The Ethereum Foundation and the Open Anonymity Project launched zkAPI on Ethereum Mainnet on October 1, turning an earlier zero-knowledge payment proposal into working infrastructure for metered APIs. According to the Ethereum Foundation’s official launch announcement, the system is initially positioned around AI inference, where persistent API keys can connect billing accounts with years of user requests. zkAPI separates the payment relationship from API usage by replacing persistent billing identity with privately funded usage credits and zero-knowledge authorization. The Foundation confirmed that the client, server and vault contracts are live on mainnet.
The release implements a design published earlier this year by Ethereum Foundation researcher Davide Crapis and Vitalik Buterin, which CryptoCurrencyMagazine previously examined as a zero-knowledge framework for anonymous AI API usage. The official Open Anonymity repository contains the client, server, indexing infrastructure and Ethereum settlement contracts behind the implementation. What was originally a research design has now reached a mainnet implementation, although the software remains experimental and adoption by external API providers is still unproven.
zkAPI Separates Billing From API Requests
In the current mainnet setup, users deposit USDC into an Ethereum vault while using ETH to pay transaction gas. The client converts that funded position into a private note and later generates Groth16 proofs showing that a valid balance can cover authorized usage. The proof does not reveal which public deposit funded the request, while nullifiers prevent the same private balance from being spent twice. Commitments and nullifiers are maintained using zero-knowledge cryptography, Poseidon hashing and a Merkle tree.
Crucially, ordinary API spending is not executed as an Ethereum transaction for every request. The zkAPI server verifies spending proofs off-chain and issues a short-lived API key with a defined dollar cap. The AI provider later produces a signed usage receipt, allowing the private balance to be charged for actual consumption. One authorization can cover an entire session, avoiding the latency and gas costs that would come from settling every model call directly on-chain. The vault contract handles proof verification around deposit, closure and escape functions rather than every API request.
The architecture also separates information between participants. In its direct runtime-key mode, the payment server sees that valid credit exists and the total session spend but does not receive prompts. The inference provider receives the prompt and response but does not learn which deposit or billing identity funded the session. Neither side independently holds the complete link between the payer and the AI conversation under the intended architecture. This payment separation adds another privacy-oriented application layer as Ethereum continues broader work involving zero-knowledge systems, privacy and AI verification.
Payment Privacy Stops Short of Full Anonymity
zkAPI does not make AI inference anonymous in every respect. The Ethereum Foundation explicitly warns that providers can still see IP addresses and other network metadata, while request timing may allow sessions to be correlated. Providers also receive the actual prompts required to run their models. Personal details, recurring writing patterns, project documents or reused conversation histories can therefore reconnect supposedly separate sessions even when their payments remain cryptographically unlinkable.
The public blockchain also remains visible. Ethereum records deposits, vault closures and withdrawals even though it does not reveal which API usage those funds purchased. Users seeking stronger network privacy would need an additional privacy layer such as Tor, while confidential execution would require infrastructure capable of hiding prompts from the inference provider itself. zkAPI solves a specific billing-linkage problem rather than combining payment privacy, network anonymity and confidential model execution into one system.
The model may also extend beyond human AI subscriptions. The Foundation lists blockchain RPC access, image and video generation, bandwidth services and machine-to-machine APIs as potential applications, making the system relevant to the emerging agent economy. Other payment architectures are approaching that problem differently, including Visa’s infrastructure for autonomous AI commerce and Crossmint’s programmable payment credentials for AI agents. zkAPI’s differentiating feature is that metered access can be funded without exposing a persistent payer identity to the service provider.
Its significance will ultimately depend on integration rather than the mainnet deployment alone. Providers must support the proof and receipt workflow, users must accept additional client infrastructure, and the experimental implementation still needs to demonstrate resilience under sustained production use. The launch establishes that privacy-preserving API billing can operate on Ethereum today, but broad adoption, production hardening and sustained usage remain separate questions from technical availability.
