The Ethereum Foundation and the Open Anonymity Project have launched zkAPI on Ethereum mainnet, a system that lets users pay for metered APIs, including AI services, without linking usage to a billing identity, according to an October 1 Foundation blog post. The launch separates two roles that account-based API billing normally combines: the party paying for the service and the party making each request.

How It Works

Users deposit credits into an Ethereum vault, then authorize spending with zero-knowledge proofs generated on their own devices. A proof verifies that a funded balance covers the charge without identifying the deposit or the user behind it.

In its runtime-key mode, a payment server checks the proof and issues a short-lived API key with a spending limit. Prompts then travel directly from the user's device to the AI provider. When the key expires, a signed usage receipt determines the charge against the private balance. Under this mode, the payment server does not see prompt content, while the AI provider sees requests without learning the billing identity behind the key, the Foundation said. That split means no single intermediary in the flow holds both the content of a request and the identity of the payer.

A simpler proxy mode relays requests through the zkAPI server, allowing that intermediary to see traffic — the trade-off for its simpler setup.

Cryptographic Design

The system uses Groth16 proofs on the BN254 curve, Poseidon hashes, and a 32-level Merkle tree. Nullifiers identify attempted double spending. Spend proofs are verified off-chain, while the vault verifies proofs for deposits, closing balances, and escape withdrawals. In practice, spending is proven on user devices while the Ethereum contract remains the enforceable backstop for balances.

Users can withdraw through the Ethereum contract even if the zkAPI servers become unavailable, so access to deposited funds does not depend on the service staying online. The local client supports standard OpenAI and Ollama APIs, allowing existing applications to connect through a local endpoint.

Scope and Limitations

The Foundation said the same design could support blockchain RPC queries, image and video generation, VPN bandwidth, and machine-to-machine services. The live mainnet vault holds USDC credits, and a Sepolia deployment is available for testing, offering a testing path that does not involve mainnet credits.

The Foundation also identified the system's boundaries: zkAPI does not hide prompt contents, IP addresses, or request timing. Providers may still link sessions through reused personal details, writing styles, or conversation histories. Network anonymity and content privacy were flagged as separate limitations. Together, those boundaries mark where the system's privacy guarantees end, giving developers a concrete checklist for what zkAPI does and does not conceal.

Development Background

The implementation builds on an Ethereum Research design by Davide Crapis and Vitalik Buterin. Open Anonymity helped develop the client, server, and contracts. Between the Foundation's stated limitations and its suggested extensions beyond AI access, the project's own documentation outlines what to track as zkAPI moves past launch.

Source: CryptoBriefing