The Open Anonymity Project, in collaboration with the Ethereum Foundation, has officially launched zkAPI on the Ethereum Mainnet, introducing a novel cryptographic architecture designed to decouple user identity from metered API consumption. The system allows users to deposit funds into an Ethereum vault and subsequently authorize bounded usage through zero-knowledge proofs, effectively eliminating the persistent billing trails that currently accompany artificial intelligence prompts, blockchain queries, and other per-use digital services.
The Problem: The Persistent Digital Footprint of Modern API Calls
In the contemporary digital ecosystem, nearly every interaction with artificial intelligence models and automated web services carries a rigid chain of identification. Under standard operational models, a user’s API key links directly to an account profile, which in turn connects to a traditional payment method, such as a credit card or institutional billing contract. Consequently, every individual prompt, query, or file upload submitted to a service provider joins a permanent record attached to both financial and personal identifiers. This mechanism enables providers to aggregate years of usage history into a single, comprehensive user profile.
The implications of this structural design are particularly acute when interacting with generative artificial intelligence. Prompts submitted by individuals frequently encompass deeply personal queries regarding sensitive topics, including physical and mental health concerns, financial planning, legal uncertainties, and private creative work. Under the prevailing paradigm, utilizing modern AI tools requires individuals to hand over a continuous, unredacted transcript of their thoughts and inquiries to whichever entity controls the billing relationship.
Historically, viable alternatives to this surveillance-heavy model have been remarkably scarce. Users could theoretically settle payments onchain for individual requests, but public blockchain transactions are notoriously slow, economically expensive, and completely traceable by any observer. Alternatively, individuals could place their trust in a centralized middleman to safeguard their data privacy, an approach that offers no cryptographic guarantees against data harvesting, subpoenas, or breaches.
The Idea: Separating Payment from Identity Through Cryptography
To resolve this systemic privacy vulnerability, zkAPI introduces a framework that cleanly separates financial settlement from user identity. The underlying concept, originally detailed in a research proposal published on Ethereum Research by Davide Crapis and Vitalik Buterin, allows individuals to deposit assets such as Ethereum or USDC into a smart contract vault on the Ethereum network through a single, ordinary transaction. Once deposited, the user’s remaining balance transitions into a private note—essentially a form of digital cash that can be spent exclusively by its rightful owner without any traceable link back to the original deposit transaction.
To authorize ongoing spending without revealing sensitive information, specialized software running locally on the user’s machine generates a compact zero-knowledge proof. This cryptographic proof mathematically demonstrates that a valid, funded note covers the requested expenditure and that the specific funds have not been previously expended. A single proof can be structured to authorize either an isolated request or an entire multi-turn session.
Upon receiving the request, the target server can independently verify the validity of the mathematical statement without learning which specific note, which deposit transaction, or which human being initiated the interaction. At the payment layer, incoming requests remain completely disconnected from the user’s real-world identity and are rendered entirely unlinkable from one another.
This privacy architecture relies on two foundational cryptographic mechanisms. First, user deposits exist as cryptographic commitments stored within a Merkle tree, enabling a zero-knowledge proof to confirm that a given note belongs to the set of valid deposits without pointing to any specific entry. Second, every authorized expenditure publishes a unique cryptographic nullifier—a one-way serial number derived directly from the secret parameters of the note. A user who remains within their funded balance operates entirely anonymously. However, any attempt to double-spend the same balance generates a duplicate nullifier, exposing the fraudulent attempt to the network while revealing no other information about the user.
How It Works in Practice
The operational mechanics of zkAPI divide the interaction into two parallel pathways: a conversation path handled by the AI inference provider and a payment path managed by the zkAPI infrastructure. These two distinct paths share only two temporary objects: a short-lived runtime key and a cryptographic usage receipt.
When a user initiates a session, the system ensures that the server processing the financial transactions never inspects the actual content of the conversation. Simultaneously, the AI provider that processes the prompts and generates responses learns the content of the requests but never discovers the billing identity or the funding source associated with the temporary key.
The spending cap functions as a secure reservation system. As the session progresses, the usage limits are enforced locally and verified cryptographically. When the temporary key expires, the provider records the total resource consumption in a signed receipt, and the zkAPI server deducts the corresponding amount from the user’s private note. Neither party possesses the ability to retroactively alter the billing ledger, and a single authorization covers an entire session of model queries rather than requiring a separate proof for every individual API call.

In addition to its fully decentralized runtime-key mode, zkAPI offers a streamlined proxy mode wherein the zkAPI server directly relays requests to the service provider. While this alternative mode simplifies operational deployment, the relay infrastructure retains visibility over the transit traffic. The runtime-key mode remains the preferred architecture for scenarios requiring strict assurances that no payment intermediary can observe the connection between the user and the service.
From a technical standpoint, zkAPI validates spending proofs using the Groth16 zero-knowledge proving system operating over the BN254 elliptic curve. It computes commitments and nullifiers using the Poseidon hash function, and manages private notes within a Merkle tree structured thirty-two levels deep. The receiving server verifies spend proofs off-chain for efficiency, while the underlying Ethereum vault contract independently verifies the identical class of proofs during deposit, closure, and emergency exit procedures. This ensures that users retain full sovereignty over their funds, meaning their ability to withdraw unspent balances never relies on the ongoing honesty or availability of the server operator.
Use Cases Beyond Artificial Intelligence
While AI inference represents the most immediate application for zkAPI due to the sensitive nature of conversational prompts and the invasive nature of traditional billing trails, the underlying client software and smart contract architecture can secure virtually any service that charges on a metered basis.
For decentralized applications and blockchain infrastructure, zkAPI can facilitate blockchain Remote Procedure Call (RPC) queries, allowing developers to query network nodes without tying their funding source or wallet address to their specific application queries. Similarly, platforms offering intensive image and video generation jobs can process creative tasks while obscuring the financial link between the paying entity and the generated media.
Network services such as virtual private networks and high-bandwidth data relays can also leverage the technology to meter time and data consumption without requiring users to register traditional accounts or provide payment cards. Furthermore, autonomous machine-to-machine agents can utilize the system to execute micro-transactions and pay for computational tasks dynamically without maintaining persistent accounts or exposing their operators to tracking.
Despite these advanced privacy guarantees, developers emphasize that zkAPI is specifically designed to obscure financial linkages rather than network metadata or content contents. A service provider still observes the raw text of a request and network-level metadata, such as the user’s Internet Protocol address, and may theoretically attempt to correlate distinct sessions based on timing patterns. Consequently, achieving comprehensive operational security requires combining zkAPI with independent network-layer anonymity tools, such as routing traffic through the Tor network or utilizing emerging confidential GPU computing environments to secure model processing.
For service providers and API operators, adopting zkAPI requires minimal friction. Integration involves accepting a cryptographic proof in lieu of a traditional API key and settling cryptographically signed usage receipts instead of maintaining traditional customer accounts. Core business parameters, including pricing models, rate limits, and underlying cloud infrastructure, remain entirely unaffected by the integration.
Limitations and Considerations
While zkAPI establishes a robust cryptographic barrier between payments and identities, the system operates within certain inherent limitations related to network metadata and content-based correlation.
On the network layer, gateways can potentially analyze request patterns and correlate traffic if a user consistently routes queries through a static residential or commercial Internet Protocol address. For individuals requiring a higher degree of operational security, maintainers recommend routing requests through the Tor browser network and establishing a fresh circuit for each discrete session to prevent traffic analysis.
On the content layer, individual sessions can potentially be re-linked by inference providers if a user inadvertently introduces distinctive personal details, unique writing styles, reused conversation histories, or proprietary project documents across multiple sessions. Shared prompt contents can function as behavioral fingerprints for any entity capable of reading the incoming prompts. This dynamic reflects an inherent privacy-utility tradeoff within language models, as standalone queries offer heightened privacy but reduced contextual utility due to the absence of past conversational history. Potential mitigations discussed by the Open Anonymity research collective include leveraging local client-side models or trusted execution environments to synthesize contextually relevant requests on top of shared memory rather than manually drafting prompts.
zkAPI is currently live on the Ethereum Mainnet, representing the practical realization of the design framework originally outlined on Ethereum Research by Davide Crapis and Vitalik Buterin. The open-source client software, server infrastructure, and smart contracts are fully operational, enabling developers to build local gateways and integrate privacy-preserving settlement into any OpenAI-compatible application using the official documentation provided by the Open Anonymity Project.

Comments (0)