EZKL is an open-source library and command-line tool for zero-knowledge machine learning (ZKML). It converts deep learning models or other computational graphs into zero-knowledge proof (ZK-SNARK) circuits, enabling cryptographic verification of model inference results. It leverages the Halo2 proof system under the hood, and generated proofs can be efficiently verified on EVM, browsers, or devices, suitable for privacy-preserving inference and model integrity attestation.