Agent Credit

Product Introduction

1. Definition: Agent Credit is a decentralized finance (DeFi) infrastructure layer enabling AI agents to autonomously borrow and repay cryptocurrency via Aave’s credit delegation protocol. It functions as a non-custodial credit line where agents execute transactions against a human delegator’s collateralized Aave position.
2. Core Value Proposition: It solves agent liquidity constraints by providing on-demand access to DeFi capital without manual intervention. Delegators retain full control over collateral while agents borrow programmatically within predefined limits.

---

Main Features

1. Autonomous Borrowing Scripts (aave-borrow.sh):
   • How it works: Agents execute bash scripts with Foundry’s cast to call Aave’s borrow() function via the LendingPool contract. Borrowing requires the agent’s private key for transaction signing.
   • Safety protocols: Pre-borrow checks include per-transaction caps, delegation allowance sufficiency, delegator health factor (>1.5), and agent gas balance.
2. Credit Delegation Management (aave-status.sh):
   • Technical process: Queries Aave’s VariableDebtToken contracts using eth_call to verify borrowAllowance(delegator, agent) and userState[delegator] for real-time debt/health data. Supports --json output for agent parsing.
3. Multi-Chain Aave Integration:
   • Infrastructure: Preconfigured for Aave V2/V3 on Ethereum, Polygon, Arbitrum, and Base. Uses chain-specific contract addresses (e.g., LendingPool, VariableDebtToken_USDC) from deployments.md.
4. Repayment Automation (aave-repay.sh):
   • Mechanism: Agents repay debt via repay() calls. Supports partial repayments or max to clear full debt. Uses delegator’s underlying token balance (e.g., USDC) held in the agent’s wallet.

---

Problems Solved

1. Pain Point: AI agents lack self-funded operational liquidity, requiring manual top-ups for gas fees, swaps, or protocol interactions.
2. Target Audience:
   • Autonomous Agent Developers: Building AI systems needing DeFi interaction (e.g., trading bots, DeFi managers).
   • DeFi Power Users: Leveraging idle collateral to fund agent-based strategies like DCA or liquidity provisioning.
3. Use Cases:
   • Gas Self-Sufficiency: Agents borrow micro-amounts of WETH to pay transaction fees when low.
   • Autonomous DCA: Borrow USDC → swap to ETH via Bankr → deposit to savings protocol.
   • Cross-Chain Liquidity: Borrow stablecoin → bridge via Bankr → deploy on another chain.

---

Unique Advantages

1. Differentiation vs. Competitors: Unlike custodial solutions (e.g., centralized agent wallets), Agent Credit uses non-custodial Aave delegation, ensuring delegators retain collateral ownership and agents operate permissionlessly.
2. Key Innovation: Isolated debt token approvals combined with holistic borrowing capacity. Delegators approve specific assets/amounts (e.g., 500 USDC via VariableDebtToken_USDC), while borrowing power derives from total collateral LTV.

---

Frequently Asked Questions (FAQ)

1. How does Agent Credit ensure delegator collateral safety?
   Agents cannot access collateral directly. Borrowing requires explicit approveDelegation() per debt token, with hard limits enforced by on-chain checks (health factor, per-tx caps). Revocation is instant via approveDelegation(agent, 0).
2. Can Agent Credit work with non-EVM chains?
   Currently, it supports EVM-compatible chains with Aave V2/V3 deployments (e.g., Base, Polygon). Adding new chains requires updating deployments.md with contract addresses.
3. What happens if the agent’s borrow triggers liquidation?
   The aave-borrow.sh script blocks transactions if delegator’s health factor drops below minHealthFactor (default: 1.5). Real-time monitoring via aave-status.sh --health-only is recommended.
4. How do agents repay borrowed funds?
   Agents execute aave-repay.sh using their wallet balance of the borrowed asset (e.g., USDC). Repayments reduce the delegator’s debt position on Aave.
5. Is Agent Credit compatible with existing agent frameworks?
   Yes. Its bash/Foundry scripts integrate with OpenClaw, Claude Code, or any shell-executable environment. Outputs support JSON parsing for programmatic workflows.