Trustless Agents (ERC-8004) Call #3

Transcript

Loading participants...

Okay. Hi everybody. The music is getting a bit mystic. I think we can start. Welcome to the third community call of 2004. Um well it is like three weeks since the last one but every community call it feels that ages have passed in terms of uh progress. Uh that's the plan for today. So we will um discuss updates in general on the implementation and on 804 from a specs perspective and we will talk a lot about the agent registration.

Community Call Overview: Discuss updates, agent registration, dev tooling

Second in the last call we promised that the focus of the community would have been dev tooling and we will discuss progress here. Uh we will share the first version of ecosystem map that it's populated thanks to your contribution. We will have as usual the community time with a few demos, not too many because as you have seen the the agenda is pretty packed and today we have like a specific running list already defined and then uh we will finally present the agenda of the in-person event that will happen next week on Friday at Dev Connect. David will pitch this. It's a it's a very exciting agenda and if we

will type we will do some Q&A. Let's get into uh it. So I would say that the the takeaway is that we're really learning by testing together uh which is uh exactly the plan. That's why we are on test net and um as we anticipated uh the specs are almost consolidated. So there are a few things that are changing but they are very very small. So that's a bit for 8,04 nerds. So we the only maybe slightly relevant is that we are removing feedback because based on your requests and suggestion um it was creating like too much friction for reputation and uh it was like a bit complicated to implement it and so we

will just remove it and uh in order to defend from civil attacks people will probably just filter uh reputation by the different client addresses watchtowers reputation system aggregators all the others are very small things. So like changing name of variables to avoid the collision with Viper uh set agent theory was already implemented actually just [clears throat] wasn't in the spec and then really deacing from bytes to string to enable like easier uh readability from dev tooling.

Um the PR hasn't been submitted yet for these small improvements but it's ready. We are sharing this now in the chat if you're like uh very technical with 804. While these small changes in test net that could still require uh uh updates to SDKs will be released like super soon. We wanted to make it for this call. We didn't manage to do that but I suspect it can be like maybe tomorrow and for sure this this week. So all very small uh changes. Um mainet has been postponed for this reason. We want to make sure that we are future proof. Uh we want to continue to listen to your feedback. Uh also at the event at dev connect we want to work with

registration. So we prefer to to take more time uh to also work on audits. Neithermind and other organization are helping uh with that. So we we we postpone to to to uh December. Um where we are right now. So during the last community call if you remember we were between implementation and dev tooling. So what happened is that now we have three SDKs working and uh we have actually it's four or five probably more it's it's just impossible to count the number of explorers that are popping up every week in the builder program. Uh but uh we're really going into the right direction. Um dev tooling I mean it's

Agent Registration Focus: Importance of discovery and taxonomy usage

not done. We will I think continue to have strong improvements over the next months but it's I would say that the urgency that was there in the last community call has been like uh partially uh sold and so now we can move to the next step which is I think uh probably the first one which is very exciting which is like registering agents um and uh I think this will like be the focus of a community for some time for the next few weeks also. after dev connect and then of course we will move to reputation but before thinking about reputation we need to make sure that agent agents are there right and registered and um um so a couple of words on this so discovery um as a cold

start problem u but but that's fine I think because uh I mean uh everybody in the web 3 and um space at the intersection is talking about it so we are a strong community so I'm not concerned about that second discovery it's two-sided You need both explorers and agent to create value because agents want to pop up in explorers and and explorers of course want to want need agents to make the explorers meaningful and of course both sides uh needs to speak the same language. Uh what does that mean? Well 84 does the job of course but it's also important to use it in the best way and uh and also use taxonomies. So um uh we have worked in the last few weeks and if

you also talk to the for example to a A2A ecosystem uh there isn't like a strong skill taxonomy that projects are using and what we uh um and again it's not mandated from the from the spec it's just like an improvement it's just a dev tooling thing which is what we are working on right now but instead of reinventing the wheel um we um we think that what Linux foundation now but it's originally coming from Cisco it's called OASP is um is a good place to start um you have a skills it's two dimension skills and domains all this is open source but it can really like help you

explaining with clear unique keys if you're doing an agent which is doing education if or if it's doing training and so what we um we suggest is just to leverage uh and uh and just tag your your agents to make sure that that they're uh well described in explorers. We actually are launching today a nice guide. It's a it's 804 compatible. It's just a way to make things uh very easy to understand. So, how to make sure that your agent looks great. Of course, don't forget name, image, and description. And these are how you introduce yourself to the world. how you describe which uh what you're what you do, what problems you

solve. Second, don't forget about endpoints at least one and um and if you use MCP uh you can put your tools names if you use it to you can put skills. Here we are using O field skills. You can use ENS. It's just a recap. It's stuff that probably you know but that's in just one page you can really like pick what you want. Don't remember to put at least one one wallet can also be on another chain compared to when you're registering. It's where you want to receive money basically. And and third, that's where taxonomies could be useful.

Um tell us what you can do. So we need to really want to know what your agents can do and you can pick domains and skills. Um there is a in this repo you find we prepared a nice JSON with the skills and domains already packed that you can import and and pick your own the public website this one that I was showing is linked here and um um and by the way in one of SDKs that we will pitch later this is already supported so it should be like actionable uh immediately and finally don't forget about putting the it wasn't for registration to link back to the chain that's for security reason to make sure that the owner of the server and the um and the person that registered the agent

is the same person. Um it's not and um maybe a couple of more things you can also advertise if you are using X42 to receive money and if your agent is active or not. So I by default SDKs will consider it false and it's active. If it's active it means that you're ready to to receive requests. Um so the the this is how the file looks like. Uh again it's it's nothing different from uh the past version is the same. O was already by the way in is already online since beginning of of October in our specs. Maybe um one thing guys please don't put there are many teams that have insert the entire JSON in the agent URI

field. So instead of a URI URL some projects are directly installed the JSON onchain and uh for example the builder of 804 scan that we meet we will meet him later today in this call got crazy because of that. Um so the message is really the moment is here we are now ready in the next few weeks from now to Christmas let's make sure that we all register the best agents uh that we have. Okay. Um and um somebody David is probably sharing the links that I'm sharing now in the chat. We also share all the links afterwards in our telegram channel.

Dev Tooling and Piñata Introduction

>> Now let's move to dev tooling. Um we have many talks so please make sure you stay uh under a few minutes three four minutes depending on uh your presentation. And I'll leave the floor to Matt from Piata that will tell us more about how Piata can help us uh with of course the IPFS resources that we need to register agents and to share feedback. The floor is yours. >> Awesome. Give me a second. Oh, turn my camera off. Not to share. All right.

>> Can you share information? Okay, great. Yes, I sure can. Yeah, I actually I turned my camera off when I was trying to share. That's uh never gets sold. Cool. Um All right. So, my name is Matt. I'm from Piñata. And last week, I gave a little bit of a sneak peek in about 15 seconds. Marco so graciously uh let me speak again this week. Uh so, Piñata is running an AI builder program. And what is Piñata? Uh Piñata is crypto's file storage solution. We run on top of IPFS which is kind of like the gold standard for how you work with metadata and off-chain files in the blockchain space and ecosystem. Uh this is being utilized uh more and more

heavily by Agentic Systems. Uh a lot of people building 804 are using this for a variety of reasons and yeah our whole sales pitch is we are very easy few lines of code. uh sometimes with even out code we have a UI you can get fast secure and scalable agentic file management uh without complex setup. So uh again like how is piñata being used for specifically in the 804 community and agents uh if you look at the initial proposal for 804 you're going to see identity uh validation and feedback registries IPFS plays a part in those uh this basically allows for a tamperproof uh file layer for storing

that type of information and it's crosschain uh no cryptocurrency needed kind of works everywhere in a very agnostic Okay. So, uh we are also seeing agents utilizing IPFS to transmit data amongst themselves. Uh even outside of registries, agents are using IPFS as a way to save state to uh transfer files in a way that is tamperproof and uh distributed and decentralized. And then this is a little bit more of a piñata specific plug. We do have a uh X42 plugin for private IPFS which is allowing agents to buy and sell uh data using X42 in an IPFS compatible way. So still get those same CIS. Uh you're just

able to monetize them now which we think is pretty cool. So uh what we're offering for everybody, this is our AI builder plan. All you need to do is email teampata.cloud. Just give us your account email and your project name. Uh, and then tell us what you're building and we're going to give you a free year of our Fiesta plan, which is $100 a month. So, yeah, full year of free IPFS, free piñata. Um, just let us know what you're doing and we kind of want to uh see what people are building. We're really curious. So, looking forward to it. Um, that's what we got. If you're asking yourself, do I care? And we ask, are you building with agents? Then I think the answer is yes.

So pretty simple question and answer and I will leave it at that. Thank you for your interest. >> Thank you Matt and thank you for like supporting this very early phase of uh Gentic Economies and we will continue to talk about IPFS and the floor goes to another Matt actually Matt Hamilton. Nice to meet you from Protocol Labs and Fcoin. >> Right. Hello. Uh hopefully you can see my screen and hear me. Okay.

Filecoin PIN for IPFS and Agent Storage

>> Yes. Perfect. >> Yeah. Okay. Brilliant. Right. So yes, you have the benefit of a double header of Matt's uh here both talking about IPFS and storage. So my name is Matt. I'm work within the uh Falcoin and Protocol Labs ecosystem. And what I'm going to talk about is a tool we have here called Filecoin PIN that allows you to pin your uh agent cards to IPFS and also backed up uh on you know durable storage um archival storage on Filecoin. So we've got a in in the docs here we've got a tutorial on Filecoin PIN uh using it specifically for 804 agents. Uh so I'm going to give you very quick snippet

of that. Uh so the first thing you need to do uh you need to install the filecoin pin uh command. It's an npm command. So npmg uh install filecoin pin. Uh that's all detailed in the tutorial. And uh then we set up payments. So filecoin pin payments setup auto. This takes a couple of minutes to run. So I've run it ahead of time here. And it connects to network. You need some uh test fill and some USDFC which is a USD stable coin used for payments. Uh both of those are described in the instructions and there are faucets that you can get those for. So I've done that. So the next thing I want to do is I'm just going to pin a uh card to

um here. So I've got a agent card in JSON for GitHub uh for an MCP agent. And so that's just going to go ahead and do that. Uh so as that's doing it, takes a couple of minutes. You see it packs it. We get a root CD here. and it's going to store it with a storage provider on Filecoin and handle all of the payment for that for you. So, we have the CI um which means I can take that CD uh ends 7DI um and I have it here um 7DI and we can see it loaded here. So, here is our um card here. But as you can see, it's actually um storing this on chain here. Um and

you can actually look that back up and get confirmation of that. There's a command again, this is all detailed in the instructions here, Falcoin pin CLI. Um there is a command there that you can run that will get this back. So, IPFS um Falcoin pin data set and it will show you uh what data sets there are and what is available. So, that's still running there in the background. Um but uh yeah, so we'll just just wait for that a second. See if that should finish in a minute. And then the next thing I'm going to do is just call the once we've got that I'll just call the cast command to actually cast that just to show this

um using base and the registry the um uh kind of implementation of the 80004 registry that is on base. Uh just to show that I think this is just taking a little while to confirm here. Um yeah. Okay. Uh well, live demos. Uh so yeah, this is just uh like I said, this is just confirming it with a storage provider. Ah, there we go. It's it's finished there. So that's now stored and confirmed on the network. We've got a data set ID. So I can actually um uh put that there to data set 30. Um,

I can actually call Filecoin pin data set and it will actually confirm that we actually have that information stored on Filecoin so that you know cryptographically that information is all stored. Uh, so there we go. Um, that was a quick whistle stop tour of Falcoin PIN. Um, like I said, if you go to docs.falcoin uh.io, io. We've got here in the builder cookbook the uh Falcon pin CLI and also the specific agents demo there as well. And that also looks through uh walks through to a GitHub repo that has a few scripts in there to help you just check you've got things like prerequisites set

up and just go through the the steps that I just did there about uploading, registering, verifying. Um there's some scripts there to automate that. So there we go. And I believe I'm h handing over to my colleague Jenny who's going to talk about some offers that we've got on at the moment with uh Falcoin. So I guess over to Jenny. >> I'm I'm your local Jenny uh or standing in for Jenny. She just got off this morning. Um >> y >> but yeah. Hi friends. um here to give you a a tiny bit of background about um how uh Matt's awesome demo um storing all of this onchain agent data in onchain decentralized storage um is is working. Um and so this is all building

Filecoin Onchain Storage Overview

on the new Falcoin onchain cloud that is running entirely in smart contracts in Falcoin land. Um it uh layers a number of services and smart contracts on top of your friendly Filecoin storage providers. Um we have onchain payments using stablecoin or ERC20 based payments. Um and we have a warm storage service contract that is um doing all of the the managing of storage contracts. Um and then filecoin pin is a uh an application built on top of this whole onchain um set of contracts that is enabling the the IPFS uh storage. And so this is now live. You can go to pin.falcoin.cloud.

Still we're still in uh our production testing. This is still all operating on the Falcoin calibration test net, but this uh has a free set of of wallets uh for early builders to um start uploading their assets. You can use this as a web app. You can clone this um but there's also a GitHub action and the CLI that Matt was demoing. So, as builders, you're probably going to be using the CLI anyway, but just if you want to get started and and give it a try, um you can do that today. Um, and I mentioned that this is all running on the calibration test net right now. This is what's happening under the hood. There is a funded wallet. We funded a wallet on your behalf to make life easy. Um, you can upload files. You can verify them in real time. Um, as Matt was

showing and these are getting published to the interplanetary network indexer and then can be retrieved through any IPFS gateway, Kubo, whatever route you please. And you can also um directly fetch from the Falcoin storage provider. Um, and it's all being served fully on chain by a network of decentralized nodes. Um, and coming next week, we are going to be also bringing this to the Falcoin mainet. We'll have a sponsored uh contract for 8004 builders. Um just in case you don't want to go and uh start purchasing your uh Phil and USDFC uh stable coins uh just to get started um you can start using the the um calibration endpoints today. But if you

want to u stay aware and and uh be be an early part part of this as we go to mainet. Um if you're in bonus Iris we'd love to host you at our launch party at deepend day. Here's the QR code. Um otherwise you can subscribe to Falcoin Cloud or follow uh the live stream on the Falcoin Twitter account um on November 18th. Come be part of uh bringing this to um mainet and having fully endto-end decentralized storage for decentralized trustless onchain agents. Um and we're we're excited to be on this journey with all of you um going going to GA in uh real time alongside all the work you guys are doing. So check it out.

Thank you Molly and good luck with GA. >> Very exciting days. >> Okay, so the next speaker is uh YQ from Allayer. Then Nad will go next and uh to make sure we stay within around like the three minuteish uh Jesse will like show countdown a countdown after like uh two minutes and a half. Let's try. Uh so YQ is actually is is a improving dramatically the Devax experience not just for 804 but also for X42 and we'll present today I think 804 scan. The floor is yours. Thank you.

Upcoming Filecoin Mainnet Launch and Events

>> Yeah. Thanks Marco. Uh can you see my screen? >> Not yet. We can hear you and see you well. But >> um I just share this one. >> Let me do it now. Can you see it right now? >> Yes. >> Okay, great. Um yeah, thanks for Marco and David and um the EF people for the invite. Um yeah, so w here. Um I printing this uh AO4 scan just uh because I previously I spent a lot of time on digging into all the details about this uh video for design implementations. Um I really to help my um understanding like sort of how it

works and u um how we can really uh like explain the details of a lot of agents. Right now it's pretty straightforward like for all the developers. A lot of like sort of project developers asked me like how they can sort of register for this gun. I told them like if you register directly on like sort of this AO4 contracts either on the u etherum polia or base uh we will auto index these agents. Um you can directly see like sort of right now uh all the information for each agent and also the like sort of this contract uh registration uh information. Meanwhile um as Michael mentioned like so I tried

the best to show the metadata uh like because a lot of developers users they directly invite all their data everything into the token URI. So in all different formats HTTP and H IPFS and sometimes also this B 64 and all different formats uh we are trying the best to sort of uh uh elaborate the details. Um it just yesterday we sort of enhanced this um information like sort of um um showing a little bit better. Uh right now if you really want to see the details uh we also further load the metadata for a lot of these agents. Uh you can even for some images um like all the MPC, MCP and A2A if they already

have some content uh we uh you can also go to this identity section uh to check all this detailed information. Beyond that uh we also started to index this u uh feedback and also validation uh informations. Uh of course like a lot of the things are still on test net. um we just try our best to index as much information we can get from this um uh from the contracts and also the registration information from the agents as possible. Um and later on like once this uh AO4 like sort of contract on the midnight and once we have more agent right we will later we will probably have more this kind of interactions with

some of the agents especially later we may have a lot of the XO to payment agents right we will we already start started to index the data for it uh and beyond that um we also have something a little bit interesting um but it's still in This kind of a mode is basically uh sort of we call builder as uh is just like quickly help you to build this uh um like able for like agent to reg for the registration. Uh it's not for the agent itself is like help you to uh to quickly uh fill in a bunch of information and then you can uh register based on some of the best practice. Of

course we will further improve um this kind of informationational description based on the best practices u provided by Marco and the team and after that we try to make as straightforward as possible. So once user just fill in all these different informations they can just register. There are this very simple we can call wizard or builder um like sort of this um yeah uh this this kind of platform. So uh right now it's on Solia. So it's a pretty uh straightforward and fast and uh just after a few seconds or minutes you will see like sort of the agents

will be uh index into the scan. Of course, if you have further like sort of uh requests like what kind of things we should show on this AO4 scan, feel free to uh DM me uh in this uh TG groups or just um uh we um on our side or on my side, we um we really want to build this like A4 scan as a public good tool and to help like sort of the developers to quickly register their agents uh on like sort of AO4. So in the future um we can quickly like bring a lot of agent to interact with each other for better agent economy. Yeah.

Yeah. That's all on my part. Thank you. >> Thank you. It's cool to see like the product that changes every two days. The velocity is is astonishing. Uh next in the agenda is Nad from AEN Labs talking about how AEN cloud can be of support in building trustless agents. Not sure now that he's with bus.

Agent Zero SDK Overview: Discusses dev tools for agent economies

>> Um I can ping him but we can just go on. Yes, >> in the participant >> please do that. And the next one in the agenda is me again. I'm sorry. And uh with a different hat. So um I want to speak about agent zero. The general idea of agent zero is that in terms of dev tooling, I mean as a developer, I don't care about an ERC or another ERC. Who cares about ERC? I'm just a dev and I want my agent to take part of agent economies. So the we have a lot of SDKs focused on how to build agents, but how about how to make agents connect it with to other agents to find

other agents to coordinate other uh with other agents and so yes of course 84 is important on that but this SDK could also potentially uh cover other feature and uh and other part we we want to cover the the whole uh flow of agentic coordination. We launched this 12 days ago and uh we have received since then almost two community PR per day and that's why actually we just released new docs because the product has changed a lot compared to uh two weeks ago. Let's get into it. I will try to go super fast because we only have two minutes but the the docs are very structured and with this SDK you can abstract it4 and do

stuff with just one line of code. Um the SDK is available as a Python SDK or as a type T type SDK. It's called agent zero. Uh and uh let's start with a quick start. You select a default chain but now we support multi-chain. So you have a default chain that you use to register your agent. But then as you will see you can use multiple chains to um to query agents on another chain. You have an RPC you can put a private key or directly the signer. It's compatible with a filecoin CLI that we have seen today with pinata and directly with an IPFS uh client. You can register your agent with name description image with one line. Uh

and this is sorry this creates the agent doesn't register it yet. You can add MCPS and A2A endpoint DNS. You can add the wallet. You can add the T models. You can add custom metadata. You can add skill and domain as we were discussing earlier. Somebody in the chat wasn't able to find the finance skill just because finance is a domain. So if you look at the domains you will see many uh domains correlated to >> and uh and finance >> you can if you can mute Yu please you can active or the agent directly from uh from with one line of code. And finally you can

register the agent. As you can see, it's just one line. So after that, you are fine with all the parameters. You can just register the agent and you can update it later after having uh loaded the agent. So you can load the agent or also get agents built by other. You can use feedback. So you can um sign the feedback code. This will be removed to make stuff simpler. You can define the feedback with one line. You can submit the feedback. And let's move to my favorite part which is search.

This is a multi-chain global search. We leverage subgraphs. So we are also giving like free subgraphs with infinite query for everybody. We now support polygon testnet base test net and sepia test net. We are going to support network soon. So here you can find all agents that know how to code python. You can query agent by uh reputation um with a mean average score. You can have multiriteria search. I want the who wants to agents that are good with JavaScript and can generate code. You have pagination and uh um this is it was just like a a very fast go through but

walk through but you have like almost 50 methods. It's very complete. Uh we just went through the example section. I suggest to go to the usage section. If you have time, I will share the link now because it's it's it's more complete and you will see like all the end points. And this is um um growing and changing very fast. And I will share in a second the link to the agent zero kitchen which is um a telegram group pretty boring where we just discuss PRs and who wants to do that, who wants to add this. Uh but it's uh to to to see the product evolving live. The product has been already downloaded by uh more than 1,000 builders that are uh using it.

Okay. Uh let's give a floor to and we are looking for contributors. So if you want to help uh that would be great. Jesse, do you know if Nad is with us? >> Uh I don't think so. I paint him but not yet. Okay, >> cool. I think I just go next with the EOS app and get it started. Awesome. So, would you be able to share the slide?

>> You can share it. >> Okay. Let me try to see how to do this from my screen. Sure. Can you guys see the screen >> perfectly? >> Okay, perfect. Awesome. So, lots of movement happening for the ecosystem side of things. Um there is we have about uh 1100 builders in the 804 builder group already and I think we try to push a lot of people to go to the right link to submit for the ecosystem app and currently 67 projects have submitted uh mostly ranging from

Ecosystem Updates: Overview of projects and categories in development

development testing and production. At these stages there's a lot of different categories everything from AI infra like data hosting memory context eval wallet privacy to reputation scores different people are playing with different metrics onchain offchain to uh sort of have a more um comprehensive overview of the reputation having a lot of different types of discovery layers explorers um different uh format and and methods of doing validation doing agent orchestration there's obviously a lot of defi agents. There's some, you know, decentralized robotics projects are uh also emerging. And then definitely a big category is payments, commerce, even like agent stable coins and credit cards

for agents that we've seen, which is very interesting. Some community media projects as well. And uh last but not least, many agent frameworks and marketplaces. So uh we did a preliminary um ecosystem map. This is nowhere near being comprehensive. This is just mostly taking ones that are um devel uh testing and and in production and the logos that our designers are able to find. Um so if you want to like be a part of the map, please uh this is like uh actual like permissionless open uh version of the ecosystem map. Um if you want to scan the QR code, you can find it here in the first QR code and and then if you want to be a part of it, feel free to fill in

the second QR code. But yeah, we have many cool projects um happening and then we'll be inviting some of these projects to speak at the trustes agent day which we'll cover um in at the end of this presentation. That's it from me. Thank you Jesse. I think we're moving to demo time and uh we have uh three minutes each demo and then so we we will have then time to discuss the dev connect event happening next week. The first person is uh Robert Robert Brighton from Braxis that will present making 804 feedback

Demo Introduction: Transition to demo time and upcoming dev connect event

frictionless in one click. Hey guys, I've uh pre-recorded just to make sure that I'm going to stay within the time given that everything ran over, but we've um cooked up a demo to showcase the practice explorer which is integrated with the ERC 80004 onchain reputation standard. So what you will see is new agents being registered, their identities being viewed, metadata and capabilities. Um [clears throat] and then feedback is left after each interaction which then updates the agents onchain reputation score in real time. Um so let me cook that up for you guys so that we stay on time here. Bear with me.

C can you see my screen? >> Yes, perfect. >> Yeah, let me mute myself and then I'll play it. >> I think you have to unmute for sound. We can't hear anything. Robert, I don't know if you want to speak on it or you can just look at the subtitles.

>> Sorry. Is the audio not working? >> No, it's not. >> Given the time constraint, it is a subtitled. So, if you're okay, I'll just continue there to make sure that we stay on time before I try and figure it out if that's okay. >> Yes, please. Would be great. All right. And I think that this demo is a good

reminder that 2004 is the infra but in terms of reputation how this is really valuable for users how experience can be frictionless that's up to apps that can make it like one click easy with like stars and ratings as we have just seen. >> Yeah. And the other plug is that um actually the way you leave ratings uh in 804 is like very generic right it's like a very simple uh structure and you can be creative on like what type of rating like you can slice like the 0 to 100 into stars like yeah >> yeah just to add on to that the goal here is to make it programmatic so that

the agents can give each other the scores um it is possible to to add in user ratings as well, but of course that's more subjective to to you know gamification there. So we're going to keep it programmatic as much as possible. Appreciate the opportunity guys. Thank you. Onto the next one. >> Denver will speak about simulation environments for 8004 reputation. >> Sweet. Thanks Marco. All right. So we've been doing research into how to do uh ratings and rankings of agents based upon reputation. Uh we found that it was then in turn useful to have a simulation environment. We'll be open sourcing this under the Apache 2

Simulation Environments: Discussing simulation for reputation systems

license. So hopefully it's helpful to you as well. Uh to walk you through what's going on. We have a local PI EVM backend where we then spin up fully like ADA compliant agents. They register and then they interact with one another. um within our simulation loop. We will then get all of this spit back out to us. And we found that this is useful for a couple of reasons. One um you know just having this locally let us spin up our own contracts, deploy them, etc. Fund accounts, do all of that stuff without having to worry about test net connections. Uh you can also point this if you so choose to at uh any you know EVM compatible chain. So, if you're familiar with something like Anvil from Paradigm, that's also another backend

that you can use. Uh, we then have our agents themselves, which have a couple things going on. Um, you know, they're ADA compliant. They have wallets. Uh, and then they also have a full set of, uh, tools that they can interact and use. Um, those are just, you know, a variety of functions. And then the peers being other agents registered within the simulation uh that they can actually interact with. Uh, an important thing to note is you can really bias your simulation output mainly just through a prompt. Uh, our goal is to, you know, be able to simulate what does a malicious um, reputation provider look like? So, if I'm trying to artificially inflate a

certain agent, are we able to pick that up and determine that u based upon the activity that they show there in um, at the end of the simulation? Like I said, you get the full output of everything that happened uh just as you would within an EVM system. Um so yeah, we found that's useful in our research uh in being able to simulate, you know, certain desired outcomes. Um this really can like scale up pretty much as much as you want. Uh and you can just kind of have these guys run in the background and continue to communicate with one another. Um hopefully this is useful to you. If you want to get in contact, uh, like I said, we will be open sourcing this, so anyone that wants to can use it

in turn. Uh, and then you can also just follow us on Twitter, um, Simak Labs, or reach out to me on Telegram or on GitHub. I'll share this in the chat. Should be fun. I think this is a big problem broadly when it comes to just how do you actually make meaningful insights out of reputation uh, so that you can effectively rank agents. we see manipulation in things such as El Marina um in a web two context and I think we have a interesting opportunity to actually make something that's uh I don't know let's call it less scammable.

Thanks everyone. Thank you. So next one is Marco from Oasis but we'll talk about trusted execution environment validation for agent keys. >> Exactly. Thank you. It's a pleasure to present. So, we've been building actually trustless agents with quite a few projects. Um, oh, someone is already sharing. Um, what we generally advise projects and actually users too is to think about three questions and anyone that has talked to me knows I keep on shilling this. Everyone should be able to answer who's running the agent, who's controlling the agent keys, and who can upgrade the agent. And I assume everyone here has now built some agents and you

will realize it's really not that trivial to answer these questions properly. Uh so for this demo we have actually built a validator agent that takes not only TE deployments checking at the stations but also checks if the agent is utilizing keys that were derived in a decentralized key management system meaning the private key has never left the secure enclave. So I don't actually know who's sharing the screen now. >> Yeah, I think I can take over it now. Yeah, sorry. >> Basically, for those of you who don't know, Oasis is running this privacy focused blockchain. And here is an example from our explorer of an application. And the one thing I'd like to expose today is basically every um TE

replica has this um access to this metadata key value store. So only within TE you can store stuff into this data structure and it's public like anyone can read it right. So now uh every Oasis app can also register an ERC80004 agent of course. Uh but how do we prove you know that some agent you get from the 80004 is now really running inside TE and that the keys that it it derived that really derived from the TE and yeah that's where we use this metadata data structure. So you store the public versions of the keys inside. Uh and then we need to prove you know that that keys never really left the TE and uh first you need to audit your application of

Application Validation with Blockchain

course but this is outside of this scope. You need to audit all previous versions of your applications uh as well. So blockchain is really useful here because all application upgrades are stored on chain. And the second step uh is that we need to prove that the current instance that's live that it really um running uh from the source code we audited right. Uh so how we do it is um we um when the replica spins up uh it stores this public uh key to the metadata and when you register an agent uh to the year 80004 on Ethereum uh you store this Oasis app ID inside the metadata as well.

Uh so when uh the app is started it stores the public key and then triggers this validation request. Okay, this is the validation registry uh part and our validator agent then basically fetches um the raw metadata. Uh it checks whether the app really exists on chain uh if it's alive and also compiles the whole image and checks whether the hash matches the one that's stored on chain and then this results in the validation score. So if everything is fine, the score is 100 and you also get this uh nice validation uh response. Uh so this is the example live on the explorer. And yeah, I admit I'm also one of the sinners who stored the B 64 uh encoded JSON.

So uh in our case uh we stored this raw metadata copy uh inside the transaction. So yeah, you can see that this ERC804 signing key for example here is now really uh that matches um the one that's stored on chain. Yeah. So the goal of what we what what we we've been doing actually is um the user needs to check that validation response score and the desired public key is now present there. So uh that then they can trust the agent. So this simplifies the whole you know technicalities around the TE. Uh but of course one still needs to audit the actual source code. So if the key you know if the app is broken then um yeah the key left uh the TE and maybe

developer posted it somewhere and leaked it but yeah this is outside of the scope. Thank you very much. Thank you to you and to the whole Oasis team. So Nad is with us now. This is a preview uh uh of just two three minutes of what uh Nad will do in at the trustless agency day. We will have an entire product talk dedicated specifically about what a cloud uh is doing for trustless agents and then David will close presenting the agenda of event.

Preview of Trustless Agency Day Event

>> All right. Thank you. Um let's see here. I will share my screen. Actually, can I go next >> or last? >> Yes. Yes. And Louisis, if you can unshare, I still see your Okay, David, up to you.

Hey guys, this has been great so far. Uh, let me share the agenda for the trustless agent day event. Um, yep, we've been hard at work um, organizing this uh, in parallel of all the updates um, on the dev tooling side. So, the event is going to be on Friday, November 21st. Um it's a whole day event. So we start in the morning. It's at Lural which is kind of the main um venue um of Defcon connect. And uh as you see here we put like 804 and X42 in the new logo. So what happened is that

the X42 uh community wanted to do like um an event at um uh Dev Connect around the same time. We kind of invited them. So we are kind of hosting also like um people from the explor 2 community. Um I think these two communities are very aligned. Uh when Marco and I started talking about uh what then became a 2004 uh few months ago like in the early summer actually we were thinking about payments and then we looked into explor and we're like okay uh maybe we can use this protocol for payments. Let's solve kind of the hard problem of like um discovery and trust. And so since the beginning these two were

complimementaryary and uh we want to continue like essentially uh building this together. Um we are just sharing the um agenda for the first time. So we'll publish it u after uh the call but just like a quick um view uh we start at 9:30 a.m. with with registration. At the beginning we have a series of um uh talks which are kind of uh more um tech talks. So there will be like ch from deep mind talking about like mass gen which is this framework that they're building a couple of talks like focus on like cryptography

and security uh for agentic AI and then we continue with um a few uh product related talks. Uh there's going to be like some X42 in there, some crossover between like 804 X42. Um um yeah. So then um lunch break in the afternoon. Um we are going to have like again like some X42 talk like as we start in the afternoon and then like a lot of like product talks from like um ecosystem builder in the 804 ecosystem uh that will showcase like what they've been

building. Some of them have joined like the uh builder program that we ran. Um uh and then uh towards the end uh there will be um like a couple of panels and then we have a closing panel with a surprise guest that we'll announce um uh later. And um yeah, so this is uh uh the agenda. We'll publish it like if you are there please make sure you register. It's going to be a great day. Kind of a celebration of like this beginning of 2004. Like it's just the beginning.

>> Yeah, maybe I can pass it on to another. >> Sorry. Uh sorry about that. I was actually making some changes to my presentation and I was not finished yet. So now I'm finished. Okay, cool. So um let me go ahead and share my screen. Okay. Portion of screen. All right. Cool. So, um the talk that I'm going to be or what I'm going to be kind of really diving into is essentially the like trustless agent stack. And it's it's kind of like with um AI in general, like where we currently are today in terms of uh being

able to build and and develop products and agents, we're starting to kind of see a lot of standards and and you know, I would say agreed upon protocols that developers can follow. Now if you look at kind of like traditional the traditional AI space um you you see that we have like MCP and you see that we have now with Google they have A2A and now AP2 and in the crypto world we have like X42 and we have ERC 8004. So what I'm trying to do is kind of explore the entire trustless stack like what does the actual stack look like for building trustless agents. At the core it's definitely going to be 8004. Um, obviously X42 fits in there really, really well. Um, and then a couple of

things that we're building at IGEN we think also fit into the stack, but we want to expand beyond just the stuff that we're building and kind of look at what's being built not only at the protocol and standard level, but also at the product level outside of our company as well. But the original kind of like tweet that kind of got me thinking about this was this post here where I'm kind of like breaking down a very maybe basic look at what the trustless agent stack might look like. You have ERC8004 for identity and reputation. You have X42 for payments. And then you have two products that we're working on. IGEN AI for verifiable inference and IGEN compute for verifiable execution. Um, and when you start looking at the properties of a blockchain and why you can't traditionally bring regular

Verifiable AI and Deterministic Inference

inference on chain, you have properties that a blockchain has that obviously aren't available in traditional LLMs, LLM inference. You have uh deterministic execution. You have the ability to re-execute and verify the the state of the chain. Uh you have deter you have u I would say like pure functionality in the sense that you have identical outputs uh return identical outputs. Identical inputs return identical outputs. Um you have transparency and you have immutability. But with inference today none of these things exist. So you can't really just say I'm going to make an inference call and bring that result on chain. What you need are two things. You need deterministic execution. you need verifiability. So that's kind of what we what we've tried to build with uh IGEN

AI, which is essentially verifiable AI. And you can kind of play around with what this looks like in practice at this new website that I put together last week called deterministicinference.com. And you can go in and you can like compare traditional AI models with um what we offer with verifiable and deterministic inference. So you can kind of run any inference with a seed and you're guaranteed to get the same output and anyone in the world can also re-execute this and get the same exact output very similar to kind of how a blockchain works. Um and then the other thing that I wanted to dive into is the compute version uh the the compute part

Trusted Execution and Compute Verification

of this. It's essentially being able to kind kind of like run a agents verifiably in a trusted execution environment um with the ability to verify the code that's being run and the ability to kind of verify that the agent that you're saying is being run is actually like being run as opposed to you know it being run on kind of like some centralized server. So this is kind of some of the stuff that I'm putting together for the talk and I'll probably share this in a Twitter post kind of all together at some point, but I'm still kind of finalizing all of it. So that's kind of it for me.

>> Thank you very much. So looking forward to your tweet and to listening you on stage on Friday. And it's a miracle. We're perfectly on time with such a dense agenda. It's exactly 60 60 minutes now. Uh well, oh yes, we we had the Q&A but uh >> we can do it offline maybe unless >> we can do it in the group unless there are burning qu or we can also do that in person for who is around in Argentina next week.

>> Awesome. >> Have a good one. Thank you all. >> See you next week. >> Thank you guys. See you guys in Argentina. >> See you. Bye-bye.