Remote Brainstorming

Summary

Project Concept

The team is developing a decentralized alternative to AI services like ChatGPT (Infernet). The platform will allow users to run open-source AI models (like DeepSeek and Llama) on their own computers or connect to nearby nodes running these models. This would eliminate dependence on centralized services and enable direct micropayments between users and model providers.

Technical Architecture

• Decentralized node system where users can run models or connect to others
• Implementation of X402 to manage per-token-usage payments
• Command-line interface (CLI) for configuration
• Possible web interface similar to ChatGPT/Claude Code
• Node reputation system based on reliability and price

Payment Mechanism Options

• Direct micropayments: Blockchain transactions per API call
• Prepaid system: Users deposit funds that get consumed over time
• Batch payments: Transaction processing at specific intervals (daily)
• Gas station: The protocol sponsors gas fees to reduce costs

Technical Concerns

• Gas cost could exceed transaction cost for small micropayments
• Smart contract optimization options to reduce gas costs
• Possible implementation of “Delta Accounting” to manage balances

Action Items

• Arturo will work on the terminal interface and CLI (tui-v1)
• Research and implement the smart contract by the weekend (node-registry-smart-contract)
• Antonio will research what types of projects have won prizes in similar hackathons (last-winners)
• Create and share an architecture diagram on Notion (project-architecture)
• Define the scenario for handling gas in transactions

Hackathon Goals

• Participate in a hackathon (possibly sponsored by Coinbase)
• Aim to win prizes that could add up to $10,000
• Set a minimum goal of $1,500($500 per person)
• Prepare a compelling project presentation

Notes

Transcript

Yeah, you’re right. I should have understood what I was doing. I mean, first I was going to hook up with another band. They were like, oh my god, that’s so cool, we have to go, let’s see if… I don’t know what, and we went. They got excited and said they were going, and at that moment I was like, yeah, of course, let’s go.

There we see that we build, that I don’t know what. I gave them a hard time. But then I regretted it. Because at the moment it was cool, but I thought it was going to be like… Well, it’s uncomfortable. I know you guys were going with a more relaxed plan and I wanted to study too. I was like, fuck, I think I screwed up. But it’s the other week and those guys didn’t tell me anything. I don’t think they went.

I hope you get it this week. Hey, that thing you put in there, is it talking to everyone? Yes. But it’s just the audio. It’s like a transcript. Something like that. Well, let us know. What do you say to Banderas? If they haven’t contacted you yet, who knows? Maybe they’ll contact you at the last minute. We’ll see what’s going to happen. I really like this plan. If that happens… Oh well. I’ll have to…

In a week I have this fucking exam that sucks, so let’s see what they say. But if it happens, we’ll see. If it doesn’t happen, which I think is the most probable, we’ll kick their asses with this project and if it does, I’ll leave them. You implement it and that’s it. I don’t think it’s a bad plan.

It was like, how can we make an AI more decentralized? Instead of everyone using the models from ChatGPT and all that stuff. There are open source models. I think you can download DeepSea and run it on a computer. That’s correct. There’s one called Gwen, right? Or Cuen, or something like that. I saw it’s Chinese and it’s really cool.

I don’t watch it, but yeah. There was a day when ChatGPT crashed like a month ago. Several APIs crashed. I think Cloud also crashed. And I saw a lot of memes on Twitter saying that I can’t work anymore, I don’t know how to program without IA. And it’s like a meme, but that could be avoided if you ran your own chat GPT on your computer. But not all computers can do that. For example, on the Mac I’m on right now.

Honestly, I don’t think the heaviest model can handle it, but what it could do is that, I don’t know, Artur, let’s say he’s going home right now. It starts running the model, and I imagine an interface where all the nodes close to it appear, and you can filter them, like, this one runs LAMA, this one runs DeepSeq, this one runs this, so you can see which models are running.

You choose from these models, and then you look for the node that has the best reputation among those who are running it. It’s the one that has dropped the least in the last month, I don’t know, whatever the metric is, or the cheapest one, whatever you want. It connects, and then it sends my requests to that guy.

And so I don’t have to pay Claudio or OpenAI anymore, but when I’m using it, maybe in a good code session, I get my 5 dollars. But for the rest of the month I didn’t do anything hardcore, and I don’t pay anyone anymore. And instead of giving my code to this OPENAI guy so he can train his IA…

Well, I’m running it in an open source model, dude. There’s not that kind of… And what open source models do we have for… There’s this DeepSeek, or the other one that Dio said was cool. And there’s another one called Lama. There’s several. We downloaded them, they’re pre-trained.

In theory, if I understand correctly, the idea is that they continue to train and that there is a network of many points to which you, as a user, can ask questions. For example, I would like to raise someone who is specialized in music. And I give them the idea of eating pure music things.

And I say, look, ask him about music and he will train, and then there is someone else who wants to do something very specific about DevOps or process optimization or whatever. And there we would not have to rely on a general model. It’s very well trained, but it’s also very general. In part, the band can have their own nodes, and you can choose who to ask for information, and blah blah blah. That’s what I do.

I have a question. The idea would be to have another UI for the terminal. It acts more or less like Cloudecode or Gemini CLI. I mean, what would it be doing? Pure question and answer. Well, maybe we don’t have to get into that right now. I have an idea for the architecture. I’m going to make a diagram and upload it right now that we have the call. But it would be cool to define what it’s going to be, if it’s a half-copy of that in terms of UI.

Or if we want him to do something else. To be honest, I imagined something else. Like, I was thinking like… Well, ok. Did you lift the knot? And how do I send him requests? So, on the one hand, we’re going to use the X402, and we’re going to say, ah, look, so that…

You, who are my node, answer requests, first you have to go through a paywall, that’s what x402 does. So in the header comes a hash. It says, here is my payment check, I’m sending you a request for 1000 tokens, so it’s 5 cents, whatever it is. So, since it already has that in the header, the middleware lets the request go to the server, the server does the compute and answers it, right? So, that part is already there.

But there’s still a lack of… well, most of us use an editor or chatgpt or something like that, and you’re not going to connect it like that. This is a voice memo. It has been edited to include proper punctuation. This is a voice memo. It has been edited to include proper punctuation.

Or this is my wallet and here I receive my payments, here I put that my node is on or off, so that terminal is like a gateway. And you can request it as an API or something like that. And then from your editor or something else, you send a request yourself.

To your port 69420 or something like that. And so, let’s say that this program is the one that is connected to the whole network. And you and in your settings you put that, ah wey. I like the cheaper ones. I don’t know, obviously it can’t be that bad. It has to be something very simple to start with, but in my mind it would look like that.

You open your damn terminal and you say, look, I’m going to run my model or I’m going to send requests. So I open this damn program, here I already configured my credentials so that I can send payments or receive money to lend my computer. And yes, you can use Apikit, because if you don’t, no one will be able to consume it. Yes, yes, yes. Ok, so the CLI is going to be more for the configuration.

I imagined that, but maybe we could do two views. One that is a chat GPT, similar to Cloudy Code, but maybe not agent. So it’s a chat, so we don’t get in trouble. So you can drag a picture of your cat and say, is it a dog or a cat? It could be web, it could be based on the terminal. Because you can drag a file and read it. I have no idea. Yeah, you can send him photos.

I didn’t know. Ok, tell me, tell me. I have a thought. Go ahead, say it.

Ok, first question. Is there a way to make contracts that you only have to sign once? I mean, they get signed every time there’s a transaction, but as a user you only have to sign once so that… For example, I’m thinking about the payment mechanism, right? Every time you make a request, well, to the server that manages payments, right? And then, well, I imagine there’s going to be a server that manages.

and it already brings a wallet number and the request itself, reads it, then gets into the blockchain and says, oh, there’s going to be a payment. Or you pay this person and the contract gets signed. And once it’s successful, the request goes to the AI and comes back, right? What I mean is, you probably don’t want to open MetaMask and click sign contract on every run, and they’re going to be micropayments, so are there contract types that don’t necessarily require signing each time, but rather a one-off and then a flow of money.

Yes, there are two options. I think we can do something called a server wallet, where you take a brand new wallet of which you don’t even know the twelve words, the private key, or anything. Your server just knows it and you have the keys to recover it. And you can name that server wallet something like AI credits so you know that address is where your credits are, and you can deposit five dollars into it, and in the configuration we can make it so that…

Since it already has the private key, and what pops up in MetaMask asking you to sign is really just composing an object with your private key to create a hash, and the other side can verify it cryptographically.

And that’s it. So really, the signing thing is for user control, but you can tell users: look, in this product you don’t have to worry about signing anything. You just need to know that every time you send a request, it’ll calculate based on the model and tokens how many cents you need to send, and it sends them.

So that can be done.

Alright, I see it. There are many things I don’t know how we’ll do. Well, I have to research them, but more than that. What about gas? If we’re going to do microtransactions and we get charged gas, there could be cases where gas exceeds the transaction cost. In that case I’d say…

it wouldn’t work like maybe the model where you deposit money into the service, I don’t know, 10 dollars or whatever, and we manage it as a general money pool and then at the end we make settlements. I don’t know how to manage that and I think you have better…

the chains with cheap gas, but even so if they’re small tokens maybe gas will exceed the transaction cost. Right, right. I’ve seen in Coinbase examples that they send… well, there are two options. One is let’s say the gas price…

Is a problem that other people are trying to solve. We could say look, we’re not getting into this detail… It could happen, right? If you send a “Hi, how are you?” to ChatGPT, maybe for the tokens and being a cheap model, you’re screwed because gas will cost the same as answering the question.

So we could say the gas price is something we’re not actively solving. We can look for gas optimizations so each signature is as cheap as possible, because cost is calculated by how complex the contract is. And there are optimization methods where now there are super cheap operations, certain exchanges, like Uniswap and such, that used to use a lot of gas and now use almost none. So sometimes they charge fractions of cents. It’s not negligible if you do it a thousand times because it adds up to a dollar, but I think it’s not that bad.

So we could do that, or we could do what you said, which would change the architecture so payments are made in batches, like every day at midnight UTC the payments go out, you charge those who used the model and pay out the providers.

But the problem with that is you need either a very sophisticated contract that manages all the states, or a solution like a database where you keep the whole ledger in, say, a Postgres, and at the end you just sum up and make one transaction. That’s something, I mean, I think I like both options. Declaring “hey, we’re going to take on this cost” solves a problem, but I also like having a transaction ledger.

But as prepaid, like OpenAI. You put money into the AI service manager, we consume it, and at the end we make the payment, right? But the user pays, I mean… I don’t know if that’s something people would… well, the product needs to be good on its own merits, but that would eliminate the gas problem because yes.

Well, even if you send a “hello” request, no worries, but I think that could work. Setting up a Postgres to log those things shouldn’t be too hard. We make a per-user account contract, or actually we’d already have all the money like a bank, and at the end of the day we make contracts per provider…

Yeah, that could also work. The option of doing it in the contract isn’t that far-fetched either. It’s a pattern called…

Damn, I forgot the name, but they call it Delta Accounting. What you do is make a table of addresses and just keep a balance of deltas, so plus one cent, minus one cent. The first time you interact with that contract you authorize it.

And from there it keeps your accounting as plus one, minus one, plus whatever, until you decide to terminate and that’s when it executes. I’ve seen that too. Never implemented it obviously, but we could find a pre-made one and base ours on it. And if not, we can do it with the database. I don’t think that’s bad either.

As you say, everyone puts in money and at the end of the day payouts happen. The only thing we’d need is that if you just deposited 5 dollars and immediately want to withdraw, it doesn’t work because we’d need to wait until midnight, check the database that you haven’t spent anything, and then ok, since you haven’t spent anything, we return it.

So it would be locked for a maximum of 23 hours and 59 minutes. We need to think about this because, I mean, I get it, but I’m not sure I have the best visibility on business and product finance topics. But I like both options and my only tech concern about using this contract is whether it’s heavier than a very simple one.

In terms of gas, because in that case I’d probably opt for the cheaper one. But anyway, I don’t know, we need to think about it. Is there a third option? Which is that we as a protocol sponsor the gas. Obviously that won’t be sustainable forever, but we can say…

We believe the business model could hurt us if gas ends up costing more than using it, so dear Coinbase, give us credits on your blockchain to be our gas station. That’s very common. You tell them: look, I could deploy this protocol on Base, or Polygon, or Arbitrum, and I’ll put it on whoever gives me the biggest gas reserve for my users. I won’t even keep it myself. You put it in a very transparent contract nobody can touch, and it’s called a gas station. The gas station works so that when transactions happen…

Instead of the user paying gas, the gas station pays on behalf of the users. So that’s another option and I think it would work too. Damn, I went super deep but what I wanted to tell you guys is don’t worry so much about the contract right now. I have a bit more knowledge about contracts and X402 now. So more than anything, my thing is just that we’re all three thinking about the idea in case we come up with cool stuff or ideas about how it could look or be used, we share them. And the rest we’ll solve along the way. But the contract topic and this transaction stuff…

I’ll get it done 100%. Maybe even by this weekend. Hey dude. No, sorry, sorry. Well, mostly I already have the… time has always been humble but… well, a bean broth. Also define the scenario for the gas topic, how that’s going to be handled. I think we’ll figure that out as we go, it’s not a big deal because it’s just adjusting a thing or two in the contract.

But the core logic is more or less already there. So dude, I’d say what’s important is understanding who has won those two prizes. Because the idea would be to aim for both prizes, and I think if it goes really well we could win, what do you think? 10k, right? I mean, 10k is our best-case scenario.

Yeah, I think so. Among 3 people? Yes. And our worst-case scenario should be that we don’t leave with less than $500 each, meaning winning a thousand and a half among the three. I think that should be our floor. So we need to check the prizes.

Well, I’ll drop an architecture diagram once I get home, and if you want, tomorrow or whenever we have time we can review it again. Yeah, put it on Notion. And Artur, honestly, I also wanted each person to do what they enjoy.

But I’m not that familiar with terminal tools and I know you are, dude. So if you take on the interface and if we can connect it to a Node server or something familiar. That’s exactly what I’m thinking. Before the call I was…

Downloading the things and seeing how to start the server and how this thing for the interface works. I more or less understand it, but I need to dedicate some hours. But from now until the weekend… Yeah, I’ll put in whatever hours I can and have a basic product that makes requests somewhere and looks good, and we go from there. The Node part isn’t that bad. I don’t think so.

That we can get done. Yeah, we’ve got more experience with that. Exactly. And Anto, dude, if you can handle… Really understanding the product, dude. Like understanding a bit… Technical details don’t matter. Just be like, oh it’s an alternative to OpenAI that runs on Ethereum. And start drafting the narrative and all that because if everything goes well, you’ll be the one doing the finalist presentation, dude.

I saw they posted the flight leaves at one, right? Yeah dude. At two something. One o’clock here, dude. Ah well, and what time do you need to get to the airport? Like 11. But you know what? Yeah. But we could use my Uber budget from LED Global for the ride there.

Who knows if it’ll be faster or slower, man? Good point. I don’t know, we should look into it, but yeah. Go ahead Antonio, dude. So to summarize… We need to be super fucking optimistic, dude, and really see ourselves winning everything.

Ok, ok, deal, deal. The real deal. In the flesh.

Alright, so Anto do that research dude, also about who has won prizes from these guys. Ok. To understand, because those SuperIntelligence guys are really weird dude, I feel like they don’t have much product, so…

If you can research who has won their prizes, we’ll get a good idea of what we need to sell them to win. Ok, agreed. And for Coinbase too, see what they like, but I know they’re interested because they’re betting on agents.

Agreed, I’ll be putting it on Notion. Camera, band. And well, the other goal is to have a great time, that’s the most important thing. Yeah, exactly. Be careful with those hotels. It’s like the one in Tuxpan, right? Yeah, yeah, exactly.

Take care, see you.

Brainstorm Meeting

Click the button below to add a new question block

Note

What alternative products we think solve this problem well?

Click Add an idea to automatically create a bullet point with your name.

• @Matt Piccolella -
• @Yiwen Zhan -
• @Sohrab Amin -

Note

What are the biggest problems facing our users today?

Click Add an idea to automatically create a bullet point with your name.

• @Matt Piccolella -
• @Yiwen Zhan -
• @Sohrab Amin -