On October 1, the Ethereum Foundation and the Open Anonymity Project put zkAPI on Ethereum’s main network. It does exactly what it says it will do: you pay for AI, receive a token, and no one can trace the payment to the token.
GitHub hosts the working code, while the design roots go back to a February 2026 idea from Davide Crapis, who runs the foundation’s dAI team, and Vitalik Buterin. This version is experimental, carries no audited label, and its documentation remains open for all to see.
Here is how it actually works.
The Vault Contract
Credits enter the vault contract on Ethereum as a single ordinary transaction. After it settles, your balance becomes a private note — digital cash that only you can spend, with no record leading back to the original deposit.
The foundation laid out the issue plainly in its announcement, stating that “Prompts are personal. People ask AI models about their health, their finances, their doubts. Under the current model, using AI means handing a running transcript of your thinking to whoever holds the billing relationship,” which it argued was the crux of the matter.
“zkAPI separates payment from identity,” the foundation continued. “You deposit credits (ETH, USDC, etc.) into a vault contract on Ethereum, one ordinary transaction. From then on, your balance exists as a private note: digital cash that only you can spend and that nobody can trace back to the deposit.”
Zero-Knowledge Proofs
When you wish to ask an AI something, your device’s software generates a proof that shows you possess sufficient funds and have not yet spent them, all without exposing the specifics tied to those funds.
A temporary API key is returned after the server examines the proof. It functions as a password granting software access to an AI service. The key comes with a set dollar limit and exists solely within your device’s memory.
The request moves to the AI provider bearing only that key, carrying neither a name nor any payment information.
Once the cap expires, the provider hands over a signed record of your actual use and bills you for it, rather than charging the full limit. What you requested shows up on one side, and what you paid appears on another, with no way to connect the two records together.
The OpenRouter Connection
The documentation for the project describes temporary keys that operate with OpenRouter, which offers access to hundreds of AI models through a single connection. An app that follows the OpenAI format can hook up simply by pointing at a local address.
The protocol isn’t locked to one supplier. Instead, it backs several AI services via a common routing layer, with the foundation’s dAI team pushing to make Ethereum the top layer for settling and organizing those AIs.
The Proposal Behind It
A proposal posted by Davide Crapis and Vitalik Buterin on Ethereum Research in February gave rise to the idea. Crapis runs the foundation’s dAI team, whose goal is to make Ethereum the preferred settlement and coordination layer for AIs.
Before August 2025, Crapis had pitched Ethereum as plumbing for software agents. Then Crapis and Marco De Rossi proposed the ERC-8004 standard, a shared rulebook that lets autonomous AI agents find each other, verify identity and transact. Crapis predicted most Ethereum traffic would come from machines within three to five years.
On February 10, Buterin posted a thread laying out an Ethereum-AI roadmap, and it included zero-knowledge payments for private API use.
In April, he revealed a different approach: running an open-source model locally on his own Nvidia 5090 graphics card rather than entrusting his life to cloud services. When you host your own AI locally, your data stays put. zkAPI addresses the situation where your prompts still end up on someone else’s servers.
Other Uses
The post notes that the same plumbing could apply to several metered services, including blockchain data queries, image and video generation, VPNs, and payments between machines. For providers to integrate directly, they would need to accept a proof rather than an API key and settle signed usage receipts.
The repository calls the protocol experimental and makes no mention of a formal audit.
The Privacy Context
Privacy fears in AI aren’t theoretical. A federal court ruled in May 2025 that OpenAI must hold on to its output logs, including chats users had erased, in a copyright case filed by The New York Times and other publishers.
The ruling compelled OpenAI to retain logs beyond the point when users had removed their messages. The matter has not yet been settled.
OA Chat
The launch also brings OA Chat, which is a private chatbot available through a browser with no need for installation. Its code sits open on GitHub.
This project relies on the Open Anonymity Project framework. The foundation’s statement sets out the larger issue: “Prompts are personal. People ask AI models about their health, their finances, their doubts. Under the current model, using AI means handing a running transcript of your thinking to whoever holds the billing relationship.”.
What We Know So Far
- zkAPI launched on Ethereum mainnet on October 1
- The design follows a February 2026 Ethereum Research proposal by Davide Crapis and Vitalik Buterin
- The project’s repository labels the protocol experimental, and its documentation points to OpenRouter as the AI provider behind the keys
- The protocol separates payment from identity using zero-knowledge proofs
- The system is built with the Open Anonymity Project
- OA Chat is a browser-based private chatbot that needs no installation
- The code is open on GitHub
Our View
This is a clever piece of engineering that addresses a real problem: paying for AI without exposing your identity. The design has already been published and reviewed.
The experimental label holds true. The protocol has not been audited, and it bears no formal certification. Yet the design is public, the code sits on GitHub, and the problem it tackles is genuine.
Adoption remains the larger issue. Providers must accept a proof rather than an API key and settle signed usage receipts, which alters how existing AI services handle billing. That is a significant undertaking, and there is no certainty that providers will want to take it on.
The foundation has shown it can advance standards before. Crapis’s ERC-8004 proposal shows the team is willing to build infrastructure for machine-to-machine coordination, and Buterin’s plan shows he views private payments as central to the Ethereum-AI mix.
Currently, zkAPI exists as a functioning prototype running on mainnet. It remains experimental and has not been labeled as audited, though its documentation is openly available. Should it prove durable in actual use, it could serve as the foundational layer for a generation of privacy-preserving AI interactions.
The foundation begins its statement with a straightforward pledge: “You deposit credits (ETH, USDC, etc.) into a vault contract on Ethereum, one ordinary transaction. From then on, your balance exists as a private note: digital cash that only you can spend and that nobody can trace back to the deposit.”. It is a strong assertion, and the source code is open for independent verification by anyone who wishes to examine it.
It is early days, but the pieces are on the board.
Source material: “Ethereum Now Lets You Pay for AI Without Revealing Who You Are,” Decrypt.
Get the Notebook.
The day's best stories and every fresh verdict, in plain English, in your inbox by seven. One email a day, no more.