Agent Credit

The second major feature group revolves around delegation management and user control. The user sets per-asset borrowing limits by calling approveDelegation on each variable debt token contract. Each asset on Aave has a distinct debt token, so the user can independently approve, say, 500 USDC and 0.1 WETH. The agent can only borrow from the assets that have been explicitly approved, even if the user's total collateral would support larger borrowing. The approval amount is expressed in raw units—for USDC with 6 decimals, 500 USDC becomes 500,000,000. The user can change an approval by calling approveDelegation again with a new amount, which replaces the previous value. To revoke delegation entirely, the user sets the amount to zero. This isolated, per-asset control ensures the agent cannot borrow unapproved tokens, giving the user fine-grained risk management.

The third feature group covers compatibility and integrations with agent frameworks. Agent Credit is designed to work with OpenClaw, where it can be installed as a skill, allowing the agent to borrow autonomously within the OpenClaw environment. It also integrates with Claude Code, enabling developers to run the scripts directly from a Claude Code session. Because the scripts are plain bash combined with Foundry's cast tool, they are compatible with any agent framework that can execute shell commands—no special dependencies required. Furthermore, Agent Credit combines naturally with Bankr, a DeFi automation platform. With Bankr, the agent can swap borrowed USDC for any token, bridge funds to another chain, or deploy capital into yield strategies. This integration transforms a simple credit line into a full autonomous DeFi toolkit, enabling complex multi-step operations.

The overall workflow of Agent Credit follows a clear two-party setup: the delegator (user) and the agent. The user first supplies collateral on Aave using the standard Aave interface. Then they locate the variable debt token addresses for the assets they wish to delegate—either from the Aave UI or from the provided deployments.md file. Using a block explorer like Etherscan, the user calls approveDelegation on each debt token, entering the agent's address and the maximum borrow amount in raw units. After that, the user sends a small amount of ETH to the agent's wallet to cover gas fees—a borrow on Base costs about $0.01. Once these steps are complete, the agent can run aave-borrow.sh with the desired asset symbol and amount. The script runs the safety checks, then submits the borrow transaction. The debt appears on the user's Aave dashboard, and the agent can repay using aave-repay.sh or the user can repay manually.

Concrete use cases demonstrate the power of this system. For self-funding agents, the agent borrows stablecoins as needed to pay for on-chain operations, eliminating manual transfers and keeping capital productive as collateral. In autonomous DCA scenarios, the agent borrows USDC periodically via credit delegation, then uses Bankr to swap into ETH, gradually building a position without any manual intervention. Gas self-sufficiency is another use case: an agent monitoring a position can borrow a tiny amount of WETH to cover its own gas fees when its balance runs low, ensuring uninterrupted operation. For on-demand liquidity, the agent accesses capital exactly when needed for trades or liquidity provision, avoiding idle capital. Borrow+swap combos allow the agent to borrow USDC, swap to a governance token, and participate in voting—all in one autonomous flow. The outcomes include reduced operational overhead, efficient capital utilization, and 24/7 automated DeFi participation.

Agent Credit is designed for developers building AI agents on EVM-compatible chains where Aave V2 or V3 is deployed, including Base, Ethereum, Polygon, and Arbitrum. The primary audience is DeFi users who already have collateral positions on Aave and wish to delegate borrowing power to automated agents for advanced strategies. The tech stack is minimal: bash scripts and Foundry's cast tool for Ethereum JSON-RPC interactions. The toolkit is open-source and free, with no paid plans. Users must have an existing Aave depositor position and be comfortable using block explorers to approve delegation. Safety recommendations include starting with small amounts, never approving type(uint256).max, monitoring health factor, revoking when idle, preferring stablecoins for borrowing, and testing on testnets first. In summary, Agent Credit provides the infrastructure for AI agents to become financially autonomous while preserving user control and safety, transforming Aave depositors into credit issuers for the next generation of decentralized automation.