zama-ai/fhevm

FHEVM, a full-stack framework for integrating Fully Homomorphic Encryption (FHE) with blockchain applications

What it solves

FHEVM solves the problem of data privacy on public blockchains. While traditional smart contracts are transparent, FHEVM allows for the creation of confidential smart contracts where transaction data and state remain encrypted, yet can still be processed and updated on-chain without being revealed to the public or the network operators.

How it works

FHEVM leverages Fully Homomorphic Encryption (FHE), a cryptographic technique that allows computations to be performed on encrypted data. It integrates this capability into an EVM-compatible blockchain environment. To maintain efficiency, it uses symbolic execution on the host chain to represent FHE operations, while the actual heavy cryptographic computations are offloaded asynchronously to a specialized Rust-based coprocessor. Decryption is managed securely via a Key Management System (KMS) using multi-party computation (MPC).

Who it’s for

It is designed for blockchain developers who want to build decentralized applications (dApps) with programmable privacy—such as confidential voting, blind auctions, and private token transfers—without needing to be experts in cryptography.

Highlights

  • Solidity Integration: Write confidential contracts using standard Solidity and existing toolchains like Hardhat.
  • Quantum-Resistant: The underlying FHE crypto-scheme is designed to be secure against quantum computing attacks.
  • High Precision: Supports encrypted integers with up to 256 bits of precision.
  • Full Operator Support: Provides a complete range of mathematical and boolean operators (e.g., +, -, *, /, <, >, ==) for encrypted data.
  • End-to-End Encryption: Ensures that transaction data and state are encrypted from the moment they are submitted until they are processed.

Related

  • Project
    zama-ai/concrete-ml

    A privacy-preserving machine learning library that uses fully homomorphic encryption (FHE) to enable inference and training on encrypted data using scikit-learn-like APIs.

  • Project
    OpenMined/TenSEAL

    A library for performing homomorphic encryption operations on tensors, enabling secure computation on encrypted data without decryption.

  • Dispatch
    Towards Encrypted Large Language Models with FHE

    Hugging Face and Zama demonstrate how Fully Homomorphic Encryption (FHE) can enable Large Language Models to process encrypted user data while protecting the model owner's intellectual property.

  • Project
    zama-ai/tfhe-rs

    A pure Rust implementation of TFHE that enables boolean and integer arithmetics to be performed directly on encrypted data without decryption.

  • Dispatch
    Running Privacy-Preserving Inferences on Hugging Face Endpoints

    Hugging Face and Zama have enabled the deployment of Fully Homomorphic Encryption (FHE) models via Hugging Face Endpoints, allowing users to perform machine learning inferences on encrypted data without decrypting it.