The Pulse
Ethereum Foundation Launches zkAPI for Private AI Payments
The Ethereum Foundation and Open Anonymity Project launched zkAPI on Ethereum mainnet on October 1, 2026. The system uses zero-knowledge proofs to separate API payments from user identity, but does not conceal prompts or IP addresses.
AI.info Team · Written with AI ·
The Ethereum Foundation and the Open Anonymity Project announced zkAPI on October 1, saying the system is live on Ethereum mainnet. It lets people fund metered API use without tying each payment to an identity-linked account. The privacy has limits: AI providers still see prompts, and network details such as a user’s IP address remain visible.
A zero-knowledge proof replaces the billing trail
Users deposit credits, including ETH or USDC, into an Ethereum vault contract. zkAPI represents the balance as a private note, then uses a zero-knowledge proof to show that the note can cover a proposed spend without identifying the deposit or the person who made it. A single proof can authorize one request or a session of requests.
The system uses commitments in a Merkle tree and publishes a one-way nullifier for each spend. The nullifier helps prevent double-spending: reusing a note produces a duplicate that exposes the attempted repeat, the Foundation says, without revealing other details about the payment. zkAPI uses Groth16 proofs on the BN254 curve and Poseidon hashing, with notes held in a 32-level Merkle tree.
The zkAPI server verifies spend proofs off-chain. Its Ethereum vault contract verifies proofs at deposit, close and escape, allowing users to withdraw on-chain without depending on the server, according to the announcement. The project is a working implementation of an earlier Ethereum Research design by Davide Crapis and Vitalik Buterin.
Short-lived keys keep billing apart from prompts
A local client requests authorization from the zkAPI server, which checks the proof and issues a short-lived API key capped at a dollar amount. The key stays in the user’s device memory, while prompts travel directly from that device to the AI provider in the runtime-key mode. When the key expires, a signed usage receipt records the actual metered cost, which the server deducts from the private balance rather than charging the full cap.
“The server that handles money never sees content, and the provider that sees content never learns the billing identity behind a key.”
Vittorio Rivabella, dAI Team
The Foundation also offers a simpler proxy mode, in which the zkAPI server relays requests to the provider and can see the traffic. The runtime-key mode is designed to keep the payment intermediary out of the request path. That distinction matters: the privacy claim depends on which mode is used.
OpenAI- and Ollama-compatible apps can use the client
The local client exposes standard OpenAI and Ollama APIs, so existing applications can connect by pointing to a service running on the user’s device. The same payment and authorization approach could also support blockchain queries, image and video generation, VPN bandwidth and machine-to-machine services, the Foundation says. Providers would accept proofs instead of API keys and settle signed usage receipts; their pricing, rate limits and infrastructure would stay as they are.
The Foundation says users can close their vault balance and withdraw on-chain even if zkAPI servers disappear. Its launch post also lists a mainnet vault holding USDC credits, a Sepolia test deployment and a browser chat client. The announcement does not provide adoption figures or name AI providers already accepting zkAPI payments.
Payment privacy does not hide a prompt or an IP address
zkAPI separates the billing link from API usage; it does not make the service itself anonymous. The provider still receives prompts and responses, and a stable IP address or request timing could help correlate sessions. Reused personal details, writing style, conversation history or project documents can also make separate sessions recognizable to an inference provider.
The Foundation suggests Tor with a fresh circuit for each session for users seeking stronger network privacy. But its own caveat is direct: zkAPI hides who paid, not what a user asked or the network metadata accompanying the request.