AI Agent Custody and Fund Access - Dejan Radic | ResolverSys.com

Now, let's stay in the in the AI realm. So, so here we talked about the ways not to feed junk to AI uh in order not to get junk as an output. But let's look at a different application of AI and AI agents that seem very attractive. But there's also a big question of whether or not you should let those AI agents have custody over your assets given that they're full of hallucinations. So Dan Radic will uh walk us through this topic.

[Applause] Thank you for the warm introduction. Um well AI agents and access towards your crypto. Um ever since um we got algorithmic trading being there and accessing your funds and doing your fund management to you. We have been asking a lot of questions around it. uh in the last two years um I've been working on many I would say uh decentralized AI applications and accessing the funds and fully managing of them has always been a pretty I would say nasty question around it so we're going to talk about agents and custody like towardal aspects of this speech we're going to talk about authentication of AI agents regulatory aspects custody models market mechanisms and at the and multi- aent environments.

So, uh A2A protocol of Google came few weeks ago. So, what are AI agents? Um basically if we think about them and if we want to give them the access to crypto basically we want to enable the agent agentic finance. AI agents um basically have three components. LLMs, they have memory of the previous conversations for consistency of giving the answers and they have tools like web search, accessing the database and so on.

So when they make the trades altogether, what do we want to achieve? We want to achieve some level of explanability. So they need to explain us why they are doing these things. And if we as humans want or actually demand to have control over it, how much of our decisions should be embedded inside of the agentic workflow of making decisions and in general uh when we talk about LLMs, we we we have like many aspects of hallucinations happening and in case of AI agents managing funds, those can be irrational traits and misplaced traits. On the other end of the spectrum, we have a totally different defined concept of crypto custody either through mechanisms like private keys or API keys or centralized decentralized custody which we are going to talk about.

So it's about enabling the operations over your funds and basically when it comes to AI agents it's about them authenticating themselves towards the blockchains to access the funds. So we have regulatory aspects of these and we have the always the question on the custody where are the private keys how are they stored so basically that can be either a self custody or a third party custody. So regarding the authentication, we have two ways. Token based authentication through JSON web tokens or like API keys which are suitable for custodial systems and private keybased authentication which is based on digital signatures and novel concepts like agentbound tokens which are basically bas based for suitable uh suitability of the non-custodial applications. What are the most important aspects when when we talk about the managing of funds in this kind of context?

When an agent authenticates himself to the system either through providing the IP API key and doing some kind of I would say challenge response scheme or by signing certain messages and like relaying them to to any kind of blockchain system. Basically after this point of authentication comes the step of authorization and authorization means know like what can that agent do with our funds. So does he have the custody of all of the funds? Can he access only certain amounts of funds at certain points and so on. Now AI agents are still machines but they need machine level of authentication.

Usually when we do the I would say multifactor authentication it's either based for humans on something humans know something humans have or something humans are. So biometrics for particular usage of AI agents basically we need to know and we need to rely on something what they know. So AI agents do not have biometry. AI agents cannot possess physical things. So usually when talking about these schemes we are talking about requesting some kind of access to the resources then getting the access to it and then thinking about authorization further on.

Now we see that agents are considered to be pretty autonomous. We do not need need to have like high levels of controls over them. On the other hand they are not physical persons. And can we say that regulatory aspects are actually being brought to the face of light regarding them? And we are opening a lot of questions on responsibility.

So can we say that AI agents are in identity crisis? Uh who is responsible for their actions? Are the creators of the AI agents or the operators of them? So that's basically the the boil down question. when somebody wants to access the crypto funds the first thing being asked today from the regulatory standpoint is to do any kind of KYC KYB so that is also like high level legal requirement on the other hand if you want the utility of AI agents then some persons uh is going to be responsible towards all the actions of an agent um irresponsibly that can be any kind of thing that the agent does So we might introduce a new new thing towards AI agents from this standpoint.

And basically if we have a proof of humanood by doing KYC and KYB, why shouldn't AI agents have the proof of agenthood? So right now if you're a human, you would pass the capture test. In many cases, AI agents can still pass some of these, but in in general, it's about presenting yourself as somebody who has created an AI agent and that that AI agent can actually operate without any kind of obstacles fully autonomously. So, regulatory landscape on both AI and crypto is still being developed. EU is the first of course in all of these cases with AI act with Mika.

Now combining these two to be able to have operations of AI agents accessing the crypto funds is on a responsibility level is the next step for regulators. Now when AI agents access the funds and for example they are doing it towards the centralized exchange where we are holding the majority of liquidity we have the best performance of doing any kind of trades and so on. Usually we would access some concepts regarding fire blocks, coinbased custody and centralized exchanges and it can be either through u cold storage mechanisms hardware security models multi-party computation by fire blocks and so on and all of these are basically centralized systems in which we authenticate ourselves by using API keys or token based authentication schemes. In the case of MPC, basically a human can be a controlling point for some trades or some decisions that AI agents are doing. So basically it's a token based authentication where parts of the threshold signing can be done by somebody else.

On the other hand, if we want to give the private keys to crypto u access to to the AI agents, we open up a whole new level of questions. It can be a standard wallet where you just keep all the funds on a single address and execute trees. Further on it can be hardware wallet which can be in due in possession of a certain person who has created or who is operating the AI agent. An AI agent is requesting the signing mechanism which can be approved physically by a person if a person wants a control. On the other hand, we have proxy smart contrast.

For example, we want to be transparently open towards what the AI agent is doing to show that our AI agent is the best trading AI agent out there. And uh we want to execute all the trades fully transparently over the proxy smart contract by also implementing some kind of investment policy like you can spend 10,000 on a trade but you cannot invest in anything regarding Monero. if that's aligned with our values. So that can be done through proxy smart contracts or if we want multiple agents to um decide on what the next step would be that can be done with multi-IG wallet and of course account abstraction which is at first created for people to have abstraction around what we can do with smart contracts. In this case it perfectly aligns also with the things that AI agents can do.

Now for example when when we want to access the the liquidity of any kind of the system we are always thinking from two dimensions. Usually centralized exchanges have central limit order book or club which you access by authenticating yourselves through certain APIs and AI agents can choose on what kind of action to perform based on let's say liquidity depth based on previous trades based on the current price and so on. With the AMMs everything is transparent, fully visible like unis swap style a liquidity pool with like two or three or n assets inside where you know what's the difference of doing the trade there or some other place. On the other hand, we have request for quote systems which is suitable for longtail assets where we would ask certain solvers or market makers and so on to give us quotes regarding the executed uh trades that we want to like relay on chain and so on. So it's a question of what do we want to achieve?

The problems with clo is for example partial fill. The problems with AMM is the amount of liquidity in liquidity pool. The problem with um RFQ systems is for example, nobody answers on your trade. You I want to swap one bitcoin for $1,000. Nobody would go and give us an offer for that.

So basically, we need to think high level. Right now, if we want to create a holistic agent to access all of these, it's not that easy. We need to adapt to different systems. But today we have the option to use intents. Currently working on near intents which is a system where AI agents can actually access the funds by utilizing simple APIs and have the access to all of these which where trades are actually executed in behinds by by solvers and market makers.

Now single agent is cool but it has its own limitations. What about you know like mixture of experts multiple agents having their own domain expertise and so on. So we're entering a new new environment and in the previous talk um from origin trail we've seen a lot of things um being mentioned from this image in particular you can check the paper by scanning the QR code uh about the governing agent to agent economy of trust via progressive decentralization from January this year and it basically proposes the way how multiple agents can um execute the environment in a decentralized way where they can of course like possess some funds and stake some funds as a proof that they are going to do everything up to certain standards. On the other hand, all of them are having domain expertise. They have a reputation system with a decay of reputation through time and they have the ability to vote and with quadratic voting of allocating the votes towards different options in different amounts to actually run the system where they can decide between each other how these things can be done and for example by having multiple agents.

One is like doing the sentiment analysis. The other one technical analysis. Third one the fundamental analysis of the trades. Fourth one is searching for long tail options, ICOs, lounge pads and so on. We can conclude about for example making some trade at c certain point of day based on some kind of um I would say touchable information provided by agents.

So AI agents are about utility. We are using LLMs and then autonomy of agents because they are pretty like touchable. We can sense the way how they are doing things up to certain extent. It's hard to have full explanability on one side but on the other we really like certain things around them. On the other hand, crypto custody has always been about security.

It's always about having the procedures, processes about keeping the keys, about having layered protection and so on. So we can say that in security there is always a security utility tradeoff and these two are on a counter sides of the spectrum and basically by having these two authentication mechanisms which is like API based and private key based uh concepts we can access the funds by utilizing different kinds of aspects and they also align with different levels of custo custodiality I would say one is centralized systems and the other one is decentralized systems. If we want to balance out the control and autonomy of the agent for some swaps, some trades, some decisions, human can be responsible by for example signing an additional signature in a multi-IG or participating in a threshold signing of an MPC. custodial and non-custodial moments are pretty tied to the way of centralization decentralization spectrum that exists regarding these traits and um AI agents are still I would say not trusted enough so the balance of this control is going to be higher right now but as we move through time it's going to be more and more intents are pretty good abstraction mechanism of different kinds of market mechanisms that we are having out there to abstract out the complexities of uh placing the trades in different kinds of systems. And of course, if we want to differentiate on the different kinds of areas of expertise by the agents, we need to enhance the way how we we are thinking about agents from the standpoint of multi multi- aent environments.

But it opens up a lot of questions that were mentioned before on this stage today regarding reputation and decentralization. That would be it for this speech. [Applause] Any questions? No, there is one

just a non non serious question. Uh do you trust any of your financial operation to agents?

Right now it's on a level of giving advice about the options. So it's not fully autonomous. There is control about you know like kill switch decisions. So yeah it it is a pretty direct question and so far we are at the point where where everybody's thinking about the way how these things are done know and in general with with algo trading mechanisms you know why certain trades have been executed with AI agents you're just thinking from the level of explanability why they have placed certain things and why these things have happened in some kind of order and why these for example particular trades were directed for example liquidity wise and volumewise on some parts of the system. So basically it is a question on you know like when is the focal point of people actually start trusting these agents to to manage the funds.

So far from my standpoint I'm still waiting for the point where where the level of hallucination is being dropped down and uh we are seeing that happening every day

and having just a next question sorry uh having your uh experience in this like uh AI finance stuff uh what is your prediction like how much time years or months maybe should pass before people start trusting their

well uh to be honest um it boils down to the hardware level of support when we we see for example with every new chain every new um LLM coming out when it comes out they allocate a lot of power to it and this power at the beginning for early adopters they are getting really good results then everybody jumps towards this LLM and then after that they need to reduce these coefficients of the depth of the search to to and then system starts producing some hallucinations and so on. So from my standpoint it's about solving the hardware problem and from my from my view I feel that in next year basically we are going to have pretty good level of I would say accuracy regarding all the answers that we are getting and right now for example with with AI agents doing the crypto stuff usually the prompts are much longer than the answers. So it's about summarizing a lot of information from different sources and giving the proper answer. Right now systems are doing pretty job with it but till the point that we are 100% sure it is going to do a good good stuff regarding our funds most probably roughly a year. So that would be my prediction.

Yeah. Quick question. Um are there already examples of uh agents doing some significant volumes? uh I mean has you know somebody already like dedicated significant amount of uh funds and those like being uh seeing like a great let's say uh return of investment uh so far that I've heard of that that somebody has been allocated a lot of liquidity to an agent I haven't heard of right now definitely um there has been advancement with virtual with other protocols and basic basically um from terminals of the truth uh happened at the end of October last year. We are seeing much and much more shift from just shilling on Twitter towards you know like fully management of funds and at this point you know like early stage is always like couple of enthusiasts who put like couple grand just to test it out but so far not much liquidity has been put on an agent to do it autonomously.

issues.

Well, basically um we talked about different kinds of custodial mo models and about authentication. Authentication is still a layered thing. So you need layered level of protection of your private keys or an API key on or accessing the funds in some kind of way. For example, with these concepts that were mentioned here, you can for example limit out the ways how things are being done on chain and then based on authentication and giving the level of access to a particular agent, you can access these funds in a more secure way. So it depends on like what kind of security in this kind of case like it's combination of cryptographic and layer security which are basically Yeah.

So basically if you go towards private key authentication and seed phrases as a concept or pass phrases and so on basically you need to keep it either in hardware security module of a cloud or at your PC and so on. So you're depending on the security of the machine where the keys are.

Last but not least um agents are in open, how to say. Um, and u in the end you have the security breach going to asking agent to get your security keys and everything. You can have that uh as a problem because the other people build the agents for you that you use. Do you have some kind of of uh thinking about that? How to bypass this?

Well, for example, if you build an agent and you open it up with agent to agent protocol and somebody else can use it. Uh basically, you need to do it on a prompt level. So it's on a prompting side and it's about like having the final check of the amounts of I would say the accessed funds regarding like executing the trees further on this has been more from the standpoint of having the autonomous agent run in a safe environment by yourself in order to manage your funds. If you want to open up your agent towards the access by the others and then others can influence your agent based on certain I would say decision making aspects that's opening up a lot of questions regarding the system and that's why this slide was presented where like we need reputation mechanism to actually deal with agents because if if we like give a lot of influence to somebody uh basically he can access our funds. So it's it's as if for example you you meet somebody new and he tells you like invest in Solana you know like are you going to trust him but if you know him well and you know his expertise you're going to trust him.

So basically similar with AI agents, it's about having the reputation scheme that's general and where where basically we're going to have scoring system about AI agents and their own productivity, their own decision making and accuracy and so on.

Okay, leave it at that but ask a lot of questions afterwards.

Thank you.

Yes. Yes. Did you mean that uh a agents uh can become uh intentbased mechanisms actually solvers itself or uh will they use the intentbased mechanism that exist in the current market?

Both. So um if you if you create an agent to answer to other agents requests then you're a solver but if you're an agent who just wants to do a swap from this chain to this chain from this token to this token like that's straightforward thing like you just place it as a request and then you get the I would say the quotes and then basically somebody else is going to execute it for you. So you have couple of market making solvers architectures in behind that can happen on this side like it's about only accessing the funds through the API level of like an intentbased system.

Do you know any teams or projects currently working on this specific connection between intentbased mechanisms and AI agents?

Yes. Well, basically all the swapping systems right now have some API level access and right now I'm working on near intents which is enabling this functionality in precise

you mean near in by near protocol right

near protocols intent system yes

are you currently uh using it right or working to integrate their intent mechanism to your agent right

yes you can integrate it over the API there are a couple of ways you can do it thank you

thank you I think that was Last question. A big round of applause. And now for our last speaker for today, we have we have Defi Nikolola. If you don't know him, he's doing the same topic for three years now, and I bet he refreshed it to to catch on the the latest developments in DeFi. So, Nicola, tell us what's what's new.

[Applause] All right. Um, first of all, um, thank you for staying up till the till the final talk of the day and, uh, yeah, welcome welcome to to Belgrade. Hope you're having a great time so far and hope that you're going to enjoy this talk as well. Um my name is Nicola. Uh I'm coming from a block analytica team team behind uh risk team behind maker and um and dice stablecoin now also USDS.

Um and today we're basically going to talk about latest developments in DeFi. As uh told in the during the introduction, this is actually the topic I've been covering for the last three years. So yeah, it's going to be it's going to be interesting uh to actually compare the agenda from 24 and 23 uh which kind of can bring an insight of how at at what pace this whole indust industry uh is moving. So yeah, today is basically going to be like a comprehensive overview of like the latest developing trends in the D5 space. Um on this year's agenda, we're going to cover new protocol iterations.

This was also the topic last year as you can see, but here we're going to focus on the protocols uh that are relatively new coming in the last uh uh in during the last year. Uh but yeah, we're also going to mention some that were mentioned also last year but may may not be launched. Then we're going to head to um the D5 lending innovations. So to put a bit focus on um lending protocols and D5 lending space. Uh then we're going to move to how basically yield products have evolved in D5 recently.

Um and then we're going to head to uh dashboard that we actually at Block Analytic are uh developing. It's a stable coin dashboard covering ma all major default lending protocols on mainet. Um and as as the the last topic uh we are going to cover privacy in Ethereum which I think has unfortunately been um let's say sidelined or under discussed topic recently and this is um my attempt to to bring it up to the discussions on events like this like um this kind of conferences and community meetups. We're going to briefly uh go through a a recent um privacy project on Ethereum called called Privacy Pools. It's a relatively small as well, but yeah, it brings some uh new features on top of the well-known tornado cache.

Uh and then I I would like just to wrap up um by mentioning DI tools uh you may consider using nowadays while wandering through DeFi. Okay. So new protocol iterations. Let's start with the one coming from the InstaDub team. It's called Fluid and it's basically a new lending protocol that uh combines lending and the decentralized exchange.

uh they do it by something called smart collateral and smart depth. Now let's start f first with the smart smart collateral uh side. So basically uh let's say you as a user want to deposit some collateral to a lending protocol and you um basically say okay I don't care if this collateral is for example in rep BTC or Coinbase BTC. Uh and what fluid does is basically leverages that fact by using a pair of tokens as collateral um and then using it as a trading pair. Uh so that basically enables you as a deposit of of depositor of collateral to earn uh trading fees on top of the like the vanilla lending APR you can now uh earn on let's say a or compound or protocols like that.

Uh so let's go through through an a simple example. Let's say I put one rep BTC in fluid. Um and then a trader comes that wants sorry let's say I put 10 rep BTC in fluid and then a trader comes who wants to swap one Coinbase BTC for one rep BTC. This is done in a you know familiar fashion. Let's say unis swap like uh fashion.

Uh so now my collateral composition is changed from being 100% in rep BTC to um having nine rep BTC and one Coinbase BTC and that is fine uh that is fine by me right uh and in return I kind of get the the the swap fee from from each trader of course um so the similar mechanism can be applied to the depth side as well it's called smart depth and it's kind of a bit more interesting because you can again utilize your depth as a trading liquidity which is kind of cool and but the uh like the underlying principle is the same in a sense that a user um uh can say okay I don't really care if my depth is in let's say USDC or or USDT and the protocol basically leverages that by using that um depth in a form of uh token pair as a trading liquidity. Uh but the the let's say the protocol flow of actions is a bit different than on from the collateral side. So let's say uh I generate USDC depth, right? Um I get it on my wallet and get to use it however I however I want uh in DeFi. And then a trader comes who wants to swap USDT for USDC.

What the protocol does under the hood is basically um it uses USD uh it uses uh sorry the trader comes and wants to swap one USD uh one USDC for uh USDT. Um, and then the protocol basically uses the USDC to pay back uh my depth and then generates new depth in USDT and get it back to uh to to the trader minus of course the the swap fee. Um so yeah from from the UX perspective as a borrower you may end up um needing to pay your debt in like some um proportion of USDC and USDT. Uh so not only not 100% in USDC but yeah in the meantime you get um you are earning trading fees um which can effectively lower your borrowing uh borrowing rate. Um but yeah it's a again from the UX perspective this can be solved uh like one step before you actually pay uh back depth by swapping everything to USDC and then paying it back in USDC fully.

Uh then we of course have liquidity v2. It was uh already discussed last year here um but has not uh been uh launched. Now it's launched and actually relaunched. Um I forgot to mention at the beginning of the presentation uh feel free to scan those QR codes. Those are meant to provide some additional information if you're interesting in reading more about the protocols and the you know recent developments we are going to cover today.

Uh because you know it's impossible to to cover everything in 20 minutes. Um then we got Oiler V2 a new very modular uh version of lending protocol alongside with Oiler Swap which has been announced recently. Oiler swap is very similar uh to to fluid decks which we just explained. Um they they're basically keeping all of the liquidity in Oiler vaults and then using it as a just in time liquidity uh for trading on on oiler oiler swap. It's actually built on top of unis swap v4.

Um and of course a v4 uh it was introduced also last year. It's not still launched but yeah something uh some very interesting features are coming up. So yeah uh feel free to read more about it as well. Um when it comes to define lending specifically there are some of the things that I would I would say like in my opinion are expected to be um implemented in in newer versions of define lending. The first one being separating LP LPS into the trenches the junior and the senior trench.

Now this is actually a very popular thing uh in Treify right and has not this has not been implemented so far in defaing protocols. Uh it actually has been tried but yeah in my opinion I think u the the industry um did not was immature too immature at the time. So yeah, we'll see uh if this uh will work uh this time. Um my assumption would be that it would uh we'll speak more about this at the next slide. Next up, risk premiums.

This is actually also something already available in Trefify. Uh but not yet yet to be implemented let's say in uh uh on a smart contract level. the the the mechanism is basically you you pay higher borrow rate for uh borrowing against the riskier collateral right so it kind of makes sense uh and I think this is actually uh going to be uh introduced in uh in a v4 uh then fixed borrower rates so currently we have fixed yield thanks to pendle and some other protocols and those are based on uh maturity right you have expiry date and then you need to roll over if you want to if you want to keep like this um expected like predictable predict uh yield. Uh so I think this is the next thing we're probably going to see very soon that those are fixed borrowing rates. Um my assumption would it be that those are also going to be maturity based with kind of um uh automatic rollovers.

Um, next up, composite oracles. This is actually um risk management improvement that has been somewhat adopted so far among defi lending protocols, but I think it should be um like it should experience mass adoption because it's effectively like making lending and borrowing defy um significantly more secure. So, let's give it a few examples. um what protocols can implement of what protocols can we implement is basically upper bounds for stable coins. So let's say I mean most of the defend lending protocols are currently pricing uh the stable coins which are the most borrowed assets uh within the protocol by market price and this actually let's say if um for some reason uh some of the stable coins uh go price goes above $1 you can you can have uh those highly leveraged positions being liquidated you know due to their depth uh increase and this is something that can prevent that.

So the upper bounds for stable coins or LSTs then redemption or exchange rate with market fallbacks. Um for example, you can use exchange rate for a stake teeth and then um use a a fall back to a market price um based on some threshold for example a deviation between the exchange rate and the market price because you don't want to you don't want to use exchange rate um permanently if the crash is real if it's not temporary right you want to avoid bad debt and like this excessive borrowing then of course uh different provider fallbacks let's say you want to use chain link as your primary oracle and then um for example redstone as a as a fallback oracle or time weighted average price. Um then medians between various providers of of oracles. This is actually something if I'm not mistaken that sparkland is currently using and even some custom feeds for specific purposes. Now of course this uh would be a new code and you need to audit it but yeah if it fits the purpose um is you know highly advised to use it.

Uh we for example uh at block analytica did a custom oracle on base on on a morpho market uh for the lumbard BDC. So it's LBDC. Uh we used proof reserve feed from Redstone and combined it with um a market rate and of course had a threshold um being c certain deviation between those two and then u made it a batable. So we can so we can adjust the threshold based on market conditions. Of course, um the the custom feeds uh do not have to be of course developed by you or your team.

those for a good example of that is Pendle developing uh linear discontrac tokens that are now massively used throughout D5 as collateral which is a pretty pretty cool example of how you can um cleverly price assets in uh within your protocol and thus create this u create more capital efficient environment. Right? So basically those PT tokens are going up to to the level of one uh and then basically you can price them uh similarly so so so capital efficiency grows. Yeah. Then I want to talk about more about uh the yield products which I think uh is a very interesting thing because the segregation of riskreward as a product has has been tried previously in DeFi had has not to be honest seen much adoption but now we are seeing again this uh wave of um sort of dividing the higher risk and lower risk yield like the trenching trenching process.

A good example of that is um the proposed updated version of safety module of a protocol called umbrella. They effectively they're effectively adding trenches via uh staking a token a tokens which are like proof of deposit tokens. when you deposit to AB you receive let's say U a USDC or a USDT and when you stake those tokens um you basically can um cover uh I mean you're you're being exposed to bad dep risk from a currently this is done by staking a tokens and this is kind of an improvement in a sense that no tokens need to be sold when in the event of bad deb uh but rather a protocol will just burn a tokens that were previously staked by by stakers and in return stakers are receiving of course um boosted yield which is uh coming from a treasury itself and yeah u like vanilla deposits um basically another improvement um is that vanilla deposits like if you just deposit on a get a tokens and not do not stake it uh you get the same yield basically for less risk because now you are kind of your risk of um covering for the bad depth is uh is lower because the stakers are covering for you. Um another good example is product called summerfy which is uh which has launched the new protocol a very with a very simple UX where they aggregate yield vaults for and rebalance between those for boosting the APY. Um and they basically also have like the the high risk vaults of course offering higher yield uh and the lower risk risk vaults uh and yeah so far it has um seen like a constant growth.

So yeah, that's why I mentioned you know maybe uh I mean I want to just put draw attention of uh the importance of timing especially in in DeFi when you launch a certain type of of product also uh I think it's worth um noting that pendle pt token looping which we already discussed uh probably even last year I think it's now safe to say that um this kind strategy is sustainable that it's keeping up um due to various factors including um our ability to like keep the borrower rate always lower than the than the yield of the PT tokens which kind of makes the strategy profitable all the time. Another factor being that the one of the protocols tokenizing some kind of yield strategy like Athena uh adding a mechanism to rebalance between the basis trade like the core business model and and the yield from treasury bills. So it kind of makes the whole process more um more sustainable. Then you can also have uh projects like DeFi saver uh who can build additional features on top to ease the management of those kind of uh positions like the automatic roll over to the next maturity as as I mentioned uh before and then yeah this is um this is actually a stable coin dashboard we're uh currently building in block analytica I mean it's it's uh it's live and it's actually something we decided to build because uh we needed something like this uh as a users of defa landing protocols and CDP holders and leverage users uh it's uh called sphere and yeah uh you can basically find a lot of things there including the net APYs for leverage positions let's say you want to long ether or bitcoin um benchmark rates these benchmark rates are actually used by block analytica when deciding what should uh when deciding about the adjustment of the die savings rate or sky savings rate. U you can also estimate the unwinding price impact of your CDP.

Uh you can find lowest lowest borrowing rates in DeFi. Uh you can see the the supported protocols. You can find lowest liquidation prices and basically even find some trends if you if you uh dig deep. Um yeah, I think that should be it. Now the yeah the topic I also wanted to draw attention a bit and I think this project is actually pretty cool.

So privacy pools uh it's a new and non-custodial non-restrictive privacy protocol. You can see it as a as a cache but with one important addition and additional feature and that's zk proofs for uh which you can use to prove that you are not part of a certain uh data set or or list. um basically proving that you are not laundering money that you are not um hacker that stole that has stolen funds from some protocol or criminal or similar. Um important thing to mention is that this exclusion list so-called uh is not centralized like in tornado cache. Uh in other word there are no hardcoded like compliance rules.

Uh so anyone can basically um define their own list and then prove that they're not part of it. So let's uh go through a simple example. Let's say a centralized exchange for some regulatory reason needs to needs uh its users to prove uh that they are not um like malicious actors, right? So they can be define their own exclusion lists based on law enforcement data, sanctions list and basically require your proof that you're not part of part of that uh data set. Now on one end I mean it's obvious this is like a improvement um compared to tornado um but still uh criminals can still still use it for laundering money as it's completely permissionless and uh and open sourced.

Uh but yeah, they're kind of uh less incentivized to do it, right? Because if you if you want to do something with that money, you would kind of need also to provide a proof uh proof of exclusion from a certain certain list. And yeah, to to to wrap up, uh I just I'm just going to leave this here. Uh this is basic those are basically some tools um you could consider using nowadays while wondering through defi. So for transaction submission consider you know using private um private RPCs like me blocker and fleshboard RPCs.

Um of course simulate first before submitting a transaction for lending and borrowing. Yeah, you can use sphere to find the best stable coin rates and net APYs to minimize liquidation and front-end downtime risk. You can consider focusing on protocols of course with a proper risk management and defy saver automation which is pretty cool. Um, of course use multisig key rotation and similar and yeah um I think we'll have uh enough time for the Q&A session. So feel free also uh to ask any questions you may have and maybe even uh suggest some more tools that you may have uh so everyone can can hear.

Yeah, that will be it. Thank you for listening. [Applause]