Projects
![]()
No projects found
Investors
New
DAO Treasuries
DATs
Lend / Borrow APRs
Chains
Categories
Spaces
Build a trading bot using MCP
Recorded: April 3, 2025 Duration: 0:51:18Space Recording
Short Summary
In a recent workshop, the Medis DevRel team explored the future of Web3, emphasizing the importance of AI in portfolio management and addressing the barriers to crypto adoption. Pavel highlighted the need for innovative solutions to improve user experience and drive growth in the ecosystem, while also discussing the role of community governance in maintaining data integrity.
Full Transcription
And we are live. Amazing. Welcome everyone to another workshop with the Medis DevRel team.
Today I have the pleasure of hosting a special, I would say, episode with Pavel.
And I want to touch base with you right away, Pavel. So how are you doing first?
And second, what is it that you're going to be showing us today?
Yeah, yeah. Thanks, Victor, for the introduction. Yeah, yes, I'm doing pretty good. Just recently,
doing like a lot of flights and, you know, catching up with how like, you know, the world's been operating, touching grass, let's put it that way.
So today I wanna give a little bit of a demonstration
in regards to building a AI, let's say AI trading bot.
Of course, the concept of a trading bot
is that it performs trades on your behalf.
And you basically just tell it, just, hey, go with this type of strategy.
And then it manages that.
Now, this demonstration isn't necessarily a sophisticated example, but it's more of like a proof of concept, but also things to consider when you're looking at
building your own. And then we'll have certain solutions that we're working on and that we're
going to expand on in the future. So hopefully, you don't have to worry about those problems
about those problems if you encounter them.
if you encounter them.
Hey, and tell me a little bit more
before we get started with like the coding,
because we have people that are just turning in,
so it's still good to like give them
a little bit more of time.
I'm wondering what is like the motivation behind,
why did you decide to choose this particular topic
over other things?
Yeah, yeah.
So one of the big hurdles in Web3 adoption is essentially,
how do we get people to come on board?
And this is stuff that's been talked about across every single conference,
across every panel, across., is like, how do
we bring more people?
And the current solution that I think companies in this space are looking at is interoperability.
Interop does a few things well, like liquidity efficiency and all of those things.
Tons of projects are working on that interoperability.
But I think interoperability helps when you're already in the ecosystem.
But what can you do to bring more people on board?
And everyone's perception of crypto, when I spoke with them, is that crypto is just high risk and it's like a casino.
So you come in and you put your money on Vingus token and then you lose it all. And then you're like, wow, okay, well this sucks. And you never
come back again. And that's basically people's experience in Web3, but also in every casino ever.
And I think even casinos aren't doing too well right now
since, you know, crypto is kind of like
their main competitor now.
So in this case, like, you know,
one of the big factors for Web3 develop
or for people coming into this space
is that they have this perception that it's a casino
and that I'm gonna lose all my money. And that you have have like you know the deathly fear that if you sign a transaction
you'll lose all your money and it's just way too risky and it requires essentially a level of
knowledge to even enter the space and start using it uh start using the tools that are involved um
even you know i i don't know if any of you have asked,
you know, your grandma or anyone,
any relative to create a MetaMask account
or anything like that.
It's like, you know, you set it up,
you say, do not share your seed phrase to anyone
and do not sign any transactions you don't know.
And then it just becomes this sort of thing
where they never touch it ever again, because the fear of losing is very, very strongly correlated with the idea of trying new things.
So what does this all mean in relation to like AI trading? So when you go to a financial advisor, they typically recommend you based on your portfolio, well, based on your personal risk preferences.
So essentially, you have no idea what goes on in the financial side of things.
When you go to your portfolio manager, when you go to your finance manager, you basically say, here's my risk level. Here's
what I prefer. Like, here's the things that I want to invest in. Here's the things I believe in. And
you know, here's how much I'll like, you know, here's how much I'm willing to lose. Here's,
you know, everything related to your investment portfolio or your strategy. And then essentially
the investment manager or whoever is responsible for your account
is the one responsible for essentially making trades and making you money.
And this has been basically a system that everyone uses in the world, right?
Everyone is familiar with this sort of system.
You speak to your financial advisor.
I want to invest in blank.
And then they essentially diversify across many different
assets. So then crypto, it works in the same way, right? Except right now, it's very manual,
right? I have to sign, I have to click transaction, sign transaction, I have to investigate all the
APY, I have to investigate all the risks. So you are essentially your own financial advisor.
your own financial advisor. And that comes with big risks. So instead of just, instead of just,
you know, picking, picking and doing essentially what, what financial advisors are typically,
typically doing instead, why not offload some of that to an agent who is specifically designed
to perform these trades in a way where
it accounts for your risk tolerance. It accounts for different things, just like how a financial
advisor would do. But instead, you are communicating with an AI agent who, again,
acts based on the parameters that you set. And the idea is that you're not interacting with a wallet
you're not interacting with anything that you know you you're you're not you're you're essentially um
you can even talk to the agent as well you can you can use natural language to describe what you want
describe the things that that that you know are are your risk preferences. And as a result, the agent just
executes on them. And then just like how a financial advisor or like the portfolio manager
would execute in that. So I guess maybe this speech title would be more so like how to build
your own AI portfolio manager rather than like, necessarily because trading bots are maybe like a whole different segment of portfolio manager.
But there's a number of risks essentially.
And we want to minimize those risks for people to just enter in the system so that they don't feel like it's a casino.
Exactly. That's a great angle, Pavel. Thanks for the motivation. I think it's quite clear.
And I love it, really. And I think that it's perfect in sync with the rest of the workshops
that we have been doing, as probably you know, and just like the rest of our audience,
we have been exploring each.
Oh, did I cut off?
Maybe Victor cut out.
I don't know if I'm still alive.
Oh, okay. Everything is okay?
I think everything is okay now.
Hopefully, yeah.
Okay, so we're going to assume that we're going to be safe.
And just to not interrupt the workshop, I'm going to please ask you to share your screen
and get started.
But I was just connecting this with the things that we have been exploring in the past with the audit ai agent framework yeah absolutely so i guess
without further ado let's just get right into it um i'll just present my entire screen and let's do
and let's do screen one.
Why not?
So, okay.
Can you see my screen?
I guess building a port.
But I need to add.
Now we have.
Oh yeah, perfect.
Perfect, perfect, perfect.
Okay, perfect.
So let me just adjust the title
because I think that might be actually a little bit better to describe what we're trying to build.
I'll share this presentation with everyone by the end of it, but essentially it's a portfolio manager rather than a trading bot.
It's not as automated as a trading bot.
So what is a trading bot?
Well, in this case, portfolio manager. Well, portfolio manager, this bot that I designed
essentially pulls yield information
and executes a transaction using that information
that is provided.
So that's just as a very simple overview.
You can make this more complex, of course,
but just those two steps.
So this yield data, right? Like, like, the information that is requested from, let's say
that the AI is requesting, it wants to know essentially like, hey, how much are you earning?
Where can I earn this? You know, where is the contract addresses that you would interact with?
And then the the risk level associated with it. Now, of course, this is, where is the contract addresses that you would interact with, and then the the
risk level associated with it. Now, of course, this is a, this is also something that that regular
financial managers would look at, they would go for many different types of sources in order to
determine this, you know, their trading strategy. Now, there's significant problems with the yield data.
So the first thing, oh no, let's do a slideshow.
By the way, can you see my screen right now?
I can see your screen.
Right now you are on the slides, looks perfect.
Okay, okay.
I'll do it like a screen share rather than the other way.
But there's some problems with the yield data.
So first is that the yield data needs to be maintained.
So it needs to be accurate, it needs to be updated.
So there's already there,
that's like a huge maintenance nightmare
for people looking to create anything
that consumes this type of information.
And there's a significant risk with this because funds,
if you're not maintaining this type of data,
it can be lost, right?
Like if I say that my, you know,
the risk factor may be updated for the DeFi protocol
and then the agent was pulling information from that,
then it can execute a trade in which,
again, like same risk factors as you have
like when you're interacting.
You don't have enough information and therefore,
or the information that you're basing your data off of
or basing your transaction execution off of is not correct,
which will lead to, you know, this
potential for bad things to happen. So, and the final point is that, you know, why this is like
a big barrier, right, is that if, let's say, the data that you are maintaining just so happens to, you know, like there's an error and that causes a financial whatever, the money ends up being lost, you are legally responsible.
So, for example, if you were trying to start your own like fund manager, by the way, that's you would probably need to like have some like licensing, regulatory stuff. So if you've provided this service, this service for others,
you would need to have like government permission in order to do this. You need to have,
you know, abide by like money transmitter laws and, and so on, or like you need a license for that.
So this is like a really, really big barrier to entry and also like huge risks, right?
Maintaining and managing the data source, you know
because you're legally liable and, but you still need it.
You need this data source in order for this bot to function.
So taking a look at this idea, this concept is that
we've created the community ecosystem governance.
Now this has been live for, I would say more than four years, right? So we've seen a lot
of interactions to go through the system. So we actually have the top 10 most interactions
on snapshot across all protocols that are using snapshot.
So, so this has been like a really big achievement.
And then we're using it exclusively for sequencer governance and the
community verified projects program diving a bit deeper into the community
verified projects, because I think this is where like, you know, the, the,
the idea of maintaining and managing a repository of information,
such as like reliable trading data from different partners
where this comes into use.
So right now, the mainly like community verified projects
is mainly used to provide like marketing support
for projects entering the Metis ecosystem.
And the current barrier to entry is
essentially just an optimistic quorum. So projects just need to have like an 80% approval vote
in order to pass this first level of CDP. In the future, we're looking to take the same
process that we're using for sequencer governance and the community verified projects and sort of make it expand that into having more stringent requirements. So for this case,
to maintain this data. So like, for example, the project needs to have audits, the project needs
to have like a minimum amount of liquidity, the project needs to stake some amounts maybe to make
sure that they're providing, let's say, highly accurate yield data
about their protocol. So there's various different ways that this can be done.
But essentially, the idea is to maintain the integrity of the data that is supplied into
the AI agents. So this is a brief diagram on how it looks like.
So essentially there's a separation of concerns.
So I have the community ecosystem governance,
which approves a project, which gets put into MCP yield,
which I'll show very briefly,
but essentially MCP yield holds the data
that was approved through governance and essentially provides the data to the AI agents, to the AI clients.
I think here we need to make some clarifications and maybe go over from basic concepts just so that people are on the same page with us.
that people are on the same page with us.
What do you mean exactly when you say the term MCP
and then when you say the term MCP yield?
Yes, okay, okay.
So MCP is a model context protocol.
And it is, think of it like you have a information
that is constantly updated, right?
Like through this process.
So you have community ecosystem governance that sounds a little bit general, I'm not going to lie, but
what you're doing here in NCP Yale, maybe I'm not correct here, is just pulling data, right?
It's acting like a server for a database or what, for pulling data or what?
Exactly.
So if I were to, you know, there is a, by the way, I'll share the repo with everyone here, I guess, like later on with the presentation.
But this is an example of like an MCP server that I created.
The purpose of it is to take like different data sources from different providers.
So, for example, Aave was really easy.
Like I could just import directly like an SDK
that Aave provides and I can look
at the on-chain data directly.
So I have an RPC provider that pulls the data from on-chain.
And then as a result of them pulling data on-chain,
I know that this is the APY that is basically
that the smart contract is saying that it provides.
For Hercules, it's a little bit more difficult, right, to get the APY data, right?
So in this case, like, they are providing this, the link to get, like, all of the APY information.
So as a result, like, this needs to be formatted, this needs to be evaluated.
And because they are providing the information, right, I have to trust Hercules in this case, like because they are providing the API.
And then if, let's say, they decide one day, hey, I'm going to like, you know, boost my APY rates on this API.
There's nothing that, because you're trusting this single source of data,
there's no real way to verify it right at this moment.
And I'll show you how this can be changed, how this can be solved a little bit later.
But essentially, there's various different ways of pulling the data. So for example, like, in NetSwap, they have a sub graph, and this sub graph contains information
related to, you know, the different pools that are present in NetSwap. Now, again, Hercules,
the reason why they have their own API
is because they're using many different types
of yield pools as well.
Like I think there's like algebra, there's gamma,
like different types of yield generation strategies,
which are very hard to essentially say like,
here's the smart contract that says all of the information.
So they sort of have to compile it into one single API
for like easy consumption, essentially.
So yeah, this is essentially just an MCP
like at a very high level.
It essentially just standardizes the information
so that it can be then consumed by AI agents.
Does that make sense?
Yes, that makes sense.
Pavel, don't trust Verify.
Come on, bro.
This is our comments, live comments from our audience here.
But anyway, if you're just tuning into this,
we're going to be doing something very exciting today
with MCPs, with AI agents and whatnot.
So this is the main diagram.
And please make sure to subscribe to our YouTube channel
and make funny comments as our friend here.
Anyway, Pavel, let's continue with this.
Yeah, absolutely. So exactly. The whole goal is to not have to trust. In this way, the idea behind this is that you would take this data that is provided by these trusted DeFi protocols,
you would execute the trade, and that would go into your portfolio. Right.
And as a result, you know, what the true APY of the trade or of the interaction would be right
after some period of time, you can then see like a normalized range. If I were to like very beefy finance,
right? So if I were to very briefly just go into like beefy, for example, and then I go into like,
you know, Metis, like any one of these, right? I see a historical rate, you know, moving average,
I see a historical rate, moving average.
You're able to see this because they track how much money they earn directly from these strategies that they perform.
So in this way, at one point, it was 14% APY, but then it dropped off to 4%.
And then you can see like,
over time, you know, like, like, how, what it's trending towards, right. And this is, again, like information related to like, how much token distributions are occurring, like from,
from providing tokens into a pool, and then you get like yield farming rewards,
or this can be like trading fees, right? Like the more people trade, the more trading fees are
earned, and then you get like, some level of of returns from that now of course there's impermanent loss
but um but but uh you know i'll since this isn't necessarily like a trading uh present presentation
i won't mention that but i'll just briefly mention like, uh, there's, there's ways that you can calculate it.
And the way that Veefee does it is that, um, you execute the trade, you take a look at
your balance and you match it up.
Like, does this, does this match up with, with, with what I am earning over time?
And, uh, as a result, you know, we, we essentially can, okay, you know, when I pulled this information,
did I did this information match up with the actual APY that I have received? If there's a
huge difference, if there's like something where it's like a huge discrepancy,
then that goes again into governance. And then the governance would decide whether or not
that project would be, you know, continue to be included
into the MCP yield, let's say as a data source,
or if, you know, like the data source was evidently
manipulated and misrepresented in a way where it didn't actually provide the
returns that it said it would.
Yeah, so that's just the overview of this whole entire data.
But what is the eventual goal, right?
Like, again, the community ecosystem governance, the ideas that it's supposed to be replaced
by our future initiative, which is LASAE.
We're still developing this technology,
but the core idea is that LASAE
would provide data validation
as well as incentives to provide accurate data.
So that you don't, right now,
like, you know, the cg has like this this whole um
i i would say like it's difficult to distribute incentives for people actively participating
in governance now there's there's ways you can do it but these systems kind of need to be like
added on rather than being inherently within the system.
So the idea behind Lease.ai is that they would come bundled up with data validation,
making sure that the data that is inputted is correct, as well as providing incentives for the maintenance and the expansion of that data set. So that's the general overview of like how we plan to maintain the MCP yield
to provide the most accurate results for the custom AI agents, which would then execute trades
for projects. And again, like the more projects there are, the more trades you can do, the more opportunities you can make, you can,
the more flexibility you have to create the agents to make money in however you think it is.
So, you know, people, I've heard of people that like, you know, there's a concept called
looping, right?
Where you, where you can deposit tokens into like a lending, um, lending platform like Aave. And you can keep on redepositing those tokens,
essentially causing your,
well, you're never going to have like more
than what you put in,
but essentially you can have more exposure
towards a single asset by consistently,
like let's say I use my Metis as collateral and
then I take out USDC and then I use that USDC to buy more Metis. So in that case, you know, that's,
that's like, you know, super, um, you know, like, like a way to maybe like a really high risk
strategy to do, you know, that type of, that type of trade. But again, like, like those novel strategies that,
you know, maybe not so many people know of these AI agents would be able to
execute in a way where it would just make sense for them.
And they can simulate the transactions. They can understand, you know, what,
what is the end result to this whole entire operation.
And yeah, I think AI agents are better fitted for this
than actual people.
Yeah, so now that you've been all been waiting for,
you know, the demo.
So I have a couple of GitHub repositories here.
So the first is, I'll just look at is the mcp yield.
So this is something that I essentially created
just as a quick demonstration.
But the idea is that we have a number of different projects
that I've integrated.
I'll just show the code right here.
So we have different things, NetSwap is here.
NetSwap has a little bit of latency,
like the data source that I'm using.
I think it's good because it's pulling like every single pool
that NetSwap has and there's over like 900 pools.
So it's pulling like way too much data,
even though like some of those pools
don't contain like a lot of liquidity.
So I decided to just leave it out for this demo,
but we have a Hercules, we have Aave.
And the idea again is to take all the information
and put it into something that the AI agent
something that the AI agent can actually read.
can actually read.
The other one is the actual agents itself.
So you can head over to my repository.
There's an agent that I'm using
that will essentially execute these types of trades.
Now, and I'll show you like just a little bit of a brief demo, that will essentially execute these types of trades.
Now, and I'll show you like just a little bit
of a brief demo, but essentially it's using this,
the tools provided by, let's say the on-chain tools
provided by this repository to do it.
So essentially they had like a huge list
of like all of these Web3-based actions
that you can just take and implement
within your own application and it just works.
So why not use it?
Yeah, so let me just briefly show this live.
Why not? Let's show this live. Why not?
Let's do it live.
So what I'm doing in the meantime is sharing the links
that we have to the GitHub repositories.
I'm doing this on YouTube.
Hopefully, you will be able to see it on X as well.
And of course, we will be sharing more information and all of the material
with you early next week, probably. But in the meantime, you can really fork this and start
playing around because I saw the documentation is quite good. So let's keep going.
Yeah. Thank you, Victor. Yeah. Again like we want more people creating these unique agents, right?
Like agents that would interact
with a reliable data source.
So we're spending a lot of time
to think of ways to maintain that data source
so that you don't have to worry about that aspect
of like, how do we even like get the data
that we need to execute the trades?
So for example, like over here,
basically this code is taking this message that is created
and then it's executing this message.
And then basically going through, how does it do it?
So going through, it then executes this message using
Gemini and it essentially does the action
that you're requesting using the tools that it is provided.
So in this case, this is a check balance tool.
So if I, you know, what is the balance of this address for chain ID of 1088, which is Metis
and the token contract of 0x dead, dead, dead, dead,
which is the Metis token.
So if we execute this, you know, we'll get this.
What is the balance?
Let me just actually increase the size of the,
eh, too big, but maybe that's good.
So over here we have user,
oh, nevermind, let me scroll down.
What is the balance of this?
And then we basically get the information,
the parameters that are passed to the um to the tool
and then as a result you know you you get a response from the tool and then you get a response
from the from the ai agent so this is basically at a high level you know how this program works
is that you would type in natural language what you want.
And as a result, the tool will execute
exactly what you expect would happen.
So let's do another one.
Let's do the, over here is,
I have another example of like,
what is the highest lending yield opportunities
on Aave for Metis?
So if I were to run this,
what is the highest lending opportunities
for Aave on Metis?
As you can see like min APY is like,
what's the minimum APY that you want to get?
Want to filter by?
You know, what protocols do you want to filter by,
and then how many does it output.
And so it gives us basically everything for Aave,
and then it outputs all that information,
and then gives us, okay,
the highest lending yield opportunities for Aave
are USDC token, DAI and USDT, right?
Now there's additional like, let's say incentives
that Aave does give out like on the interface.
Actually, if I were to quickly,
were to quickly just like take a look at the, you know,
interface for Metis, you know, interface for Metis,
you know, like these are the APY rates that are present.
But as you can see, there's like the 5.36 that is also incentivized for the Metis side.
So in this case, like that will be added.
that will be added, let's put it that way.
Let's put it that way.
It was a little bit harder to access that data specifically for this demo.
But the idea is that all of these incentives are added
to basically calculate the exact APY that you would get.
Yeah, and one thing worth highlighting is that a lot of these things
that you can build really rely on your ability to
gather data right yeah yeah but but again like the the data like we're responsible let's put it this
way like like the um you know the the the idea is that uh through this process right we have um
Through this process, right,
we have the community ecosystem governance.
Through this process, the community is responsible
for evaluating the data that comes through.
Now, of course, we're looking at ways
to make the data collection and indexing
of every single DeFi protocol on Metis.
That's a big feat, right? You can't have just one
team, one organization manage this. So we need help from the community. And we're looking at
providing those sustainable incentives to get people to essentially create reliable data sources
over the course of time. So that at the end of the day, you don't have to think about
managing these data sources yourself. Because if you do, you know, you have to think not only
on managing those data sources, but also the legal liability in case those data sources
are not up to par. At least in this case,
if another entity is managing the data source and incentivizing the correct data,
then you can rest easy.
You can sleep at night knowing that,
hey, at least the data is being maintained.
Absolutely.
So continuing with the demo,
let's see if I can, okay.
So highest lending yield opportunity.
So it gives you the examples.
And finally, the actual action, right?
Deposit this much Metis on Aave
with the chain ID of 1088
and for the Metis token address, right?
So we're coming in and we're like, okay,
this is what we want to do.
We've evaluated our different options,
even though like Metis wasn't listed there.
It's just the easiest example to demonstrate.
But if we execute this transaction, it essentially, so it needs to go through like, if let's say there's no approval transaction.
So let's say I don't, I don't, I haven't approved the balance.
It'll do that.
If, you know, some, some other thing needs to be done, it'll do that. The idea is that the agent with using natural language,
you can execute any type of transaction
that would then result in whatever type of operation.
So you're okay, the transaction has been initiated,
you'll need to confirm the transaction.
The transaction hash is this.
So let's confirm this transaction
and I have like a past one over here,
but let me just go to that one.
And we see that, boom, 33 seconds ago.
It looks like, yeah, there it is.
So I deposited, you know, 0.0001 Metis
and then I got 0.0001 A Metis, right?
This is the value of the Metis that, well,
that equivalent for Aave essentially yield opportunities.
So, yeah, that basically is, you know, how it works.
These are just some examples in case
like my live demo didn't work,
but yeah, like I tested this previously
and over here there's like two transaction hashes right here.
So there's like a semicolon.
So there was two transaction hashes,
one transaction hash for the approval,
another transaction hash for the execution, and if you don't know again
like every single like like think of it like this every single interaction that you make
with the d5 protocol is basically at least two transactions like like like at least on evm based
chains and so that in and of itself like you need to approve
your balance to be used and then you need to send another transaction to actually execute that but
like the user experience to that is very bad and and you know what and and that's the way it is
that's kind of the way it is but with using ai you can abstract this complexity you can just get
it to do what you want through this natural language and as a result it'll handle all that
complexity and and the uh what is it the like nature of approvals and all of this all of this
stuff right so quick question for you there because do you think that people actually know what's happening?
Like, why do you have to have like two transactions
or, well, it's not two transactions,
but you know what I mean, right?
So the reason why is because whenever you want to interact
with the DeFi protocol,
let's say you need to give access to your money that's in your account to the contract.
And as a result, that execution, like, for example, if I'm Uniswap,
um like for example if i'm uniswap i want to swap you know one token to another token token a to
token b like the like you don't give uniswap you know a thousand tokens and then like uniswap
would then like at some point execute the transaction and give you back a thousand so you would essentially approve uniswap to use 1000 of token a and then as a result
the uniswap will have access to pull the tokens directly from your wallet, and then make the execution, make the swap,
and then give you token B in return.
So essentially...
This is because it's self-custodial, right?
Yes.
I think that that's the idea behind it,
and many people seem to forget about that.
But anyway, this is just a rant.
Let's continue, please.
But actually, quite relevant, actually, a little bit of a rant.
So this, like, let's say we've solved the data, right?
Or we're solving the data, but let's say the data problem is solved.
Now, let's say you want to run a business hosting these ai agents for other people currently right if you try to host
these ai agents as what is it portfolio managers and something happens right you as the server owner
owner are responsible are the custodian of the keys that have that that the user then you know
like like interacts with so how do how do i describe this it's like um you know if it if if i
i had a birthday and then i gave a cake to a random person and said like,
hold on to this.
And then like,
you know,
the,
the,
and then you come back for the cake and then you see that the cake is gone or
something like,
cause the person ate it.
I don't know.
But what you know is the cake is gone.
You gave it to him.
The cake is gone.
You know, that person was responsible for the cake that i gave to them and now that it's gone i can go and i can sue them
or i can you know do do do whatever it is like they are legally in a sense responsible for whatever happens to that cake. So in this case, the main reason why, you know, this DeFi processes exist, right?
Why DEXs aren't considered like custodians is because they don't actually hold your money,
right?
You are using them to interact with pools that are set up in this way. Everyone has access to their own
funds and it's based on the protocol in this way, on this execution. So why this long ramp?
If I decide, hey, I want to set up this like a portfolio manager as a service,
a portfolio manager as a service,
the server that I used to set it up on,
let's say like AWS or Azure or whatnot,
technically speaking,
if you have full access to the AWS server,
which then contains the private keys of the user,
because in order to make these on-chain actions,
you need to have some sort of private keys.
So if you can access the server that contains the private keys,
you are a custodian in that case.
You have access to another person's bank account.
You have access to another person's wallet.
That makes you a custodian, which makes you legally liable.
And that requires money transmission license, you know, and all of these types of things, right?
Regulatory side, you don't want to mess with.
That's like scary.
So there are solutions that exist, namely trusted execution environments, which are basically like hardware based restrictions that basically the server owner cannot access the information inside of the server.
So you can spin it up for someone else, but you yourself cannot access the information
in it. So that's one. And then the other way is like ZK proving, right? Like the client runs
some type of execution, but you might not know exactly what as a result, you know,
since you don't know exactly what, but you do know that it ran correctly, but you don't know
what the private keys or anything related to that was, that you're in the clear as well.
So this, let's say the trading bots can only work right now if you are the one that's running
it, right?
If you are not the one that's running it, then it is, let's say, custodial.
And at that point, you're legally liable. So just wanted to point that out there.
But if, again, you implement a trusted execution environment, you can essentially host it for someone else without actually being able to
access their private keys. So hopefully, you know, that, that, that distinction, you know,
like, like that sort of helps with, you know, people's ability to think about, think and reason
about these problems. Like, how do we use AI to achieve
the things that we're doing just at scale, right?
And something to think about is,
how do we do this at scale,
but without exposing ourselves to additional risk
for building this?
So keeping this, hopefully that was short and sweet,
but definitely would be open to take any questions.
I don't see any questions right now.
I've been actually looking into the comments
so we can move forward because in 10 minutes,
we actually have to jump off into a nice AMA
with our dev rails on the Hyperion side for many.
So I don't know.
How do you feel about it?
Because maybe I can ask some more questions as well,
if you want.
Yeah, absolutely.
If you have any questions about this or or like anything
related to like related to this uh yeah of course so i'll i'll i'll ask you about the future where
is all of this going like we have the ai agents we have other components you mentioned tees and ck provers where is the
combination of these things going with ai agents and managing your portfolio how does that look
like in i don't know 10 years or whatever yeah yeah the the end result is that the ux is improved
is that the ux is improved right we in uh people that are in crypto right now right are in crypto
because primarily because it gives them opportunities that they don't have in the
location that they're in right that? That's how I see it.
Like people go into crypto
because they want to make tons of money, casinos,
but also people go in
because it provides them that opportunity,
opportunity for growth
and opportunity to essentially,
to not be bound by their physical location.
So giving this ability, using these AI agencies,
verifiable execution, using all of these tools,
using the data sources that are also verified
and making sure that everything is,
is the whole supply chain of of this this information
and execution is is uh managed and maintained accurately then i see this replacing wallet
this right this itself is meant to make on-chain interactions not feel like they're on-chain interactions.
And that's the end goal.
We want to make sure that anybody can use this because anybody can.
And you know what?
Web2, so the websites that we know that are built today, every single one has a way to as like a capture, right?
Like rate limits.
I can't access certain sites on a VPN.
I can't.
I can't.
Like if I am a robot, if I'm an AI agent, I cannot interact with the web as if I were a real person it just was not built for bots for
good reason because there's so many there's spam and do everything i think like email bot prevention
was like the first thing that that came out but if something is not built for bots but then we have
crypto we have these rails that are essentially built for everyone
and essentially it's indiscriminate as long as you have x token you can deposit it into y pool
and as a result you have crypto is built for ai agents there's nothing preventing them from maximizing value in this way. So I think that
the interactions that we see, the future interactions that we see, is that yes,
you'll have your savings account as maybe your hardware wallet, your ledger or trezor,
but your hot wallet or the wallet that you do all your trading on will be with an AI agent.
And as a result, you'll have a better experience doing the things that you're already doing in the Web2 space, especially within the case of financial planning.
in the web2 space, especially within the case of financial planning.
Absolutely, Pavel.
I think that you summarized in a pretty clear and very encouraging way
how the future with all of these technologies might look like,
and I can't wait to see it happen.
Thank you so much for joining us today.
To the people in our audience, in about five to seven minutes,
we are going to be starting at Twitter Spaces spaces in which we will have an AMA so
that you can ask question about the upcoming launch with Hyperion.
And if you might have other questions about, you know, the things that we have
been exploring with LAS AI and Alid as well, feel free to join.
We're going to have some fun.
And of course, next Thursday at the same time, we're going to have another
workshop so that we
can we can keep learning together thank you so much don't forget to subscribe bye bye guys and
bye bye Pavel thank you thank you Victor thanks to everyone
