Thank you for coming to our seminar. So, my name is Nolan. I lead many of our product lines at Google Cloud related to Web3. We have been developing Web3 products and services since 2018, believe it or not, starting with BigQuery datasets. We'll tell you about some of the things we're building, but we have datasets for over 40 different blockchains.
We have faucets for many different test networks. And we are also working on creating agent access to these resources. But one of the things we focus on is agents, agent payments, and one of the main issues is trust between agents. We believe this is a big gap. So we'll dig into that.
Gotham is with me too. Greetings. Greetings to everyone. My name is Gotham. I work on the technical side on the Google Web3 team.
We help support many different services, different services, like BigQuery and faucets, and we're excited to tell you about ERC-804, which is a very interesting standard, especially in the area of agent commerce and its future. So, we're structuring today's meeting like this: we'll talk about the product and why trust is important between agents. How ERC-804 is a great solution for trust, and how you can access 804 data using BigQuery, since we have all of Ethereum data, from the genesis block to the end of the chain, indexed in BigQuery. Then we'll show you some real-time SQL queries and how you can retrieve this data. We will also discuss what a successful app might look like.
We are offering a cash prize of $5,000 for the winner. Agents are doing more and more these days, aren't they? We are giving them more and more autonomy. Now they can buy and interact with other agents. But it's very difficult to give more trust or autonomy to these agents if you don't know who you're dealing with.
I don't know about you, but I imagine a world where agents delegate tasks to other agents, who delegate them to still others, right? And real-time markets will be created "on the fly." These are the things. In such a world, it is important to know who you are working with, or at least have some idea of that agent's reputation. We have a lot of different standards and protocols that have emerged, like MCP— have emerged, like MCP— obviously, everyone uses MCP to communicate with different tools and services.
Google has been working on an agent interaction protocol that standardizes communication between them. But one of the things that's still missing is the issue of trust, right? Give me one second. I actually opened the wrong presentation, so I'll go back now for a minute. If you can, please take me off the screen for a moment.
Sorry. Give me a second. It will look very different, but it will be great. OK. Well, we can continue.
OK. Dark mode for everyone . OK. So, as I said, we have different ways of communicating between agents and services, and also between agents themselves. We know how to do it, but how do you actually trust an agent outside of your organization?
If your agent hires something else within your organization, that's not a problem. Your agents can communicate with others, you trust them by default. But if you are turning to third-party agents, you need an answer to this question. Therefore, the 804 standard bridges this gap by establishing three different components. So, identity— So, identity— who is this agent?
We'll get into this in depth, but at its core, it's about issuing an NFT to an agent, who you can then verify through reputation. This allows anyone to leave reviews about an agent, evaluate their quality, and assign a rating. And validation. This is still in development, but how do you prove that the agent did the work it promised, or at least get a ZK-proof or other verified way to track the agent's usage? We created the 804 standard in collaboration with MetaMask, the Ethereum Foundation, and Coinbase.
It has been running on the Ethereum mainnet since January 29th. And it is available on many other networks. If you go to the repository on GitHub, you will see all the networks where it is already running. So the first part of this is part of this is identity. How to manage this?
We have a mechanism where you trigger a registration event in this smart contract. And when you do that, you're essentially providing key data about your agent, which I'll go over on the second slide. In the background, an NFT is created NFT is created that is associated with your agent, and you receive an agent ID agent ID . You can then use this ID in the future to verify your identity. Of course, you can own this NFT.
So only the wallet that triggered this registration event will own this NFT, proving ownership of the agent. If we take a closer look at what this file looks like, it has a name, a description, and the most important part most important part is the services. These are the endpoints through which you access this agent. This can be your ADA endpoint if you have an MCP server on top of it. You can also specify whether it supports X42, whether it is active, and what trust models it supports, where reputation is primary.
The registration field is actually populated after registration, it updates the agent URI. But you don't submit it during your first registration. You provide all other data provide all other data . You get the agent ID and global ID from the registry. This is the first part of the identity establishment process.
The second part is reputation. There is a contract method called "provide feedback" where you pass an agent ID agent ID , a value pair, decimals, and tags. This is made public because the project is still in its early stages. We decided that we shouldn't limit users' choice of tags. You can use, for example, star rating, uptime, response time, etc.
When you leave feedback for an agent you worked with, others can see it because it is on the blockchain and public. So, this is a kind of construct to create a reputation system for your agents. We will leave plenty of time at the end for questions and answers, if you have any . This was the second part. The third part is still actively being worked on, but the idea is that you can request validation or verify this agent by providing an ID, URI, validator, and hash to prove that the agent did the work it was supposed to do, right?
So do, right? So , a third-party validator could confirm this could confirm this . And there are different ways to evaluate this. And so this is still a work in progress. This is not part of the specification.
There is no registry for this yet . If you have ideas, I think it would be great to work on them, although it's not exactly included in the, let's say , reward. But this is something that needs to be worked on and where we are looking for collaborators. So, if you have any serious ideas on how to implement this proof of work, let us know work, let us know . Here are the different smart .
Here are the different smart contracts on the Ethereum network for the ID registry and the reputation registry. These are two different smart two different smart contracts. You can view them on our prizes page, everything is linked there, as are the sample requests that we will now look at. So, our Ethereum dataset in BigQuery is available for you to use. It is available to you from the very genesis block to the last one at the moment.
It's like a public good. It's not exactly free, but we have a generous free tier free tier —1 terabyte per month —1 terabyte per month . Everyone who participates in this direction will receive a coupon for $1,000 to develop over the course of this weekend. It will be in effect for a few more weeks after that, maybe for a month. You can check with us at the stand, and I will tell you the exact time limits you the exact time limits .
Of course, we don't want you to spend a lot of money during the hackathon. That's why we're giving you a $1,000 coupon. Honestly, you probably won't need much since we provide a terabyte for free terabyte for free . This should be quite sufficient for most tasks. We have the full Ethereum dataset, all the data.
If you choose this track, for hackathon purposes you will only work with a portion of them. Namely with the data 804, which is why we indicated these smart we indicated these smart contracts. Um, I'll quickly explain how to do this because it's not always obvious. You need to create a Google Cloud ID or account. Um, this requires a payment ID.
So, you need to add a payment method, but we will provide you with a coupon that you activate via this link. And then you can go to our web3 go to our web3 portal and find our Ethereum dataset. And from there you can open BigQuery Studio and start making queries. Gotham will show you how to do it, literally in a few minutes. But we made accessing this 80004 data quite simple.
So, do you want to take the floor? Of course. Yes. Yes. Thank you, Dolan.
We have a few requests for demonstration. And Nolan, if possible, it would be nice to show this in BigQuery, right? So if you go to the project card, there is a link to the BigQuery product. And if we go there, oh, let's go to, yeah, let's go to the card. Yes.
And if you scroll down, you'll see a link to, yeah. BigQuery Ethereum dataset BigQuery Ethereum dataset . Here is the documentation showing what these fields are. So we'll move on to the Ethereum mainnet. Yes.
Set. These are internal URLs. Don't worry about them. You will see console.cloud.
google.com. And these are the different tables available to you. Yes. And if you want to know what those fields are, go to supported datasets, middle tab, well, for example, highlight tab, you want to just go here.
Oh yes, of course. Everything is fine. I'll come. Yes, if you want to know the documentation on what this means, you can click here to understand what these fields represent. Um, they are also in the diagram itself.
Exactly. Details. Oh, maybe. You are right. Perhaps the documentation is the best place.
Mhm. But let's say you wanted to filter. This query shows registrations per day. So, this is the number of agents who have registered themselves as agents using the ID registry. Right.
We'll just let it happen as an example. So now we have a row with the number of new agents that have been registered. Since this ERC was created back in January 2029, it is obvious that many new agents have registered here. Um, and we can see it here. Another good example to check is the second option, decoding the registration event.
So if we go here, and by the way, you don't have to take pictures of all these requests, they can be found in this Gist with comments explaining... from the prize page. From the prize page. Yes. But coming back here.
So, this option shows a string with the agent ID, the owner of that ID, and the agent URI. The agent URI can be a link to a file that specifies a directory, or, typically, Base64 if it is on-chain. So, yes, we can do it. I think it would be interesting to check Base64, just a random one. Yes, do it.
Yes. And another way to look at this is to look at this is to go to... contracts, right? We have one for...Oh, it's the ETH Global page.
Oh, we'll fix that. GitHub for ERC80004. Yes. And we can click on... can click on...
we have a lot of datasets, but the one we want to focus on is the Ethereum Mainnet for identification and registries, which is what Nolan was talking about. And you can see, for example, an identification contract. One thing you can do is click on the transaction hash, view the details, and even decode some of the input data. So maybe we could watch through a decoder. And see that this is the agent ID, and here is the string that is associated with it .
That's another way to look at it. Ahem. Let's see, another query that can be made is a reputation leaderboard. So I'll let you... This is another smart This is another smart contract that manages reputation.
So we'll see about that. Yes. So, we have the agent ID, the number of reviews associated with it, unique customers, and the average score for ranking purposes. Um, and the last query that would be useful to know is, uh, x42 for joining. So I'll just execute the raw one and click "run".
Yes. This will show us the agent ID associated with the agent URI and the average score attached to it . Um, but ideally also whether it's onchain, offchain and whether X42 is supported. So you can actually filter, um, whether it's true or not, whether the agent has x42 support. So if I need to send this, there are three values: null there are three values: null , true , and false .
Yes. So, um, yeah, these are pretty useful search queries, um, that we've saved here. If you need help with any of these, we're at the stand. You can come over and talk to us. We can walk you through this.
Um, you see, I think there are a few more things here. Um, a little more detail about the prize. Thank you, Gotham. Um, that's $5,000 cash, not Google Cloud credits or anything. You really get dollars.
Um, and we're looking for, you know, applications that use this dataset. Um, the ones that dataset. Um, the ones that , you know, use reputation data as well as identity, um, from a smart um, from a smart contract. So, pulling this data and some kind of frontend on top of it. It's not necessary, but It's not necessary, but , um, it's always nice to visualize things like that.
So maybe Explorer is one example; analytics, um, to see, you know um, to see, you know , what's happening over time; different ways to analyze and structure this data, um; and also, you know, more information about which ones support X42 and what they do. That 's right, and when you find out that it supports X42, you can run more follow-up queries and find out how many transactions they made and what the volumes were. Um, so these are just some ideas, but you can really build anything you want. Um, you know, I think, uh, I'll be there on Sunday morning as part of the jury. So I look forward to seeing what you build, and if you want to discuss ideas, uh, I'm at the booth.
Let's open up time for questions. Uh, let me check let me check if there's anything else here. There are many resources here. Um, I think most of them are linked from the prize page on the portal, uh, inside the ETH Global, you know, portal. So check it out, and, uh, if you have any questions, let me know.
I'm ready to answer questions right now. Yes. Please. Thanks for the development. Please.
What are your plans for monetization? I don't think there's a plan to monetize 80004, right? This should be an open registry. Um, we want agents to become popular. We want more agents to be created.
We assume you'll probably be using some form of Gemini if you 're building agents, or maybe hosting them on Google Cloud, right? So, that's our goal. Um, but there 's no direct monetization plan here. Uh -huh. The official datasets for Base and, for now datasets for Base and, for now , are correct.
I think there is. So, the question was the question was , are there official datasets for Base and BNB? Um, let me check GitHub 804. I know it was big there. Oh, I understand.
Um, I think we have Base as a dataset. Um, if you don't see it, let me know and I'll help you. Good. Hmm, I'm not sure about BNB. That's a good question.
But I think we have Base . Yes. What is 8004? So I think the payload that you register with is the name, the description, and the key points are the service endpoints, right? So if you tell me, "Oh, you're this agent alone," I can confirm that you're the owner of this agent because, um, that's part of the signup event, right?
The owner's wallet address and agent ID are issued agent ID are issued . Then I can find out what those service what those service URLs are, and I can make requests to them. Is that clear? OK. Well, let's talk about it later, maybe at the booth.
Yes. No, no one here has created a cool analytical explorer yet. So, this is work. That's what we pay $5,000 for. Yes.
So the idea is that if you interact with a registered agent—I think there's a dependency on them being registered with the ID registry—you can leave a review about them based on your experience so that others can benefit from it, right? Reputation now, for example, we have some examples that people use, like stars, latency, etc., but you can leave feedback on any aspect, we left that open. So, depending on what this agent does, it might be interesting to leave feedback on different aspects of it, so we intentionally left that open. Um, but we think that over time there will be certain things that people will use, and there will be more standardization around that.
Perhaps the standard itself will evolve to adapt to the main use cases. It's still very early days, it's hard to understand what exactly should be considered values. No, we do not separate assessments or criteria by vertical, but you can do so if you want. Um, if you want to classify and index all agents based on their descriptions, I think that would be great. Yeah, I guess, right now I don't know I guess, right now I don't know if there's a good answer to that.
Technically, anyone can leave a review. So I don't think that part has been resolved yet. I think this might lead to that third aspect that hasn't been clarified yet, which is —verifying what work was actually done, right? And then combining those two things can really give a stronger signal. Um, you can technically filter by unique IDs, so if you see that some member is just spamming, you can filter them out.
Yes. Any more questions? Part of our vision is to index agent behavior and the future of what's happening in AI. Is there any other behavior besides changes in the AI itself, or is that all? That's a good question.
Um, I think this spec was created with agents in mind, but you rightly point out that the realm of AI and reputation realm of AI and reputation probably goes beyond agents. So yes, maybe feel free to contribute to the spec. We know who to talk to. Any more questions? Perfectly.
Well, you know, Amelia's back there. She will work at the booth and help you get coupons. I can help you get coupons too. So come to us. Here is another question Here is another question .
Oh, yes. I'm just curious what you think about... I'm not sure I understand what you're talking about, I'm not familiar with this topic, so I'd rather not talk about it now, but I'd be happy to discuss it on the stand. Great, friends. Well, thank you for coming.
I hope you will participate and I look forward to seeing what you create.
Automatic transcript — names and jargon may be misspelled.