From ChatGPT to ChainGPT: Shipping Autonomous Agents on Ethereum - Opeyemi Stephen

be calling our next speaker in person of Aquami Steven from G AI to speak to us on from Chad Gibbit to chain Gibbbit shipping autonomous agents on Ethereum. Please a round of applause for him as he comes on stage. Thank you very much.

My name is Aqui Steven. Um in the eyes and mouth of some people Bangkok um for some just Bangkok for others. Right. Nice to meet you. Say nice to meet you too.

Okay, now we we we we do have we do have the crowd. And today I'm going to be talking to you guys. Um from chat GPT to chain GPT shipping autonomous agents on Ethereum, right? But before that, let's I wanted to say let's get straight into the video, but you know, not Tik Tok. How does this move?

Cool. Um a little about me. Um obviously Aquam Steven Bangkok whatever. I'm global ecosystems lead at Gaia. Um Gaia is a decentralized artificial intelligence network infrastructure that enables you to ship your own code to ship your own AI agents um to spin nodes and to be able to inference your GPUs at any time and anywhere in the world.

Before that I was dev innovation lead or developer innovation lead at fuel labs. Uh previously I was TCM at consensus as well. Um work with um on projects with MetaMask and linear. Sorry, can I get water? My voice is a little horse.

And um yeah, I'm also CTO for Danzo Trade Africa. Um it's our startup here in Nigeria where the loan and remittance platform and we provide um loan and remittance services to smallcale enterprises and small businesses. I'm going to need that water, please. All right. Um yeah, at the end of the day, I'm a friendly neighborhood guy.

If you see me anywhere, feel free to say hello. Feel free to walk up to me. Yeah, that's that about that. Okay. Um so why decentralized artificial intelligence?

because we can't just go from oh um chat GPT to chain GPT without understanding whatever it is that we're doing right um I'll say it in a few words right chatbot talk right an agent act and what I mean by acting in this case is it's not just reasoning right but holding a wallet and transacting on chain if we're talking about autonomous AI agents um or shipping Ethereum agents right I'll show show you the minimal architecture or the aim of this is to show you the minimal architecture that makes this real and the exact steps that you can follow after this session to be able to build your own. We'll have a little um um starter code which you can scan and then build off on. I built like a template for you guys as well. So um just a precise reproducible playbook. So why decentralized artificial intelligence?

First of all, builders need control, right? Um, you don't want to build stuff and then you have no control over what you're building. Everybody complains these days about Google. Everybody complains these days even about Open AI, you know, um, as a platform. You know, I think there was something in the news the other time of, oh, um, all the secrets you tell GPT, the please and the thank you that it could be used, you know, in a legal case.

True or false? True or false? Am I talking to a crowd here, guys? True or false? Cool.

So, there was all of that talk and um builders need control in essence, right? They need clarity and they need the ability to be able to act on chain. So, first ownership and portability is one of the reasons um why you need decentralized AI. You should be able to run your inference, you know, where it makes sense, either on your own personal node or even if it's a rented one. and should be able to move without you writing your app.

Right. Secondly, trust and privacy. We have mentioned the legal case with or the the legal precedence Sam was talking about. Thank you very much. Sorry, just a second.

All right, cool. Yeah, it's better now. Let's get back to it. So, like I was saying, right, trust and privacy. You need to be able to audit the path from prompt to action and to keep sensitive data as a builder without or with actually with the explicit allow list that you have.

That's the second reason why you need decentralized AI. Now, third, onchain actionability. That's just big grammar for you should be able to do things on chain, right? Um, you should able to give your agent, whatever AI agent it is that you're creating, a bounded wallet and policies so that it can sign and it can call contracts safely. Control, transparency, and action.

If you're looking for three words of why you need decentralized um, artificial intelligence, control, transparency, and action, please. What did I say? Are we hungry? Please. is no you should have eaten before you came now you come to party and you are not going to eat so control transparency and action what did I say please that is what decentralized artificial intelligence unlocks now let's see the minimal architecture that makes this um practical so we're going to slow it down I don't know if you can see this I think it's it's visible enough we're going to slow it down now and define what runs where right how actions are gated And we'll give like a plain English version because a lot of the time people talk a lot about you know um oh the jagon is too much oh I don't understand it oh ah it's not clear what we aim to do here and what I've taken you know the personal miss my personal mission is to simplify everything as much as possible for you guys right so let's start with the first thing there you see agent runtime right agent runtime is simply your app or your service.

So this is like the minimal architecture of what you know your Ethereum agent should look like. You understand? Now your agent runtime is your app or your service. Now what does it do? It plans a step and chooses a tool.

Think oh if you want to say something like um the contract or what or the agent should read the price you understand or oh if the delta is more than 1%. then update all those your if and else statements all those your conditionals and everything that's where your agent um um runtime and your app or your service lives right what what does it do at the end of the day it outputs a structured intent so if I'm explaining to you like your five what your agent runtime simply means this is just the planner that is saying oh I think that we should do x here's why now how many of you have gone through like your normal GPT especially with GPT5 Five for instance, you see those thinking stages, right? Where it's telling you it's thinking and if you click it, you can see those different stages that that is taken. That is what your agent runtime is. Is he understandable?

Is he understandable? Please make my work easy. You understand? Please, is he understandable? Thank you very much.

Now, let's now go to the second path which is where I'm going to be shilling, right? Which is like your inferencing layer, your GIA nodes. You understand? This is the inference only node. And what does it do?

You can see it there, right? It converts prompts these prompts that you have put there to structured outputs. It does not hold your keys. It does not touch your chain. You can self-host or you can rent.

So I was with Hyperbolic at some time too. Hyperolic is also um decentralized artificial intelligence network. You know we inference GPUs. People could rent GPUs as against paying for this really really um you know $20. because not everybody has $20 to float a project or if you're like a big company that wants to scale but people could rent GPUs, right?

So that is where your GA node comes in or your inference layer comes in. It doesn't hold the keys, right? It doesn't touch your chain. You can self-host it or you can rent it. The end point is also portable.

The endpoint is portable. Now, if I was to explain this to you like you're five, right? what your GIA node or your inference layer simply is. It is the calculator or the translator that helps you the planner think but it cannot spend your money. You cannot do anything outside of what it is that you want it to do.

Understandable capish. Cool. Let's go to the third part now right which is your signer or your policy layer. So we've listed two parts already. Your agent runtime which I said is the app or the service.

your inference layer which you can plug in something like GIA to right I'm going to say plug in GIA because I mean that's what pays the bills right and then the third one is the signer or the policy layer now this is what we call the safety gate you understand um separate process or micros service right it holds your private key you can enforce whatever policy that that that you need before signing you can also allow a list of AI methods you can allow spend caps If you say, "Oh, um um when the price hits so so and so, as my as my agent, help me invest $20 or help me invest $10 or help me invest $5,000." Let's assume you are building like a trading agent or something. So, this is these are this this is where all of those instructions are going to live. You understand? All the safety gates that you need, even your rate limits as well, it lives on your ser or policy layer.

You understand? If the intent violates the policy that you have set, what is going to happen? It's going to reject you know that task and ask that oh let a human come and review this particular action. Is that clear? So if I'm explaining to you like like like your five this model would not sign anything unless you ask it to or unless the signer itself does that.

You understand? This is the adult with the wallet. If I'm explaining to you like you're five. So we've talked about the planner saying oh I think we should do this and this right we've also talked about the calculator or translator that will help the planner think now this signer or policy layer it is the adult with the actual wallet money spending maintain this is what we have here do you get me now the planner can ask but it's the adults that will now say oh okay yeah this applies let us spend so so and so or let us sign so so and so transaction is it clear is it there. Now, the fourth thing you have is your RPC provider.

Um, if you've I mean, if you're an alumni of Web3 Bridge or you work with Web3 or you are a builder, you would know what RPC means, right? Who's a builder here? Who's a builder here? Who's a builder here? So, for for 10,000 NRA, right?

Let's do quick giveaway. I like my talks to be interactive. What does RPC mean? Who knows? Oh, now all of you are raising up your hands, are we?

But no, let's let's do it. Who knows what RPC means? Yeah, big man. Sorry. Can you guys hear him?

What did he say? Right or wrong? Right or wrong? You just won yourself 10,000. Please see me after the talk.

Yeah, this is where you clap. By the way, even Tobu does not give you money for free now. All right, let's get back to it. Right. So for your your for for the number four part, you have your RPC provider, right?

Could be standard HTTPS. So when you send that signed transaction, you know, to maybe separia, right? You can configure your primary and your fallback RPCs. Nothing should really be custom, you know, on this layer. So again putting it like you're five right this is the road or the post office you understand that carries out is the road or the post office that carries out your already signed envelope so let's assume you are sending a letter now you've written your letter you've said I love you to that babe in for those that went to boarding school in my time right you've said I love you to that babe right and you said oh let's help you deliver the letter obviously there was no social media then I'm not that old please right but This is the one that now carries it all of things that you need and now you are putting your transaction on the blockchain whether it's signing whether it's making investments whether it's sending tokens right now after that the the fifth layer is now the contract or the minimal rights method.

Now you can call things like maybe you write a method like your update value method right this is going to emit an event and then when you emit that event you can verify by reading storage or listening for that event I mean everybody who um has run one or two integrations before knows what it means to call an endpoint if you're like a front-end guy I think even backend guys um backend guys know it right so this particular contract layer is like the vending machine you understand um it's the right sign coin and the button will give a predictable result because you've written the rules and you've told it what to do. So when you call the endpoint you expect something either success 200 or what error 404 or 500 error whatever whatever um for forget I forget those codes now but you know what you expect at the end of the day it's either successful or it rejects and tells you the reason why with your error message. So if you're going to put this flow in six bits right you have your user triggers and then from there it goes to your agent plans from there it goes GIA would return the intent via the inference layer right your policy would validate right your signer will build or it will sign or it will send whatever you need it to send right and from there you can now log your transaction hash and then you can confirm the block you know on your explorer that's how that is supposed to work so from idea to action. All you you are going to need to do in these five stages. All of all of what this encompasses, right?

Number one is to think. That is what the runtime will do, right? And then you decide. That is what the node will do. The inference layer is going to do.

And then you check your rules. That is what your ser layer is going to do. Check the rules. And then you pay or you sign or you do whatever it is that you need to do. And then your machine would update.

and then you keep the receipts and you can check it on whatever um explorer that you're that you're using. Right? So um there's something that you need to also know which is observability. You have to the observ observability will happen on the explorer side of things. You have to keep that paper trail which is what essentially the blockchain helps you to do.

So you make sure that you keep that um paper trail and maybe log like what we call the correlation ID, right? You lock that ID, you log probably the transaction hash, you log the policy decision so that should you need to check, you know, um um later for discrepancies, you can be able to do that without any problems. Um I think I've bored you guys a little bit. So let's go um to the next slide. So this is the agent life cycle.

This is how it works, right? So from trigger to plan to think to check to sign and send to verify, right? So trigger is your user action your chron or your web hook the plan the agent to propose the intent what are we doing why are we doing what we're doing right and then you think your GIA or your inference layer it could be I'm not going to shield other so it has to be GIA so your GIA node is going to return um that structured output and then you check you know with your policies and your safety measures that you have put in place you understand your allow list your spend caps your rate limits you sign and send wallet signs RPC broadcast and then you verify. You can log your transaction hash and then you can check whatever it is that you that you want to check. Please, is this clear?

Do I have the crowd or have I lost the crowd? Is this clear? Thank you very much. So now um so a quick part to shipping. Yeah, because we do not have um a lot of time.

I made a starter repo and you can use this QR code if you want to experiment on this actual project, right? You open the repo, you scan the repo, you set your environment keys, right? And then you can run um locally. I also put um scripts, right? I also put some some um demo scripts and some safety scripts for you so that you don't um end up going going going the wrong way.

Um so obviously you open the repo um and that just means that you're picking the toolbox, right? Um Oh, sorry, my bad. So, obviously open the repo, right? Sorry about that. You open the repo.

Um, and then you set your three environment keys so that the script would know that you are on the test network, right? It's a throwway wallet, you know, and the contract address, right? You label the jars so that nothing gets mixed up. And then now you now install. You install by running locally um obviously npm install.

If you're a builder, you should know this by now, right? MPM install. And then you you you also install the demo agent scripts that I put there for you. This would give you that environment for you to be able to play around, right? Um for the sake of time, we're not going to be able to go into like the code code.

But that was why I took the time to explain how these things work so that when you are on your local machine, when you're looking at the repo and then it also has a comprehensive readme that will take you step by step through all of the things that you need to do as well, right? Um you can leave a GitHub star if you are if you are impressed with it. And also you can also contact me on Twitter or whatever if you have um problems or office hours. I'm down totally down for office hours at any time of the day. I think my voice is going again.

Um so yeah um you load this up shows you what you need to do your chain starter. Um so let me let me explain some of the things that are there real quick then I close. Right. So your chain starter agent um what what you see there is your your chain starter agent which is the folder right this is just the blockchain agent start kit that I have created it to demonstrate how AI agents can safely interact with Ethereum smart contract right through a policy layer that's what the that's what the um the thing is um also what does it do it takes you from AI agent to policy gate to Ethereum transaction right it will show you the complete flow from AI decision making to onchain execution so when you get into that particular repo. I have listed all of those things, right?

I I don't I think I should have pulled it up anyways. My bad. But if you get into that repo, you would see all of it. Your chain starter kit is there. You check what it does for you and you'd see the complete flow from like AI decision making to like your onchain um execution.

Now, the key component of what you are going to find there on in in that in that particular folder, the key content you're going to find there is your agent oracle, your agent oracle um um contract. This is a simple article that would store a single value for you. It has an update value method, right? Uh and or a function. You can call a function as well that AI agents can call and it would emit events for transparency as well because you're working with the blockchain, right?

Um another key component that is there is your AI integration which is the demo agent.js that you see there. Um the second one that you have to the script that you have to install this is going to connect your AI endpoints like GIA for decision making. It can also fall back to mock data if you don't have any AI service um configured or integrated. That's what um the demo agent.

js script is for. It will also demonstrate how AI agents can propose blockchain actions for you. Right? Policy layer as well um which is the third one. It contains your function allow list that is going to only permit specific contract functions, right?

It also gives you like your gas price caps in case you are you're trying to um um spend spend gas and and stuff. And then you also have like your blockchain execution layer which is going to sign and send transactions to Sepia test net. Seleia money is not real money please. So you need to keep that in mind, right? Um it uses ethas.

js JS as well uh for modern Ethereum interaction and it also provides your transaction verification and status reporting if you need to um log incident reports and all of that. So um you might ask yourself like why does all of this matter right? First things first is safety, right? Policy layer would prevent your AI from making dangerous transactions. You understand?

Um scalability as well, right? If there are no patterns, you cannot have real defy protocol or smart contract um um doing things autonomously for you. So you have to have that policy layer. If you if you are familiar with um liquid, I don't know who knows unju here. Unjuv, if you're familiar with unju's products, liquid, right?

It has that that policy layer. if you go into like the nitty-gritty or if you have spoken to him about the product. So this is important because you need those policies to be able to put safety gates so that your AI does not um run riot and do you know what it is not supposed to do for you or then start hallucinating and stuff like that. Also transparency is another reason why this matters right all AI decisions that you are going to make with this particular project is obviously going to be stored on chain. So it is transparent for all to see and it maintains that blockchain ethos decentralization and transparency.

All right. Um dynamic parameters adjustments too for defy protocols. You should be able to change things as things go. You should not just have a point A to point B AI agent whereby you cannot stop in the middle and say oh let's do this or let's do this or let's add this. You understand?

Your AI rails should as much as it can run auto autonomously, it should also be able to say, "Oh, I hand over the reins to a human." You should also be able to take control, you know, at any point in in that particular user journey, right? Understood or not? Cool. Um, so this is not just a demo.

It's something that can help you um if you use this um starter agent kit, it can help you go to production ready patterns for building your own artificial intelligence agents that can safely interact with blockchain protocols. So from this you can spin up an agent that can interact with Solana. You can spin up an agent that can interact with Cosmos. You can spin up an agent that can interact with I mean there's like a billion and one protocols these days, right? So um you could use this to I mean get started and get building.

All right. Um so this this would position you also. It's going to give you that nice journey. It's going to give you that nice experience of getting your hands in the weeds with artificial intelligence. Right?

So you should use this. The readmi is going to explain it to you um properly. Anyways um so at the end of the day right um this is the ethos of decentralized artificial intelligence. I cannot finish my talk without letting you know that you should own your intelligence. Right?

You should start today. You should start today. Own your intelligence. People shouldn't be the ones in control of whatever you do with your data. It is very very important that you do so.

Right? Um and in the words I mean thank you. You can follow me here right? Let's build autonomous contain agents. Uh my handle on Twitter is deev VGBG but if you just run a quick search Bangkok um is the name don't ask me why Bangkok anyways and um in the words I'll leave you with the words of a great philosopher um so you know that we have a lot of philosophers right and um to encourage people it's also it's always nice to leave motivational quotes true of false true of us so in the words of the great philosopher right um where's this what happened.

Oh my days, I don't see it. But let me let me leave you with the words of this great philosopher, right? Um his name is Crown Usuzama. How many of you guys know Crown Usama? He's also popularly known as Charlie Poppy.

So in the words of the great philosopher Crown Usama aka Shelly Poppy, all this my three boys

start today, keep building, you know, pastor. Thank you very much.