pictures that come by on Twitter. We run our own sort of agent which constantly scrapes Reddit. Uh it pulls in all the posts and it just checks them for relevancy to nano GPT. There's so much cool stuff that you can build with this.

Um and I think we are one of the few places where you can build with this and do it fully natively on crypto. Um because as I said this agent used ETH obviously uh Joel from Dash was here. We accept Dash as well. Uh the Zeno guys we would love to accept Zeno as well. We're trying to support every crypto that you can use for payments and get it u get it actually usable on our service.

Oh, all right. So, um I had no clue how long the demo would take and how good it would how well it would go. So, I think I'm actually ahead of schedule. If you want to try us out, like you can scan the QR code or just go to the website. Uh you can also talk to me.

I'll gladly give you some funds to try it out with if you're going to integrate us. Uh we have some integrations already with Cakewallet who are also here. We're very proud to be on Cakewallet. They were one of the first places we got integrated into. We're on bitcoin.

com uh in the wallet. Uh we have a ton of other integrations. If you are integrating us via API, we have a very easy API to work with. Uh and especially if you are building something like an agent that pays in crypto and that actually does cool stuff. We would love to hear about it if you want to shine a spotlight on it.

I know many don't because what they build is very profitable. The poly market guy was very hesitant at first to share this. Um but yeah, check us out if you want to talk to me and figure out what is possible to do with these agents nowadays. Um hit me up here. I'll be here all three days.

Or of course just uh contact us on Twitter, contact me on Twitter or uh in the Discord, anywhere else you can find us. So thank you everyone. Thank you for watching the demo and I hope to talk to you. Amazing, Milan. Amazing.

Very interesting to see that AI agent working on its own. So you made an AI agent build another AI agent. Yes. With I guess a teacher or like you know a main AI agent controlling the rest and then so who gets paid at the end of the day? Me.

Oh, you get paid. So essentially there's a few parties that get paid. We use a few crypto project. No, sorry. Essentially, there's a few parties that get paid uh because the open source models that we run, we usually just use providers for them.

In some cases, we use like Google for Gemini. Uh we also have some crypto backend providers. So, uh what's it called? There's Bit Tensor that we use, Akash, a few crypto backends because they're actually more reliable and also it's cool to build on crypto, we think. Um, so yeah, a lot of people get paid from what we do and hopefully the people building the agents also get paid at the end of the day.

Yeah, that's amazing. That's amazing. Now, another round of applause for Milan. Now, any questions by the way? Are there any questions?

Well, I have one question for you which popped up through slido. One person asked, "Is it only eat payment?" I saw that it wasn't only eat, but I will let you answer that question. Uh, yeah. So, I'll go back to the slide.

Oh, my slide's already gone, of course. Where did it go? Yeah, that's because we already Yeah, but No, no, no. I can I can bring it back. Okay, there you go.

Okay, so we accept pretty much every crypto you can think of, I think. Um, top right top left is Nano, which is the one we started with, one that I'm personally very enthusiastic about. Monero, Bitcoin, Bitcoin Cash, Salana, ETH. Um, yeah, a bunch of different stuff. Even Dogecoin, it's not used very much, but uh it came with BTC pay server, which is what we use for a lot of payments.

So, yeah, if if um like there's a crypto that we don't support and you do want to pay in it, we also have an inbuilt swap service that you can use for now, we're always adding more crypto. So, if there's a crypto you particularly want us to add, get in touch, tell us how to integrate it, and hopefully we can oblige. Yes. Another question that I have. Yeah.

Ask a question. Absolutely. Absolutely. Oh, wait. You can have that.

Yeah. Yes. Would you tell us more about the Reddit watching and scraping? Uh, there you go. Yeah.

So, uh, the Reddit scraping bot. So, I actually built this myself. Also, for context, I am not a developer. I I I literally never learned anything about development. I just use these AI tools to build stuff.

Um, so I figured, okay, there's a lot of people talking about stuff related to our service on Reddit, and I would love to be like, well, actually, have you checked out NanoGVT? you know. Um, so I built something that essentially just screen scrapes Reddit like a bunch of different Reddits. Uh, I gave a model access to a tool to do so. So like browser use essentially.

It just scrolls down the threads. It sees what people are talking about and it checks like is this relevant to nanoGPT? Um, where is this relevant to nanog? I gave it a short description of what we do. So it just checks Reddit.

Um, and if it sees a post that it thinks is relevant to us because people are asking about like using AI privately or using AI with crypto, then it pings us. So, it sends us this is the link. This is what people are talking about. Uh, this is a suggested reply. We don't have it auto replying because that's not allowed.

Um, but yeah, you can easily build all that kind of stuff which also makes me think like entire Twitter and Reddit must be filled with autorely bots at this point because if I can build this, many others can also build it. So, uh, yeah, that's the Reddit bot. Nice question. Uh, another question that I have for you is uh, on the compliance case. Is there anything to look forward there?

Is it allowed for agents to be able to transact with cryptocurrencies? Could you elaborate more or anything that you like face in that in that sense? Yeah, so when people use these models and when people build these agents to transact with them, we don't actually even see what they do, right? Everything is fully private. So it's up to people's own responsibility to check like is this legal in my jurisdiction.

Um I don't know to be honest in what jurisdictions it's legal and in which it's illegal. I'm going to guess if you can transact somewhere legally with crypto, it should also be fine to have an AI agent transact with crypto. But I'm also going to guess the poly market guy. I don't think this shows up on his tax returns. Uh so yeah, I I doubt that he's doing it fully legally.

But uh yeah. Oh, we have another response. It says integrate hyperlquid nano GBT. What do you say? What do you say to that?

Who said this? Who asked this? I think heck he wanted to stay anonymous, I guess. Yeah. Yeah.

Exactly. um the easier it is to integrate for us the more likely that we will do it. So uh we use for context we use BTC pay server we have a custom integration for nano we use uh helio for soulbased and we use daimo pay for ethereum based u so if you're on one of those it's very easy for us to integrate if not then like literally email us or contact me and tell us how we can integrate this in the easiest possible way um for every crypto that we support I would say we have a pretty good integration like subsecond confir information whenever possible. U so for Dash it's like instant pay. Um we try to make all the communities that we have like the coins of to make them proud of using it.

We know we're used for demos sometimes for other communities which is really cool. Uh so yeah tell us tell us how to integrate it. Brilliant. Um all right that give it up for Milan once again guys.