The Ethereum Foundation has deployed zkAPI to Ethereum mainnet, converting a February zero-knowledge payment proposal co-authored by researcher Davide Crapis and co-founder Vitalik Buterin into a working implementation for private, prepaid access to AI and other metered API services.

The foundation announced the mainnet launch on Thursday. The Ethereum Foundation's dAI team built the system together with the Open Anonymity Project.

zkAPI addresses a specific operational problem: paying for API services normally requires an account and a billing identity, which ties consumption patterns to a person or organization. Under the new design, users deposit funds into an Ethereum vault. When they make API requests, they generate zero-knowledge proofs demonstrating they hold sufficient credit to cover the call, without revealing which specific deposits belong to them. The payment layer settles usage separately from the service request itself.

"Users deposit funds into an Ethereum vault, then use zero-knowledge proofs to prove they have enough credit to pay for API requests without revealing which deposits are theirs," Vittorio Rivabella, AI coordinator at the Ethereum Foundation's dAI team, wrote in the announcement post.

The immediate use case is AI inference. zkAPI issues short-lived API keys with predefined spending limits, so a user can send prompts directly to an AI provider while the payment circuit verifies entitlement to service. Alongside the mainnet deployment, the project released a local client, a software development kit and a browser-based AI chat implementation, giving developers the components needed to integrate the payment layer into existing application flows.

The design traces back to a February proposal by Crapis and Buterin for ZK-based API usage credits. That document sketched prepaid, anonymous-access credentials for metered services; the current release turns the spec into deployed Ethereum infrastructure.

The privacy model has defined limits, and the foundation is explicit about them. zkAPI does not hide prompt contents or network metadata from providers. A provider can still potentially link a user across sessions through IP addresses, request timing, or identifying information embedded in the prompts themselves. The zero-knowledge layer protects the payment side — who funded the credit and which vault deposits map to which user — rather than the communication channel.

That distinction matters for enterprise adoption. The system is well suited to metered services where billing anonymity is the requirement, such as purchasing AI inference without creating a vendor-side record of the buyer. It is not an anonymity tool for the request channel itself, and integrators deploying the SDK will need to pair it with their own network-layer protections if session linkage is a concern.

The launch also signals how the Ethereum Foundation's dAI team intends to position the protocol in the AI services market: as neutral payment infrastructure for machine-to-machine and human-to-machine API consumption. Prepaid vault credit with proof-based spending limits removes the need for API providers to run account management, invoicing or card-processing for anonymous customers, while the short-lived keys cap exposure to abuse and overspending.

For providers, accepting zkAPI credit shifts revenue collection on-chain and reduces identity-compliance surface on the payments side, though it leaves service-side data handling obligations untouched. For users, it offers a practical middle ground between fully identified API subscriptions and trust-based credit arrangements.

The codebase, local client, SDK and reference chat application are available now. Whether zkAPI moves beyond a research deployment into broad provider adoption will depend on whether major AI and API vendors choose to accept zero-knowledge payment credentials — a question the coming months of integrations should begin to answer.