Lux FHE

Fully homomorphic encryption library for computing on encrypted data -- original Go implementation built on lattice-based cryptography.

go get github.com/luxfi/fhe

What This Is

A from-scratch TFHE (Threshold Fully Homomorphic Encryption) implementation in Go. Not a wrapper around tfhe-rs, not a binding to Zama, not a port of OpenFHE. The core is built directly on luxfi/lattice (RLWE, RGSW, blind rotation) with NTT SIMD acceleration written in Go.

This library encrypts individual bits, bootstraps after every gate, and composes gates into arbitrary boolean circuits including full integer arithmetic. It supports encrypted integer types from 4-bit to 256-bit (including euint160 for Ethereum addresses and euint256 for uint256).

Architecture

fhe.go              Parameters, key generation, bootstrap key construction
                    Standard parameter sets: PN10QP27, PN11QP54, PN9QP28_STD128, PN9QP27_STD128Q

encryptor.go        Bit and public-key encryption (LWE samples, Q/8 encoding)
decryptor.go        Bit decryption (threshold extraction from LWE)

evaluator.go        Boolean gate evaluation on encrypted bits
                    Gates: AND, OR, XOR, NAND, NOR, XNOR, NOT
                    Multi-input: MAJORITY (2-of-3), CMPCOMBINE (comparison cascade)
                    Each gate = blind rotation + sample extraction + key switching

integers.go         Radix ciphertext (encrypted integers via bit decomposition)
integer_ops.go      Add, Sub, Mul, Neg, comparison (Lt, Gt, Eq, Lte, Gte, Min, Max)
bitwise_integers.go AND, OR, XOR, NOT, shifts on encrypted integers
shortint.go         Short integer operations (4-bit, 8-bit)
lazy_carry.go       Lazy carry propagation for chained arithmetic

ntt_simd.go         SIMD-optimized NTT engine
                    Barrett reduction, Montgomery multiplication
                    Precomputed twiddle factors in bit-reversed order
                    Parallel butterfly operations

serialization.go    Ciphertext serialization/deserialization
security.go         Security level estimation
profile.go          Performance profiling hooks
random.go           Cryptographic randomness
threshold_rng.go    Threshold random number generation on encrypted data

How It Works

1. Encryption: A plaintext bit is encoded as an LWE sample with noise: (a, b = <a,s> + m*Q/8 + e). The Q/8 encoding places TRUE at +Q/8 and FALSE at -Q/8.

2. Gate evaluation: Two encrypted bits are added (producing a noisy result in [-Q/4, Q/4]), then a programmable blind rotation maps the result through a test polynomial that encodes the gate's truth table.

3. Bootstrapping: Every gate evaluation includes a full bootstrap (blind rotation + sample extraction + key switching), refreshing noise so gates can be composed indefinitely.

4. Integer arithmetic: Integers are radix-decomposed into encrypted bits. Addition uses a ripple-carry adder with lazy carry propagation. Multiplication uses shift-and-add. Comparison uses a cascaded bit-by-bit approach with the CMPCOMBINE gate.

Parameter Sets

Parameter Set	N	Q	Security	Use Case
PN10QP27	1024	~134M	128-bit classical	General purpose
PN11QP54	2048	~2^54	128-bit classical	Higher precision
PN9QP28_STD128	512/1024	~2^28	128-bit classical	OpenFHE-comparable
PN9QP27_STD128Q	512/1024	~2^27	128-bit post-quantum	PQ security margin

Encrypted Integer Types

Type	Bits	Use
FheBool	1	Encrypted boolean
FheUint4	4	Short integer, nibble operations
FheUint8	8	Byte-level encrypted data
FheUint16	16	Short integers
FheUint32	32	Standard integers
FheUint64	64	Balances, timestamps
FheUint128	128	Large integers
FheUint160	160	Ethereum addresses
FheUint256	256	EVM uint256

FFI Bindings

C Shared Library

CGO_ENABLED=1 go build -buildmode=c-shared -o dist/libluxfhe.so ./bindings/cabi/

Produces libluxfhe.{so,dylib,dll} + libluxfhe.h. All Go objects are managed behind opaque uint64 handles. Callable from Python (ctypes/cffi), TypeScript (N-API), or any C-compatible FFI.

Rust

cd bindings/rust
cargo build --release

Rust bindings link against the C shared library via build.rs.

Usage

import "github.com/luxfi/fhe"

// Setup
params, _ := fhe.NewParametersFromLiteral(fhe.PN10QP27)
kg := fhe.NewKeyGenerator(params)
sk, pk := kg.GenKeyPair()
bsk := kg.GenBootstrapKey(sk)

// Encrypt bits
enc := fhe.NewEncryptor(params, sk)
ct1, _ := enc.EncryptBit(true)
ct2, _ := enc.EncryptBit(false)

// Evaluate gates (no secret key needed)
eval := fhe.NewEvaluator(params, bsk)
ctAnd, _ := eval.AND(ct1, ct2)   // encrypted false
ctOr, _ := eval.OR(ct1, ct2)     // encrypted true
ctXor, _ := eval.XOR(ct1, ct2)   // encrypted true

// Decrypt
dec := fhe.NewDecryptor(params, sk)
result := dec.DecryptBit(ctAnd)   // false

Integer Arithmetic

intParams, _ := fhe.NewIntegerParams(params, 2) // 2-bit message space per block
intEnc := fhe.NewIntegerEncryptor(intParams, sk)

a, _ := intEnc.EncryptUint64(42, fhe.FheUint64)
b, _ := intEnc.EncryptUint64(10, fhe.FheUint64)

intEval := fhe.NewIntegerEvaluator(intParams, bsk)
sum, _ := intEval.Add(a, b)    // encrypted 52
lt, _ := intEval.Lt(a, b)      // encrypted false (42 < 10)

intDec := fhe.NewIntegerDecryptor(intParams, sk)
val := intDec.Decrypt64(sum)   // 52

Precompiled Binaries

Application-specific encrypted computation demos:

Binary	Description
auction	Sealed-bid auction on encrypted bids
compliance	Encrypted compliance threshold checking
darkpool	Dark pool order matching on encrypted orders
marketmaker	Market making with encrypted spreads
mediaseal	Media provenance with encrypted metadata
vote / voting	Encrypted ballot tallying
seal	Generic encrypt/decrypt CLI
proof	Zero-knowledge proof of encrypted computation
stats	Statistical operations on encrypted data

Testing

go test ./...

Papers

• Lux TFHE -- TFHE construction and parameter analysis
• Lux FHE Smart Contracts -- encrypted EVM execution
• Lux FHE Benchmarks -- gate timing and comparison with OpenFHE/tfhe-rs
• Lux FHE API -- precompile interface specification
• Lux FHE MPC Hybrid -- threshold decryption via MPC
• Lux NTT Transform -- SIMD NTT optimization

Dependencies

• luxfi/lattice -- RLWE, RGSW, ring operations, blind rotation (the actual lattice math)

No dependency on tfhe-rs, Zama, OpenFHE, SEAL, HElib, or any other FHE library.

License

Lux Ecosystem License v1.2. See LICENSE.