Overview

  • The Ethereum Foundation, in collaboration with the Open Anonymity Project, unveiled zkAPI on the Ethereum mainnet on October 1, allowing users to prepay using USDC and acquire temporary API keys through zero-knowledge proofs.
  • This initiative is based on a proposal from February 2026 by Davide Crapis, who heads the foundation's dAI team, along with Vitalik Buterin.
  • The project is tagged as experimental, and its documentation indicates that OpenRouter is the AI service provider associated with the keys.

The Ethereum Foundation has officially launched zkAPI on Ethereum's main network, a system designed to enable users to pay for AI services without associating their identity with their inquiries, as detailed in a recent announcement.

According to the Ethereum Foundation, “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.”

Myriad: Predict Ethereum's next move.

This system was developed alongside the Open Anonymity Project, and it also introduces OA Chat, a user-friendly private chatbot that requires no installation; the code is available on GitHub.

The issue of privacy in the AI sector is quite pressing. In May 2025, a federal court mandated OpenAI to retain its output logs, including chats that users had deleted, as part of a copyright lawsuit involving The New York Times and other publishers.

With this new system, each time a user wants to query an AI, their device generates a zero-knowledge proof—a mathematical receipt that verifies a fact without revealing the underlying details. This proof confirms that the user has sufficient funds and has not already spent them.

The server then verifies the proof and provides a temporary API key, essentially a password that allows software to communicate with an AI service. This key has a spending limit and is stored only in the device's memory.

Requests sent to the AI provider include just this key, with no personal information or payment details attached.

Once the key expires, the provider generates a signed usage receipt, charging the user only for the actual service used, not the capped amount. The AI provider sees the query, the payment layer sees the expenditure, but neither can link the two.

The project's documentation describes these short-lived keys as compatible with OpenRouter, a service that provides access to numerous AI models through a single connection. Any application that utilizes the OpenAI format can easily integrate by pointing to a local address.

This concept was derived from a proposal made by Davide Crapis and Vitalik Buterin in February. Crapis heads the foundation's dAI team, which aims to establish Ethereum as the go-to settlement and coordination layer for AI technologies.

Crapis had previously suggested that Ethereum could serve as the infrastructure for software agents. In August 2025, he and Marco De Rossi introduced the ERC-8004 standard, which creates a shared framework allowing autonomous AI agents to locate one another, verify identities, and conduct transactions. He anticipates that within three to five years, most traffic on Ethereum will come from machines.

Buterin echoed similar sentiments. On February 10, he outlined an Ethereum-AI roadmap that included zero-knowledge payments for private API interactions.

In April, he presented an alternative approach: running an open-source model locally on his Nvidia 5090 graphics card, rather than sharing personal data with cloud services. Local models keep data on the user's device, while zkAPI addresses scenarios where queries are sent to external servers.

This infrastructure could potentially support various metered services beyond AI, such as blockchain data queries, image and video generation, VPNs, and inter-machine payments. Providers would need to accept a proof rather than an API key and settle signed usage receipts for direct integration.

The protocol is currently labeled as experimental, and a formal audit has not been conducted.