Cedro: The Intelligent Layer for Crypto
Introduction:
Liquidity Provisioning (LPing) in DeFi has mostly been unprofitable. The Total Value Locked in DeFi sits around a humongous $94B with its peak at ~$194B. Uniswap is the de facto Liquidity Provisioning platform for DeFi with over $5B in TVL. A study reported that over half of the LPs in Uniswap are losing money due to Impermanent Loss. On top of this, Loss-Versus-Rebalancing (LVR) adds another layer of loss for them. This is due to the nature of LPing in on-chain protocols which lags behind the off-chain market. So whenever there’s a price action, it first affects the off-chain market. Now, the off-chain market is informed, while the on-chain market is uninformed. Due to this, actors with the information are able to arbitrage and earn profit at the expense of uninformed LPs. Because of the volatile and fast-paced nature of the web3 space, such events happen more often than usual which makes the scene better for informed actors at the expense of the uninformed actors i.e. on-chain LPs.
One way to mitigate the loss is simply to act as soon as the LP obtains the information. This is in line with the idea of Just-In-Time (JIT) Liquidity. JIT works by providing liquidity to a protocol only when it is needed so that the capital is more efficient and the LP has less exposure to the impermanent loss. However, there are complexities around this architecture, the main one being the responsiveness and information level of the LP. Hence, there’s a tradeoff between Informed Activity vs Profit. However, DeFi, and in general, LPs are mostly passive in nature. Hence, there’s a need for a verifiable intelligent layer for DeFi to help bring and retain maximum value for on-chain LPs.
We introduce The Intelligent Layer, a marketplace for Intelligent agents with varying risk profiles and LPs while utilizing EigenLayer to create a crytoeconomic coprocessor AVS for state aggregation and LayerZero to transfer assets across chains. This delegates the hassle of cross-chain UX to our platform, providing a true Chain Abstraction experience to the LPs.
Anyone can deploy their AI agents permissionlessly. For LPs, it means, instead of providing liquidity to a specific protocol or manually rotating the liquidity, they can choose an AI agent deployed by the operators that suits their risk-reward profile and deposit their liquidity on the AI agent’s on-chain vault. The AI agent provides an on-chain commitment to a predicted return based on the amount and time chosen by the LP. The LPs are guaranteed to receive the committed return because of the stake provided by the operator. The time variable is mandatory because a prediction is impossible in an unbounded time. Similarly, the accuracy of return prediction decays with the increasing value of time inputted because of the increasing number of unknowns that could take place between now and the end of the period. This number depends on the robustness of a particular AI agent which should be transparently communicated to the LPs. Most of such variables are delegated to the agent operator themselves to let them choose their place in the risk tolerance tradeoff spectrum.
There are several ways to operate the AI agent. The protocol provides a set of pre-configured AI modules which are essential building blocks for a complete AI agent. These include twitter sentiment analysis module, long and short term price chart analysis module, etc. A non-sophisticated instantiator can easily pick and plug these modules with varying weight to spin up their own AI agent. For a sophisticated instantiator, they can create their agent from scratch to generate maximum returns. The output from running the models for all the AI agents is posted on data availability layers like Celestia, EigenDA, Avail, etc. for historical and slashing proof purposes.
Another important component is the agent specific on-chain vault. The vault acts as a constraint mechanism for the agent since it’d be hard to keep them accountable without sacrificing capital efficiency in an unconstrained environment. These vaults have in-built whitelisted actions that the agent can perform. The list of such actions depends on the risk tolerance level defined by the agent during instantiation. So if the risk tolerance level of the agent is low, then the number of actions they can perform (i.e. the number of assets and the number of platforms usable) is more on the conservative side. This vault is where the LPs deposit their liquidity to. Due to the constraints in terms of actions performable and inability to withdraw funds arbitrarily, it is impossible for the agent to rug the LPs, as long as the underlying protocols are safe. Therefore, for a given time T, the stake that the agent ideally has to have in the system is equal to the yet-to-be fulfilled return promises plus a buffer instead of the whole deposited amount. This makes the system extremely capital efficient and affordable for agent instantiators.
At the end of the committed period, the LP is allowed to withdraw the expected funds from agent-specific vault. If the agent-specific vault doesn’t have enough funds, a portion of their stake is slashed to fill the hole. Since enough stake is available to be slashed in such condition, in the worst case, the LP gets most of their promised return by slashing the agent. It is important to note that, there’s a delta factor which gives the range for the committed return rather than a fixed value. This gives a small breathing room for the agent to account for the unknowns. This delta factor increases with increased risk tolerance of the agent and the period of LPing. On the other hand, any profit made in excess to the committed return is the profit for the agent.
The existence of such a slashing mechanism makes sure that the agent is economically aligned to provide an estimation for a return on the lower side, otherwise they risk getting slashed. Similarly, the existence of a marketplace of plethora of permissionless AI agents makes sure that a particular agent is incentivized to provide an estimation for a return on the higher side, otherwise they risk losing LPs to other agents. Such a situation creates an equilibrium for agents to provide a commitment for realistic return, while the premium charged by the agent would be a race to the bottom to attract as many LPs as possible.
DeFi has suffered a lot because of its reactive and isolated nature. By enabling on-chain actions based on the insights from both off-chain and on-chain data, Cedro establishes itself as a coordination protocol between off-chain and on-chain worlds. Maximization of capital efficiency and user experience was also the core motivation for the omnichain lending protocol that we were building. However, we realized that we need many levers to control in order to provide the maximum possible return to the users and this wasn’t possible with our previous product. Hence, observing the market dynamics and needs of the users, we decided to build a revamped Cedro.
Social Media and Channels
Discord | X | Medium | LinkedIn | Telegram
