EarnYld

EarnYld is an autonomous DeFi yield layer where AI agents act as personalized, on-chain quant managers. Unlike traditional yield aggregators, EarnYld agents are represented as Intelligent NFTs (ERC-7857), encapsulating strategy ownership, execution permissions, and a persistent, decentralized "on-chain memory."

Every agent continuously scans thousands of Uniswap v4 pools across 18+ chains. It doesn't just look at nominal APY; it runs a complex 9-dimension scorecard—evaluating TVL depth, adverse selection risk, hook safety, and gas break-even thresholds. Decisions are made through a sophisticated Seeker-Critic LLM pipeline: a Seeker agent proposes optimal rebalances based on market regime labels, while a Critic agent stress-tests the proposal against volatility and Impermanent Loss (IL) risks.

The core innovation lies in the 0G Network integration. Agent decisions are not ephemeral; they are written to 0G Decentralized Storage (KV), allowing agents to build a long-term "experience" record. Furthermore, high-integrity inference is handled via 0G Compute, ensuring that the agent's "brain" is decentralized and verifiable. EarnYld turns DeFi from a manual chore into a seamless, autonomous experience where your iNFT works for you 24/7.

EarnYld is built as a TypeScript/Solidity monorepo designed for high-integrity autonomous execution.

Smart Contract Layer: We implemented ERC-7857 (Intelligent NFTs) to bridge AI state with on-chain ownership. The iNFT holds a storageUri pointing to the agent's configuration on 0G and manages delegation to execution services. The Marketplace enables trustless leasing and trading of these yield-generating identities.

AI Architecture (Seeker-Critic): The agent runtime is a high-concurrency TypeScript service. We built a custom InferenceProvider to route LLM calls through 0G Compute (Sealed Inference) via the 0G Serving Broker. This ensures that the agent's "brain" is decentralized and the inference is verifiable.

Memory & RAG: We integrated 0G KV Storage to persist decision records and reflection logs. By using similarity-based retrieval on past 0G records, agents "learn" from previous market regimes (e.g., high volatility or sideways trends) to improve future risk-adjusted yield accuracy.

Yield Discovery & Execution: We utilized Uniswap v4 for multi-chain liquidity scanning, specifically analyzing v4 hooks for potential yield-dilution risks. Reliable execution triggers are handled via KeeperHub, which monitors agent signals to trigger on-chain rebalances. The frontend is a Next.js 14 dashboard using RainbowKit and TanStack Query, providing real-time SVG-based performance visualizations.