Davide Crapis and Vitalik Buterin outline a solution for API metering that balances privacy, security, and efficiency, particularly in the context of AI inference with large language models (LLMs). Current APIs either require user authentication linked to real identities, risking privacy, or rely on on-chain payments, which are slow and expensive. The authors propose a system where users can deposit funds once and make multiple API calls without revealing their identity or linking requests to each other. The protocol utilizes Rate-Limit Nullifiers (RLN) to maintain user anonymity tied to their financial stake. A user deposits an amount, say 100 USDC, into a smart contract and can then make numerous requests. If they stay within limits, their identity remains unlinkable. The system also includes a refund mechanism that allows users to recover unused credits while ensuring that the server is protected against spam. The protocol's design allows for variable costs per API call, with refunds issued after each request, enabling users to maintain solvency. A dual staking mechanism adds another layer of accountability. Users deposit an RLN stake and a policy stake. The RLN stake can be claimed if double-spending occurs, while the policy stake can be burned if users violate terms of service. This structure prevents servers from profiting from false positives, as they cannot reclaim the burned stakes. Furthermore, the article introduces an alternative using homomorphic encryption for seamless updates on user credits without compromising privacy. This approach aims to enhance the overall integrity of API usage while protecting user identities.