Overcoming Complexities while Building the AI Agent App Store - Aleksandr Kuperman

Transcript

[Music] Hi everybody, my name is Alexander Coopermanman. Um, thanks for the introduction. I'm a senior smart contract developer at Valerie building on the list and I'm super excited to talk about overcoming complexities while building the AI agent app store called Pearl. Things we're going to cover today include understanding of what is an agent of all stack is a novel concept called proof agent things that are involved in protocol inflation and bridging when doing agent staking um integration with the Pearl app. Then we're going to touch base on um

successfully run uh apps on uh the Pearl platform and also talk about incentivizations for developers for the OLS accelerated program to incorporate deps into Pearl. All agents are essentially programs run on chain computation wise um computing stuff communicating on chain with specific set of parameters uh and um skills represented by NFTbacked u metadata configuration.

Agents have access to required resources and are defined using the following key parameters. Financial account obstruction with safe. Obviously, we need finances. Agents also need them. No change there. So, that enables autonomously pay for uh specific services and receive payments. Also, devs and users can set budget limits and uh guardrails. Uh everybody who worked with safe contracts knows that there are a lot of things they can embed using the contract into interaction and libraries.

The agent registry is the onchain set of contracts that meets and lists u all agents. Um basically agent get verification by their ID and the configuration metadata. Um embedded in the in the NFT. Um also uh here for the agents we will talk about the abstracted payments in Mac marketplace. So agents can access services programmatically and whatever payment is there the type of payment it just works and we build our stack such that agents are curated um for the use cases uh all the developers can build on

or pick up fork from and continue developing. There are several types of agents. Sovereign agents, those little buddies here, lightweight agents, anybody can run locally or in the cloud. Um, usually run by one operator. There are also decentralized agents. Uh, those are highly transparent because they are um synchronizing via the uh consensus run by multiple separate operators um and ultimately designed to to provide more workload from different areas.

The set of interactions between different types of agents is what we used to define agent economist with. Agent economist basically uh utilize different sets and types of agents to uh specialize in different areas and tasks. They manipulate u and interact with each other, produce meaningful results, hopefully profitable, hopefully. Um so agents function as market participants. They can collaborate. They can compete with each other. Um and supply demand dynamics basically uh dictates economic uh incentives and and shape the

outcomes. All has created the largest agent economies in crypto to date. Um there's been more than 7 million transactions by agents out of which more than 4 million transactions are from agents to agents. Uh all protocol is deployed on uh nine chains in counting not only EVM ones. Um there are more than 2k agents deployed by operators and um 60 plus types of agents registered uh by builders. Um all of this also forced us to define a so-called DAS daily active agents that represent a number of

active agents with a seven uh day uh activity agent activity. Agent economies interle on all protocol such that one agent um uh is able to provide input output for another agent outputs input etc. the dynamics keeps uh going by the define mech marketplace. Um max are basically entities that serve requests that come from developers and or users and or AI agents. Um and so when posting a specific request for a specific tool to

be executed the um users and or agents they provide a fee and Max compete in the marketplace um in order to um execute the request and get the fee. And so the OS protocol is a coordination agent for autonomous agents um to jumpst start and run the agent economies. Right now there are six roles in the protocol builders. They provide useful code for the protocol. They earn dev rewards for that. Um and also that goes that that goes go that code goes uh

towards agent utilization that uh helps define a economies types of um types of agents uh for various purposes. There are bonders they provide additional liquidity to to the protocol. Launchers here I would like to consider a little in more detail. Launchers define so-called staking contracts that shape agent economies. Staking contracts are basically entities where operators are able to stake their agents to perform specific KPIs in order to earn rewards. So it's like coin staking

but with some meaningful workload and it's done by AI agents basically. Um then there are contributors. They earn points and now also rewards via staking. Uh this helps us to promote the the network and the protocol. And there are governors. Last but not least, um they lock uh protocol tokens in order to get um voting power to vote for different incentives and also for um obviously taking contracts that need to be subsidized by the protocol. And so basically uh everything runs around uh the utility token called Olas that

drives the protocol from governing decision making on the protocol itself to understanding which taking contracts need to get inflation um and be distributed onto defined staking contracts such that those can perform in Asian economies uh with mic APIs and um basically earn earn rewards. We call the agent activity paradigm as proof of activation.

Basically that gives the proof that a agent is indeed active and also in principle is economically in uh with the protocol objectives um and or other economies. um self- sustained and and wellsupported economies can obviously uh jump start their own economies and that flywheel of economies keeps on keeps on going um can go in finite number of layers deep and even come back to the primary u age economy where it was

launched. Um so here's a little bit more technical stuff about minting and bridging associated with uh agent staking. So since everything needs to work seamlessly and autonomously, the only starting point is uh governors having voting tokens to vote for specific staking contracts and after that the inflation is correspondingly um minted to subsidize staking contracts. Uh since the majority of those sit on different chains

um outside of the governing chain then um the connection goes via different set of contracts uh ultimately leading to bridging and bridging is something very interesting. Uh there are a lot of providers different providers of bridging uh both on the app layer and the um uh more low-level layer. However, we found that uh it's better to use native bridging at least you know the the protocols providing bridging capabilities they they care about that one and uh even though this is the case we still need to take take care about several fault toal tolerance

cases um for example messages u arrive to to uh L2s faster than uh funds what happens in this regard. What if something breaks on the way? So all those things need to to be taken care of and uh hopefully they are. Actually, I'm very excited everything about bridging and messaging. So if you have any questions uh about your protocols, etc. uh come talk to me after the presentation. I'd be very glad about that. Um and so staking contracts define KPIs that coordinate agent economies. uh agents are staked onto the staking contracts. They either perform

KPI and they earn rewards or they pass on. They don't they get kicked out. Well, not necessarily right away, but it's up to the design of staking contracts, but obviously nobody wants to keep workers that don't do anything. Um this all sounds nice and fun but at the same time KPIs are not easily accessible to the general public right there's a lot of technical stack there is a synchronous stuff to to take care of there system requirements etc etc can somebody just do this for me please yes we can um we introduced the agent app store called

pearl It's basically designed in a web two UX where everybody is able to just um get the uh AI agent working for them for participating in agent economies performing KPIs earning rewards and so on. So in addition to owning your keys, you can now own AI agent and uh participate in in the proof of of active agent uh paradigm that that controls uh activities bringing values to uh agent economies.

Pearl is hopefully a collection to the um to to to this stack that uh is being developed today and currently uh enhanced. It's based on three key principles. Let me please read those. On your keys, on your coins. This is a well-known one, but with Pearl in addition to that one, I've mentioned that already. Um you are also um owning your uh assets as AI agents own your weights own your brain.

Agents reflect your preferences and strategies. The models are part of your personal stack and agents also learn on on that stack and and keep keep on enhancing their performance on your code on your destiny. This one is very important because through open service definition of program programmable autonomy users shape and co-own the AI services of tomorrow. Um as you know lots of companies they uh prefer to keep their AI models private. We believe it's it's not the case. The same is for decentralized agent uh assets. And so we believe that um the AI

power better to be co-owned as well. Let's talk about some of the examples that currently operate on the pearl ecosystem. Uh first one is all predict. Basically there is an interaction of different agents. The prediction agent looks for uh prediction market opportunities. Um other agents make intelligent predictions. Um then the uh max provide smart bets to traders um use using the marketplace basically.

So traders can get get a quote and get uh requests executed about about possible trades. If everything goes well, they earn rewards. If not, oh well at least but at least you're on rewards also from staking an agent, right? That does um uh acts with a um prediction markets. O predict is largest aged economy. uh it's uh out of um all the for example Gnosis chain transactions uh 50% plus are executed by OS predict and so there are five roles market creators they observe markets create u create

prediction markets at capital uh market closers symmetrically close those collect um rewards uh prediction markets um obs observe prediction markets and um uh provided information for researchers which are mechs. Those use AI tools that u provide um provide provide possibilities for predictions and traders uh obtain those in order to to make a decision whether to bet on something or or not. Another successfully integrated app is called agents on fun. It

started as a launch pad for coins at some point where um I mean it's it's a well-known concept right now right still in high demand uh everybody can uh summon their their coin um then uh heart it then unleash at some point um however this app quickly morphed into the AI influencer interaction app where agents just you know have fun uh react um on each other's tweets u launch coins as well um interact uh with different

messages basically having fun everything like average degenerates would do but uh just that they do it for us right and we own them and we earn rewards by them doing so and hopefully providing also some um interesting insights and um launching coins. I mean let's see what happens. Another example is the modus agent. This is a personal personalized portfolio manager that operates on uh the mode network. basically the AI agent that tracks your portfolio using several DEX

resources and obviously the idea there is to keep the portfolio growing. Um this is an example of how the modus portfolio manager looks like. U the assets um and the the uh DEX is basically used also still under active development but there is the motor strategy control where users or agents are able to provide the feedback to to the motivist agents for example change the strategy using such and such parameters etc. and modus strives to adjust the strategy according to that. Um another agent which is under development right

now Optimus it's that agent uh looks like modus and it really is it just operates on more network networks on mode optimism base and probably more um also manage actively manages your uh personal portfolio with the idea to uh to to perform better on the markets. Um, one thing I wanted to mention about Pearl agent app store is that um, I've mentioned that we try to to create the web two like UX but also Pearl abstracts

all the crypto complexities from the users. They just need to know about I don't know how to send coins from from one address to another. uh because we try to take care of of other crypto complexities like bridging uh like wallet management uh ramping on while basically maintaining the the self-custody um and user just needs to provide initial funds uh in basically the quantities that are written on a specified address and u pearl takes care of everything else including bridging.

At the same time also I wanted to mention here that bridging is a very sensitive topic. There are a lot of literally like we're working on it right now. Um there are a lot of bridging providers uh even more v APIs but again always have a plan to fall back to native bridging because you never know like services tend to crash sometimes RPC issues um um I don't know somebody logged off the server by by mistake etc etc um and so we invite you to download Pearl play with it and for developers specifically ally to integrate and in order to incentivize developers

um all those protocol attributes um funds uh to build their depths and in integrating in the uh Pearl ecosystem in order to increase uh and enhance uh a economies and Um, here's for the co-owned AI hopefully bright future. Thanks so much. Uh, let me know if you have any questions if we have time for [Applause] that. Uh, I took a look at the website and apparently there is no Linux version

of Pearl. Why is that? There's no Linux version. All right. It's Well, how many people will build from source? Sure. I don't like my laptop to be like, you know, like to build myself because because I I prefer some packages, you know. Yeah. And uh second question, what about computational power? Okay. uh for the agents. Uh yeah, if I you say I run I run them basically myself, right? So I need some computational power for uh for this. I mean it's not a sometimes it is a right

because if it if it can post Twitter so it it requires normally talking about links actually uh some of our developers run them even on Arduino. So um I mean not so much comput power uh but basically all the current um you know systems they they they support it's probably some interaction is going to go slower obviously with with the resources. Uh the stack is written on the offchain stack is written mostly on Python and the onchain stack is basically provided by your internet connection RPC. So that also tends to misbehave sometimes, but uh we're actively looking

for those solutions. Okay. Okay. We have uh time for another question if someone has one. No. Okay. Thank you so much. It was Yep. No, it was a lovely talk. Thank you so much. Please round of applause. [Applause] [Music]