Trustless Agents (ERC-8004) Call #1

Transcript

Vision for a Decentralized Internet: Discussing open, permissionless web

The web was meant to be open and distributed, but over the decades uh intermediaries became gatekeepers of attention and trust. And now that we're building the new internet, the internet of agents, I think we have the opportunity also the responsibility to uh not to make the same mistakes and to create economies where information, reputation, market signals are common good and public and anyone can create, discover and verify agents without any company in the middle deciding who deserve trust and attention. And this will let communities express their full potential. It will let startups grow. I think it will also help democracy staying healthier. That's why we wrote

804. That's why I think we are all today. And so it's not just about a technical specification is really about converging all together on the same credibly neutral infra based on Ethereum of course and create an open permissionless internet of agents. So welcome everybody. I am Marco. This is the first 804 community call and I will leave Mick to my co-author Davidid who has been by the way just last week been appointed like head of AI at Ethereum Foundation. Congrats David but Mick is yours.

>> Great to see everyone. Thanks Marco for the inspiring intro and Marco is actually the one who chose the theme song as well. Um let me present. Um [Music] Just checking. Can everyone see my screen?

>> Yes. So we're quickly uh going to introduce the team and then um we'll discuss what are the next steps for uh 804 and I think that's why many people are here as well like uh uh with this first goal we want to bring together uh the entire community start working together um we have builder program announcement and then um Marco will actually like dive uh on what's happening uh on the spec and then we have a few demos and Q&As.

Yeah. Uh myself and Marco already introduced. So we have Jesse. Uh she's uh coordinating like um efforts around 804. Um she recently joined uh the DI team. I'm super excited like um about her joining like she's already like uh very familiar with like many teams in the ecosystem um being in this role before. Uh Issha um she works on the app support uh team at the EF and she's also joining our team. Uh and then Leonard is um one of the main expert in in engineering is currently a consensus. So

we assembled this team like basically from uh myself and Maros um network to kind of really focus on 804 which is the priority for like both of our teams right now. just quickly on what happened right so uh we launched ERC 2004 uh trustless agents um on August 15 and uh it's unleashed tremendous energy in the community many project uh came together around these new standards I was just talking to Marco before uh the start of the call and like we are looking that there were like almost 10,000 unsolited me mentions of 804 like uh uh on socials

like uh after launch um uh we had like eight independent implementation of the very first version of the spec uh so many teams reach out to build or publicly endorsed saying that hey this is a great idea we want to build on this and many of those teams are here today which we're excited about and then like uh um it's been like the number one conversation on Ethereum magicians in terms of protocol uh upgrades uh in the past here every week. So many comments and actually we're really grateful for these uh because most of them informed the next version um of uh the spec that we are working on and we will discuss today.

So why now like um just a reflection on this right? So there is standards on like agent to agent and then agent to tools like um really like emerging and getting traction. Um for example MCP marketplace and registries um are popping up everywhere. They have one problem that they are centralized and then now there is like this agent to agent protocols. There is another framework um called Agentic um essentially like trying to establish the base protocol for like uh the internet of agents like communication between agents. uh but there is uh a lot of trust assumptions baked inside at

least before uh this proposal before like we also worked on like adding X42 um uh payments to this and then at the same time there's many project in the Ethereum ecosystem that are launching um uh agents uh protocol for agents uh um it seems though that everyone has their own framework uh which leads to from fragmentation uh silos like um uh it doesn't seem like um an outcome that uh can really lead to like um bootstrapping exploring this um um agentic economy on Ethereum.

Introduction to 804 Standard: Foundation for agent economy on Ethereum

So what uh 804 does it sets the foundation for this agent economy is a credibly neutral standard. Um it enables discovery and trust. Um it um also like um essentially like removes the silos, right? Like uh now you can have composibility between um all agents, not just by agents deployed on a single protocol or a single marketplace. Um and then it's extensible of chain, right? So the way we're thinking about you probably seen like the proposal is very lightweight and then there's going to be like um a corollery of like infrastructure around it um that will uh

really enable functionality and like a lot of the focus there is on ux and also like what Leonard has been focused on uh with the team and if you're not speaking can you please mute So briefly like um uh um what uh4 does there is three registries uh identity, reputation and validation. Um and these are like essentially like entry points and they provide like uh the relevant information. We just see it as a very like strong but very lean foundation uh layer credibly neutral as

I said and then around these uh essentially like there will be agent developers that can like uh register agents, application developers uh validation services uh which will come as like uh um these um agents start being uh verified and validated and then Um this information can be used uh for a variety of services. Explorers uh marketplaces um rating agencies could pop up right in the sense that even the reputation registry just provides like the baseline um for uh recording like um um trust scores. But uh this probably will need

to be processed like different economies will pop up. And then um one thing I'm excited about in the future is that of course like we have financial rails. So uh this will be also like um a lot of application in the info or like agent file space. Um yeah and um I won't spend much time like talking about like um type of types of agents like Marco has a post on it. We can repost it in the channel if people like um want to go and see it.

Suram a friend um also has an interesting post on it and uh I guess um the idea is that um in the month of October we'll be working with a lot of like um uh builders in the community to like um um deploy new agents, new uh interaction between agents. Uh so hopefully like we will uncover like a lot of these together. Um and in terms of what's next um there is um we're going to do a workshop at the virtual hackathon next week. Um but this is like um kind of an intermediate step. The main milestone is

on October 3 uh 3rd. Um we release like the new version of the 804 spec. There is a few changes uh but uh we think that this can be the stable version where then like uh we feel confident that okay people can start like building on this like deploying more production use cases on this um and then uh October like focus on building arathons workshops and then um the other main um event is uh on November 21st uh this is the Friday of the dev connect week uh we'll host the trustless agent day in Buenosirus and then uh we will invite like a lot of

Event Planning and Trustless Agent Day

like the teams deploying agents on 8,04 as well as like uh researchers in in both AI and intersection of blockchain AI uh to essentially like look at where the state of 2004 is and uh decide together like uh the path forward what is Okay. So on the builders program, let me do a quick intro. Hi guys, I'm Isa. I'm on the app relations and research team at the EF. My role is basically to support all the teams with anything that

they need to succeed in the ecosystem. Be it visibility, be it connections, anything ad hoc. So when I saw all like the discussions and the momentum that was going on in the 80 for builders community I thought that this is like a great place where we can support teams and getting to the next stage. So with the team we thought like let's take in like the momentum let's take in the discussion and let's show the world what 804 can do right and so with dev connect as David said with dev connect happening in November on the 21st we are hosting trustless agents day and I know between when the proposal came out and now to dev connect it's not enough time so to

accelerate the progression of the demos with the builder program We want to provide you guys with all kinds of support. So with David and Leonard and Marco helping you with technical feedback and any technical support that you guys need. So answering your questions, getting on calls with you with me and Jesse on the GTM side of things and to make any intros with infra providers with L2s, anything else. So I would say that's like the entire builder path and again like anything ad hoc, right? because it's not like a onesizefits all thing at this point to get you guys to the demo day. We can always like hop on a call and discuss what we can do. So yeah, I think that's about it. What we're looking for is

again like there's no specific criteria as to like what kind of app should come out of it. But it's anything that makes sense. Anything that shows what 804 like looks like in practice and you know new use cases or again like just like you know if you want help in brainstorming ideas with us we are here to support too. So we've shared the form we've shared the details with you. So have a look at it and if you have any questions let us know and yeah looking forward to see all the applications and the demos when they happen. really excited to see how we take this to the next wave. Thank you. That's it for me.

Builder Program Support and Technical Assistance

Oh, and also like Leonard and Jesse, do you guys have anything to add here? Sorry. >> No, very glad to meet everyone here. Um, definitely do sign up for the good program. Um, we'll try to support you in any way we can. Um, I work mainly at tech side of things. So, if you any questions about, you know, the spec, um, things, suggestions for things to build, if you want a technical review, or if you're working on, um, things that are a bit more technical, like for example, TE, I can help out. Um, but very excited to see what ERA's going to build.

>> Jesse, don't be shy. >> Yeah, awesome. Nice to meet everyone again. Um, yeah, so very excited about the builder program. This is our way to focus our support to make sure that we have something that's demo ready, production ready, usable um by Defconnect. So hopefully we'll get to see you guys there. Awesome. I think we can go on to the next slide with uh the battle of the two logos or sorry not yet >> no we have the boring part of the call but sorry guys it's a protocol call so

we need to talk about the protocol and I will try to keep it short um so first of all really thank you I I don't know I've talked to so many people onetoone in group messages it has been like a a pretty tough month I have to say but just to say that everything that we are discussing here is the result of discussion with you your request, your excitement. So, thank you for everything. Um, we don't have like radical changes from a purse perspective, but we do have I think uh probably important changes in terms of how we achieve that to have better efficiency, performance and probably the most important thing where we have worked on is making the protocol um frictionless or with less friction than

before. I will briefly go to the list then we can bring the conversation in private when we will share the draft or also in the public group of course so we will support also MCP that's very important one common request was we love HA but why not MCP which is well very popular as well second um we would like to have clients expressing feedback without going or through the full registration process or removing friction there because we don't we the the the easier it is for clients to vote, the better it is. Third, for the identity registry, which is just one of the three registry, the data structure, the ID, the domain, and the address is exactly the same of a 721.

And so we will probably leverage uh 721, the URI storage extension, for different reasons. We like to use standards. It's more elegant to use standards. And this will give for free what many of you have asked which is the possibility to delegate the management of an agent to other addresses with a so-called operators. This is given for free with 721. And another important thing is that this will accelerate a lot the development of that specific registry because of course there is an amazing open Zeppelin implementation for that uh reputation. Uh many of you have asked to save optionally more information on chain instead of just the pointers to the offchain file to have better onchain

composibility. This will enable smart stuff like calculating average of reputation directly on chain. So that's for sure a reput a direction that we we will take as optional and we will also publish the um response of the validation not just the request with a commitment on chain. We also have conventions. So these are not part of the specs explicitly but it's very important to have help developers implementing that. So we'll use more IPFS because IPFS with onchain data enables you to do subgraph and so you have for free indexing which is great for having a better experience because it the page just loads faster and same for X42. will provide examples on how to

use X42 payments, proof of payments as a as an attribute in uh 804 at the station and also we will clarify how to use that with VNS as many of you have have asked and we will try to to to offer also good decks again this is a convention is not part of the specs on the different ways to implement these and stay secure inference. This last point will not arrive on the third but later and this is done in collaboration with the manyity and stake securing projects that are part of a community. So it's really run by you. Um we will suggest to run indexes as I said it will be easier with subgraph probably and we will also suggest to sponsor transaction. No need to manage this at the protocol level

because thanks to pectra as you know we have 7702. So at the application level anybody will be able to develop application where clients can give feedback without gas and maybe without registering. So super easy. Uh we would like to go on testn net around midocctober. This will enable you to accelerate your development for dev connect. We don't have a specific date for mainet yet. We for sure want to be also on loose. Uh we have many loos that are contributing to the community. So of course we will not have just one L2 uh but that's the direction it will happen before dev connect. Um if you have any feedback you can uh text me you can text in the group as soon as we have a a more

Testnet Launch Plans

structured updated document we will share it. Thank you uh for everything. The mick goes to Jesse. Oh we have a logo moment. We just received logos. >> Yeah. So, uh, our team, um, from the EF created two logo versions that we kind of like we wanted to have the community vote on this. Feel free to, you know, submit a number in the comment. Yeah, perfect. Oh, everyone likes number. Okay, I guess the results are almost very clear.

>> We don't need to count. doing a sibling attack and passing multiple shoes. >> It seemed unanimous before the civil attack. Yeah. >> Awesome. Yeah, we will also um post a final uh logo version in the group chat and turn this official. Cool. And I think next up we have demos.

All right, the first demo is Practis. To play the video and practice will be here to answer any questions at the end. Can everyone hear the audio? I don't think I can hear. Oh, okay. Yeah, it doesn't seem like people can hear the audio. Did you share the screen with a share audio option?

>> Let me see. Maybe I need to turn on my audio. >> Hey, practice community. In this demo, we're going to explore how to register agents in the 804 ecosystem. You'll see how agents orchestrate task across a decentralized network. So we're going to launch our telegram post agent from the practice interface. So we can watch as the orchestrator registers the tool and prepares the workflow. So the L&M driven orchestrator automatically generates a workflow plan. The steps will appear on screen and as you can see no manual coding is required and in the logs we see the PTP host start the tool register the workflows and these details reveal the agents life cycle.

Moments later the agent post to our telegram channel and the UI shows a message appearing and marks the steps as complete. And here's the agent's configuration. It's named P2P host and registered information is all showing demonstrating how it all works. So next we summon our ex summarizer agent and ask it to recap the latest CZ Binance tweets. The orchestrator writes a request to the correct agent. We wait as the agent processes the summarization request. The progress bar and logs update to show the script is running.

And now the X agent returns a concise summary alongside with the link. Now we open Twitter to verify the summary matches the actual post. And now we use the EOC 804 registration tool.

Some information will be hidden for privacy purposes. We enter the agency ID, P ID, and capabilities. The agent's card is published to the registry and it also appears in the explorer. After publishing, the agent appears in the agent explorer.

PRAIS Demo: Agent Registration and Workflow

His card follows the ESC 804 standard, making them discoverable and interoperable. This is how practice works end to end. Thank you for watching. >> Awesome. And practice, if you guys are on the call, feel free to also talk about it or if anyone has questions, feel free to ask in the group chat. >> Yeah, for sure. I'm not sure if my camera is working correctly there, but hi guys. Uh, great to be on the call with you here today. My name is Robert Brighton. I'm the co-founder and CEO for PRAIS. Um, not to go into a deep pitch or introduction here, but I think that

our mission very uh neatly ties into what Marco said in the beginning. We're looking to create a world where AI is owned by the people and not the corporations and uh the tech that we're developing uh neatly sits on top of 804. So, we've already been engaging with some of the other builders in the community and we're very excited for what's ahead. Um, we have some from the technical team on the call here today as well. So, please feel free to ask any questions around the tech that we're building or which way we're heading with the the organization.

Awesome. And I think we have um thank you so much um Prais and Robert and the team. Um, we also have uh Fala next. Uh I think Fala is able to potentially do a demo live. So feel free to share your screen if you can. >> Yeah. Um let me share my window right here. Okay. So, um, so what we've been working on is a code review agent, uh, because, um, there's a lot of people who don't trust how they, uh, vibe code. So, they want an agent that vibe codes for them. So uh it's

just the basis of this idea but um when we were working on it we were building out the um uh idea of an identity registry the validator um registry and the reputation registry but we were also adding a TE verifier as well to um to basically verify what the code was doing in real time. So when we were registering the agent um we tried to set up a full A to A protocol um and just to explain kind of what we're what we did is uh the A to B to C server this is the A to A process um you know these are various APIs but they're various validators that were registered on chain

using the new agent function um when we discover agents we wanted to be able to discover uh agents from the registry. So validators so that we can bring them in um in in order so that we could use those agents and they might just be APIs for example. Um and so API A, B and C uh might just power the code and do the audit and the review. One validates and verifies the quality and then the other one uh returns the reputation back to the uh uh reputation registry. Um, and then when uh when we're running this application, it was quite fun to see how it all connected. So, um, this doesn't have like a professional pay master yet.

Uh, and, uh, just so you know, but here's the general idea. Uh, you can start a code review. Um, in the code review process, effectively just asks you to start your MetaMask. Your MetaMask comes back with uh, a transaction. um the transaction comes back and says that the uh actual local host is doing this. But with local host, what happens with TEES is that the actual application domain will be uh port forwarded to the actual MetaMask or to the transaction that you're signing.

So the TE could effectively prove um from the operating system level through a zero trust domain that the uh domain that you're using is actually the thing that you're signing. So it's effectively like going online onto the internet and and sort of seeing this artifact uh this domain almost uh that has its like own life that's living breathing thing. So um we can confirm that transaction and we'll send the transaction and we'll create an A2A session. Uh it'll build out the blockchain uh transaction and we'll wait. And we can see that um we have a pretty low trust score. So, we're going to try to increase the score by just doing the agent validation. Um,

Trustless Agent Deployment: Discussing self-funded off-chain agents

it's going to be talking to that API. So, what's what's nice about the TE's environment is that you can instantiate the environment with a private key uh at the time that it was booted up. So, uh, what we've been experimenting with here at Fala is using our CBMs to, uh, deploy an agent that has its own key and then being able to fund the key so that it can be a self-funded trustless completely off-chain agent. Um, and so when we check these transactions, uh, I deployed this to the base Sophia chain.

So, uh, you guys are more than welcome to use my contracts for test net if you want to because it's only one registry. So, um, uh, yeah, and, uh, if anyone wants those contracts, I'll show you. But basically, uh, at the end of the day, I can increase my code score. So, I'm basically using my own APIs to my own advantage. But then what's nice is that you can't see my private key. You can't also see my API keys. So, you get to use my application, my APIs, but you get to pay me for it potentially, um, to service you for any particular purpose.

like said um like said trustless code review. So uh that's how we approach the the demo and the problem and um hopefully that gives you some insight onto sort of how the uh whole 8004 process works. Um if there's no questions uh you could see that I can download the code and you could actually uh take a look at it yourself if anyone's curious. But uh I'll leave it there just so that I can save on time and thank you everybody for listening. Thank you.

>> Awesome. >> Yay. Okay. And up next, um, we have, uh, Chaos Chain as well. And, and if you have any questions for Fala and Practis, yeah, feel free to send in the comments. >> Hello everyone. Uh, so I will also give a live demo. I can't share my screen. You should um Okay. Yeah. Okay.

And maybe something to celebrate. Samit is the first person that realized like an 804 implementation like one day that the draft was out. So absolute record. uh thanks for uh having me involved since the uh since the more or less uh the beginning of the release of uh of the implement uh sorry of the EIP uh okay I had to give grant permission uh to okay here we go I I thought I have to restart uh the zoom yeah so uh hello everyone so with koshin like we are building an accountability protocol for

Accountability Protocol: Introducing triple verification for agents

the agent economy and what that means is like we need to have uh a triple verification uh stack so that like we verify everything that happens from human interacting with agent. So that's where like Google's new AP2 comes in right like to uh cryptographically uh verify the intent mandate uh and and then you you need to have a verifiable execution where like agent runs its logic its business logic so on so forth and then uh you need to assess and value the agent's output. Now uh of course like 8004 provides the fundamentals right with the identity reputation and validation. So we have also uh released

the koshian SDK which allows you to interact with all the ERC8004 contracts that we have deployed on base uh cipolia, ethereum, cipoleia and optimism. Uh so with with with this SDK you also get um some more things. We have integration with x402 natively and you will also have uh IPFS storage for all the agentic work that goes into into it. Uh so like yeah you pip install kosian SDK and then to get AP2 you have to like install it manually. It's not yet available as a package. Uh so enough with this talk. Let's run the demo.

Yeah. So so here we are going to see like we have three agents like Alice, Bob and Charlie. Like Alice is the server agent, Bob is the validator and Charlie is the client who is going to request work from Alice. Uh so yeah uh oh sorry I'm going too fast I think. Okay like uh for step one is of course like uh starting the configuration for x04 and x402 uh and then it starts setting up uh all the essential primitives. We can also have facilitator for x402 so on so forth.

Yeah. So using this SDK which you can see like this is more or less the interface to register your agent. So you have your agent domain uh and and then uh this this will give you the ID. So we have ABC uh these are the wallets. So you like SDK will also set up uh wallets for you using uh coin this uh agent kit right? So this agents I have already registered it uh like I store that's the domain. So agent ID is 10. This is a wallet 11 and 12. Uh it's sequential. Now like a phase 2 is like where we start the verifiable stack and that's where the work also begins. Uh so the work is that Charlie is requesting uh the with this

description like find me a winter jacket in green you know like uh she's also like giving some parameters uh and then AP2 right now we have AP2 in installed. So it it it is going to like create uh the mandate for the card uh with it JWT token and then okay created and verified we get the uh now the second step is the process integrity where I was talking about like we need a verifiable execution so right now I'm doing it in a very inefficient way so t is something interesting to explore here with uh probably with fala uh so right now I'm just hashing the code and also the input parameters. So uh it's this is the hash.

Uh all right and then I'm using crew AI for the agents. Um and then the work has started like Alice is like right now doing the uh market analysis to find this jacket. Yeah. This is all the work that is being put in. Okay. Like analysis is complete. >> Oh sorry. So u the conference score is 73%. Uh so like process integrity uh like what Alice has done is being stored on IPFS using Pinata for storing uh and shout out to Piata uh team like they're great. So yeah you can see here it is being stored that's a uh

CD here. Uh right now next next step is like uh authorizing the payment. Now that the work is done by Alice. So Charlie has to now pay Alice and then uh our X402 gets activated and the payment is being processed. You can see the hash uh this is the production hash and this payment information will also get packaged into the evidence package that will be stored on IPFS. Uh sorry if I'm going over time just let me know uh if if my time is out. So with AP2 allows us to not just have crypto payments but also it allows us to extend fiat like you can use basic card or like

Google pay, Apple pay. This is all part of uh like you can integrate this with Kion SDK. Uh all right. So now Alice is requesting validation using 804 validation registry. Uh and then you get the data hash for that. Uh so yeah now Bob has started validating and this validation again like uh this these servers are using crew AI. These are their uh work outputs. Yeah. So the validation score comes out to be 96. Uh and then now Charlie has to also keep Bob for the work uh Bob has put into.

Yeah. So as as a dummy for protocol fee for Koshen we like we have 2.5% for every transaction that goes through it. Uh so all of this is again like going to be stored in IPFS. Uh here it is. So the validation is being stored. So we have the information like metadata on it. Uh this is the overall score 96 out of 100. Yeah, not bad. Uh now yeah so so in Kion like we want to like then package all these things and then submit this evidence package to our set of verifier agents. uh and then we want to promote like this agentic consensus uh where multiple verifier

Agentic Consensus and Workflow Overview

agents will be looking to uh validate the output and create a score vector. Yeah. So this way uh this is the workflow and we also get a monitoring summary for what how the x402 payment was carried out and >> we're running a just >> yes if you can share also materials and repo we can also like play around if you want. >> Sure.

Yeah like uh this is the explorer we built and then these are the agents which are registered here. Yeah. Thank you very much. Sorry for going over time. >> Great. And again, if you can share resources, I'm I'm sure you are all curious to learn more. Thank you, Samit and Cast Chain. Um, next demo. Um, and then we have one more uh, one person just reach out to me. This is an open call, so if you have demos, just text in the chat or text one of the team members and we can make this happen. Um, so I'm demo on behalf of Rich Patterson. uh who is now he wants you to know he's in Croatia he's biking he's without his smartphone and uh so he just sent share this video and uh but I

will play I think at a faster velocity >> allow the movie agent implementation to accept >> it's a bit long and so let's use 1.75 how I do with podcast should work >> hello here's a demo of the agent identity service for Ethereum blockchain this solution implements ERC804 and provides a seamless UI that interacts an on-chain identity and reputation registry and make sure and catch our Ethereum naming service integration at the end of this demo. The solution allows you to search for existing agents within the system. It also allows you to manage your own agents. To create and manage your agents, you log in using your social credentials.

Once authenticated, the system connects to your onchain externally owned account, EOA. And if one does not exist, an EO is created. Note the EOA address shown in the header area, you can select that link to view your account on Etherscan. Now you're ready to create an agent record and register it with the ERC804 identity registry. Let's create that move view agent. To create a new agent, you just need to define agent domain and an onchain domain is created. An ERC 4337 account abstraction is generated and registered with the identity registry. Once registered, an agent ID created by identity registry is returned and associated with the agent. It is important to notice that all of this happens without the user needing to interact with the chain wall and spend death. All of this is managed behind the scenes and paid for by a hidden pay master. A new movie agent agent is

created and registered with identity. If you clicked on the agent address, you would see an ether scan that is an ERC 4337 account abstraction. All fees needed to construct these onchain agents are covered by a transparent pay master that the user never sees. Next, we update the agent's agent card. Notice that the card is defaulted with the default ERC804 required fields of registration and trust model. The registration signature is created using the EO owner or the agent AA. The signature can be validated by the agent AA based on ERC 1,271. This agent card is maintained here and then copied into the actual agent implementation. While we are editing the move review agent card, let's look at what is happening behind the scenes. When we signed in, a user EOA is created as shown on the left of the diagram. When the movie review agent was created and we created a movie

agent account abstraction ERC 4337 that account is owned by the user's EOA that account address serves as the identity address registered in the identity registry. All this happens automatically. The agent card we are setting up is linked to this account abstraction. Delegated permission is also set up to all the movie agent implementation to accept feedback connection between the client and server via reputation registry. We next create a move client agent that uses the movie review agent service. A client agent is also created on chain to support the reputation relationship between the client and service agents. Okay, back to the video and us capturing the move agent card skills and other agent information. Notice the agent- dash card.json on the right hand side. The JSON includes the ERC804 fields, registrations, and trust models. The registrations include the agent ID, onchain address, and signature. The user can copy that JSON at any time and set

up their agent implementation. So, we have the move agent all ready to go. The user can select the arrow icon to view the agent dash.json hosted by the agent implementation. They can go to Ether Scan and see the onchain agent account. The user can also go out and view the delegation session package used by the movie agent implementation to accept feedback. Now, let's create the client agent that calls the movie reviewer service agent. The client assistant agent is a chat UI that allows our customer to query the movie service agent for movie information. Now the movie chat agent is ready to go and shows up in the list. Next, we update this movie assistant client agent card and set the feedback data URI. This feedback data URI is used to track feedback used by the ERC804 reputation framework. This client agent card JSON is copied and put in the

actual client agent implementation. Now that the client agent is ready, we can copy the client agent card to JSON and put it in the client agent app implementation. When we select the agent card JSON icon, you can see the agent card hosted by the implementation. Now we are ready to generate the movie agent delegation session that empowers the movie agent to create the accept feedback record within the reputation While that is happening, let's look at the diagram again and see how an on-chain delegation account abstraction is used to facilitate this process. It allows the agent to update the reputation registry without exposing the full power of the AOA private key. It also all this stuff to happen without

the user having to retain any gas or have any UI pop-ups getting in the way. The session package shown here is automatically created. The user copies this package and placed in the movie agent so they can talk to the reputation >> You must come today. >> Now it's time to show the movie in action. >> The interesting thing, >> this is an example application that Google Linux foundation created. They show how agent agent interact. The interesting thing here is really to show a vertical integration which includes how you can offer SRPless experiences with web freeot the delegation toolkit to delegate your agent to do stuff pay master at the station. So I really like that reach connected all the dots. U thanks Richard and enjoy your biking uh race. Uh now the floor goes to Tim. Uh Tim feel free to introduce yourself

before doing your demo. Uh you have like ideally around 3 minutes for Max. The floor is yours. >> Hey everyone, it's Tim Cotton. Uh this has just been a fantastic community call. So I reached out to Marco because we've been talking about like the reputation registry and threat mitigation for bad ratings. People who say, "Oh, that agent is competing with uh my agent, so I'm going to go do a civil attack where I pay for the services, but I leave terrible ratings." And so I was like, how do we even model that out? So at scripted, we went ahead and just built a really quick simulator.

Reputation Registry and Threat Mitigation Discussion

I want to show it to you and I want to show you what happens like before and after if we just add one additional primitive to the spec for letting agents uh register a dispute and whether they did something with the dispute or are leaving it up to the market to decide. So let me go ahead and share this. Share screen. Yep. I'm going to share the whole screen. Oh, sorry for the hole. All right. Can you everybody see this now? That is the question. Can I get a head nod from Marco?

>> Works. >> Very good. Okay. So, this uh there is going to be a link. I'll put a link in the chat in a minute. Um this attack simulator is very simple. You can set you know your initial reputation that the agent currently has. You can say how many events are we simulating? And it does some stochastic stuff to play with that and tells you, hey, if there is an attacker who has equally bad demand down here to the honest demand that's happening, what happens to your revenue? What happens to your final reputation? And so, you know, as an attacker levels up, ah, the hit is hard. The idea is if we add a dispute object that can then mitigate, you know, like you could say

something like, "Hey, I refunded this guy." uh or hey my thing actually worked and the validators can't prove it because it's image generation and it looks fine according to these people well you can you can bring that reputation back up you can prevent that by having just that single primitive so I'll leave the link you guys play with it um there's some mathematical assumptions here happy to work with everyone and actually like improve the simulator there's also like a really simple like spec for here what does a dispute object will like guys that's not final. It's just more like look want a proof of refund or we want to add a stake and say no I I I think I'm right.

That's what I I think would be really valuable as like a last mile for reputation in the registry. So thank you for your time and I'll talk with you in chat. >> Any other demo? Congrats Tim. >> Thanks bud. >> Need to dig into the details. If not, we have been super efficient. We still have 10 minutes. Very dense. >> So, >> who wants to go?

Trustless AI Agent Demo and Verification Process

>> I can go. Hello. >> Hey, Justin. Yeah, floor is yours. >> Let me share my screen. Uh, just just one second. Just just like a couple minutes. So we have been building this uh trustless AI agent for a few months as you can see. Uh I mean I mean this is the link I shared in the in the chat. You can just let me show I don't use zoom a lot. Um yeah so so like this is like a a simple trust this agent demo that use TE as you can see it works very similar to CHBT you can talk to it it will reply and uh once you reply it will give you this full verification uh thread just tell you uh that it's

actually talked actually talk to open AI use the TLS connector and it actually talk to GBT4 at this time so that you can be sure that the agent does not do something you know fishy back end and this thing can all settle on chain as well and it's like a more complicated example this agent behind the scene so I might be go a little too fast behind the scene it calls like a many a like a APIs so each of them has their own at like a signature and attestation so that the entire um uh end to end uh execution of the agent is verified of course The model is not the model is still talking to the closed

source model hosted on open AI but uh but the TLS connection to the model is verified and in the future if we can use the Nvidia confidential computing we can host the model as so so to reach entire uh AI agent end to end verification. Uh so this is just like a a a simple live demo of the what the trustless a agent will look like and of of course like the uh the this whole commitment will be settled on chain and so that uh it it will be yeah it's a immutable commitment.

It's just a super simple demo and you can play with it at agents.sparsity.ai. So that that's that's a demo. Yeah. >> Thank you, Justin. Questions? Questions on Justin's demo or Justin, if you also want to share the link in the chat so we can >> Yeah, share the link on the chat. And also uh >> by the way we will prepare a recap probably in the group of all the demos done today.

Added them as a note to the recording. >> Hey guys, is it is it possible to demo? >> Oh, awesome. Yeah. Hey guys, we're from uh Bond Credit. So we're building we're actually building the credit layer and we're working with several agent that we're scoring uh including Codex. I think I saw Justin views here. Um amazing. So yes, we already had a formula. We work with Giza and actually we have our scoring that we did

for ARMA. So, it's mostly on chain and we're actually not using ERC80004 yet, but we really want to be like the first service to um provide actually more financial data because we're focusing on credit score and we want to give credit to those agent. Um yeah. Uh what is it? Yeah, I think this is this is all I think there is a problem on my uh I don't have enough tokens for for computing the score and I apologize for that. But uh yeah, we're happy to talk to any Asians and we're happy to contribute as well. So, thank you.

>> Looking forward to on 8,04. >> If if I can steal a minute or two. Um I'd love to to get our product in front of some people and uh uh a lot of the problems you guys are solving are are ones that uh we really need help with. So, uh if you wouldn't mind, I can share my screen. Justin, maybe if you can be like synthetic and take one minute because we want to take >> easy. Yeah. On this back end. >> Yes. Questions especially about the builder program guys. Question to Jesse >> um Yeah. So if you see my screen basically um all of our agents are sharing uh uh Intel or or rather we want

them to share intel. We're performing tons and tons of analysis consuming literally billions of context tokens. And um we want to be able to open this up. We want to be able to share it. Uh we even let people add, you know, any MCP server, whatever, uh to bend down those analyses. But there are a lot of attack vectors. I just DM'd you on Twitter, um Tim Cotton. Uh but yeah, I mean really like uh we're we're trying to to break open this space and figure out how to lock down um some of these protocols. So, uh, anybody kind of building in the, um, agent security space, etc., uh, please DM me on Telegram, JB Crypto95. I'll put it in the chat. But there you go. There's there's the minute.

>> Thank you very much, Justin. Then we can for sure dedicate more time in one of our next calls. >> Yeah, maybe we can have one uh focus on like security validation. It seems like uh important enough topic. Yeah, super super super important. I think uh like Ethereum at testation service um we really wanted to use but we couldn't because uh it it wasn't super secure. >> Sorry, there is one qu one question about DNS. So how can we be decentralized if we rely on DNS which is centralized? Uh you can directly put on chain a pointer to an agent card which is on IPFS if you want. So uh without going through the DNS system

Decentralization and DNS Alternatives

that was the question from just Jake Jake guys question on the builder program or on the event or everything that uh David and others discussed. Is there like um a universal uh site that we can visit to see kind of the the results of the program and all this stuff? >> We don't have it website uh >> just the telegram >> yet the design so you can use 80004.org.

This will be the you can already find something now. >> Okay. The FAQ of the uh builder Brango will be soon published here. >> All right. Yeah, we're uh we're buyers of of some of this uh tech. So definitely uh reach out if if all of you are are building. >> Hi guys, thank you for the the presentations. It was super cool. I just have a question about um is there some form of uh layered abstraction where we can build on top of someone who's doing more infrastructure like is there like

kind of um 8004 as a service or stuff like that so we can avoid rebuilding everything from the ground up? >> Maybe an SDK would be cool. David, what do you think? >> Yeah, I think that would be really cool. And um I think this is something we'll be discussing with Leonard that like uh beyond the ERC we want maybe like a set of docs like um maybe there's even a different SDK for like registration and then like the more security aspect >> for sure. Yeah. To improve the devx or allow these things definitely.

Cool. Thank you. >> But I would add um that's maybe like going to be the focus after uh we launched the um stable spec, the V2. Uh yeah, which is kind of like the next milestone for us. >> Yeah. Just uh maybe another question from our side. Is there going to be a registry for uh third-party reputation providers? How can we actually uh connect and became like uh providing our score? Since we're pulling most metrics onchain, we also

Third-Party Reputation Providers and Permissionless Systems

have some metrics which are offchain. We focus mostly on onchain data. But uh I would love to know if it's going to be a registaries, there's going to be an approval process. How do you guys uh see the process uh to >> fully permissionless? >> It's fully permissionless. So you can add your >> So you can add your data >> as a reviewer >> reviewer directly. >> See my voice. >> You can directly add your data as a reviewer using the ordinary reputation system. And then if you want to create experiences with just filter the explorer or just the application experience by your reputation, you can do that by filtering. We are also exploring if this could be done on chain and not just offchain depending if you

want to aggregate just the onchain data that we will put or also the offchain one. >> But everybody will be able to be build on your data and vice versa. >> Okay. And can we send just our final score in or is this something should be uh transparent? >> You can also directly send the final >> Okay. Yeah. Perfect. That works for us also. >> Yes. Not to share your sec secret recipe. I guess this is a pretty >> Exactly. Yeah. We spent like three months researching agents and formulas and talking to experts uh and we finally came up with like say version one of it.

So yeah, we try to keep some uh some of it secret as I say. >> Well, we're trying to grow the pie. >> Amazing. Thank you guys. Uh great work. We love it. Thank you for joining. We have space for one more intervention. Somebody want to say anything otherwise? >> I'd love to say something. So, uh, two years ago, the decentralized AI summit happened and we saw, you know, dozens and dozens of companies together. And I just want to encourage every single one of you that what we're building, if we actually make this composable, can work together and take on OpenAI and all the big guys. So, just want to thank everybody here and empower you because

this market is a multi-trillion dollar market and we're going to build it together. Thanks, Mike. >> Let's go to close. >> Oh, an amazing >> an amazing closing remark, I would say. >> Awesome. All right. Thanks everyone. See you in the group. >> Thank you guys. >> Thank you everyone. Nice to meet you.

>> Thank you. Talk later. >> Thanks. >> How do I check it out? There he is.