ML-KEM Python Implementation

A Python implementation of the ML-KEM (Kyber-based Module-Lattice Key Encapsulation Mechanism) algorithm as standardized in NIST FIPS 203.

Features

• Complete ML-KEM Implementation: All three variants (512, 768, 1024) following official NIST FIPS 203 specification
• Secure Messaging CLI: Real-time encrypted chat using ML-KEM for key exchange and AES-GCM for message encryption
• Performance Tools: Comprehensive benchmarking and correctness/CCA-resistance verification
• Modular Design: Clean separation of KEM operations, PKE primitives, and utility functions

Quick Start

git clone https://github.com/Preterno/ml-kem.git
cd ml-kem
pip install -r requirements.txt
python test.py  # Verify installation

Installation

Requirements

• Python 3.8 or higher
• Dependencies listed in requirements.txt

Setup

git clone https://github.com/Preterno/ml-kem.git
cd ml-kem
pip install -r requirements.txt

Usage

Basic Example

from pke.params import ML_KEM_512, ML_KEM_768, ML_KEM_1024
from kem.keygen import ml_kem_keygen
from kem.encapsulate import ml_kem_encaps
from kem.decapsulate import ml_kem_decaps

# Choose ML-KEM variant
params = ML_KEM_768

# Generate key pair
ek, dk = ml_kem_keygen(params)  # ek = encapsulation key, dk = decapsulation key

# Encapsulate (sender side)
K, ct = ml_kem_encaps(ek, params)  # K = shared secret, ct = ciphertext

# Decapsulate (receiver side)
K_prime = ml_kem_decaps(dk, ct, params)

assert K == K_prime  # Shared secrets should match

Run Tests

To verify correctness and CCA security:

python test.py

Run Benchmarks

To measure performance of ML-KEM operations:

python benchmark_mlkem.py

Secure Messaging CLI

Launch the encrypted chat application:

1. Start the server:

python chat/server.py

2. Connect with client:

python chat/client.py

The client and server perform ML-KEM-768 key exchange and use AES-GCM for encrypted messaging.

Project Structure

.
├── kem/                    # ML-KEM logic (keygen, encaps, decaps)
├── pke/                    # Kyber PKE primitives
├── utils/                  # Support utilities (hashing, polynomials, etc.)
├── chat/                   # CLI chat app using ML-KEM + AES
├── benchmark_mlkem.py
├── test.py
├── requirements.txt

Performance Benchmarks

Execution Time Comparison

Variant	KeyGen	Encaps	Decaps	Full Cycle	Public Key Size
ML-KEM-512	13.13 ms	15.20 ms	40.00 ms	53.04 ms	800 bytes
ML-KEM-768	19.57 ms	22.85 ms	59.28 ms	78.63 ms	1,184 bytes
ML-KEM-1024	28.00 ms	32.68 ms	84.03 ms	115.77 ms	1,568 bytes

Throughput Comparison

Variant	KeyGen	Encaps	Decaps	Full Cycle
ML-KEM-512	76.2 ops/sec	65.8 ops/sec	25.0 ops/sec	18.9 ops/sec
ML-KEM-768	51.1 ops/sec	43.8 ops/sec	16.9 ops/sec	12.7 ops/sec
ML-KEM-1024	35.7 ops/sec	30.6 ops/sec	11.9 ops/sec	8.6 ops/sec

Note: Benchmarks may vary depending on hardware and system configuration. Run python benchmark_mlkem.py for your specific environment.

Disclaimer

This project is for educational and research use only. It has not undergone security audits and is not intended for production or security-critical deployments.