
zkAPI lets users pay for metered APIs with ETH or USDC deposits on Ethereum and authorize spending through zero-knowledge proofs, separating payment from identity.
In brief
- zkAPI went live on Ethereum Mainnet on 1 October 2026, built by the Open Anonymity Project with the Ethereum Foundation.
- Users deposit Ether
ETH or USDC
USDC into a vault contract once, then prove spending with zero-knowledge proofs instead of an account. - The design hides the payment link, but providers still see prompt contents and network metadata such as IP addresses.
What happened
The Ethereum Foundation and the Open Anonymity Project have released zkAPI, a system for paying for metered APIs without revealing who is paying, and it runs on Ethereum Mainnet today.
A user deposits credits in Ether
ETH, USDC
USDC or other assets into a vault contract in a single transaction; the balance then exists as a private note that can be spent with a zero-knowledge proof rather than an identity, according to the foundation's announcement.
In the runtime-key mode, a client on the user's device sends the zkAPI server a proof of payment with no prompt attached.
The server checks it and issues a short-lived key capped in dollars, which the user's prompts travel with directly to the AI provider.
When the key expires, a signed usage receipt sets the amount deducted from the private balance.
The system uses Groth16 proofs on BN254, Poseidon hashing and a 32-level Merkle tree, with the vault contract verifying proofs at deposit, close and escape so that withdrawal does not depend on the server.
Why it matters
The foundation frames the problem as billing identity: an API key ties every request to an account and a payment method, letting a provider build a long-running profile of usage. zkAPI separates the payment layer from the content layer, so the server handling money does not see prompts and the provider running the model does not learn who is paying.
The same client and contracts are described as reusable for other metered services, including blockchain RPC queries, image generation and bandwidth.
The privacy is partial.
The announcement states that a provider still sees request contents and network metadata such as IP addresses, and may correlate sessions by timing; it points to VPNs or Tor for network anonymity and to confidential GPU computing for content privacy.
A simpler proxy mode is also offered, but the relay in that mode sees traffic.
What to watch
The announcement lists network anonymity and leakage through prompt contents as the system's two limitations, and says the gateway could potentially correlate request patterns from users sending requests under a stable connection.
For providers, adoption would mean accepting a proof instead of an API key and settling signed usage receipts instead of keeping accounts, with pricing and rate limits unchanged.