Rmlx is an R package that provides an interface to Apple's MLX (Machine Learning eXchange) library, designed for high-performance GPU computing specifically on Apple Silicon (M1, M2, M3, or later). It allows R users to leverage GPU acceleration for various machine learning and data processing tasks, enhancing computational efficiency. The package is built with a significant portion of the C++ API implemented, though it does not support all Python-only features, such as large neural network layers. To install Rmlx, users need macOS on Apple Silicon or a Linux setup with CUDA. The installation process can automatically locate a system-installed MLX library or download and build MLX version 0.29.4 from GitHub if not found. Users can customize their installation by specifying backend options, such as a CPU-only build or forcing CUDA on Linux. The package supports lazy evaluation, meaning operations are recorded but not executed until explicitly evaluated, which optimizes performance for large datasets. Rmlx features various capabilities for matrix operations, device management, and automatic differentiation. For instance, users can perform matrix arithmetic, apply reductions, and execute random sampling. The package also includes functions for sorting, selecting top-k values, and gradient computation, which are essential for tasks like training machine learning models. Memory management is efficient, as the M series chips allow seamless switching between CPU and GPU without incurring performance costs.