The Ethereum Foundation and the Open Anonymity Project launched zkAPI on Ethereum mainnet on Oct. 1, 2026, according to an Ethereum Foundation blog post. The system enables users to pay for metered APIs without disclosing who made the payment, with AI inference as its first target because prompts can be sensitive and conventional billing records connect requests to identities. Users deposit ETH or USDC credits into an Ethereum vault contract in one transaction. The balance then becomes a private note that only its holder can spend. Local software creates a zero-knowledge proof (proof of validity without revealing underlying data) showing that a funded note covers a charge and has not previously been spent. Each payment publishes a nullifier, a one-way serial number that exposes double-spending attempts without identifying the note behind the payment. zkAPI includes a local client compatible with standard OpenAI and Ollama APIs, allowing existing applications and chat clients to connect through localhost. A server verifies the payment proof, issues a short-lived API key held only in device memory, and deducts the private balance when the key expires. The design separates information between the payment server, which can confirm a valid payment but not the payer, and the AI provider, which can see prompts but not the payer's identity. The client can also support RPC queries, image generation, VPNs and machine-to-machine services. The launch implements the usage-credit system described in an Ethereum Research design by Davide Crapis and Vitalik Buterin. It uses Groth16 proofs on the BN254 curve, Poseidon hashing for commitments and nullifiers, and a 32-level Merkle tree to store notes. The active Ethereum mainnet vault holds USDC, while withdrawals remain available on-chain if the zkAPI server disappears. The project does not provide network anonymity, so the foundation recommends Tor for users seeking greater privacy, and shared prompt content can still create identifying fingerprints. Development remains active, with confidential GPU computing for content privacy still emerging.