The Laya-MLX engine has been released for local model execution on Apple Silicon

Developers introduced Laya-MLX — a specialized engine for local inference of typed decision-making models, such as Laya. The tool is optimized to work on devices with Apple Silicon architecture and does not require the presence of PyTorch or connection to cloud APIs. The engine's architecture provides high data processing speed when performing calculations directly on Apple chips. According to technical specifications, the latency when processing English-language queries is 13.4 ms. For multilingual input data, the performance indicator reaches 7.4 ms. The project is distributed with open source code, which allows it to be integrated into local applications without the need to use third-party cloud infrastructures. Using Apple's native MLX framework allows for efficient distribution of the load between CPU and GPU cores, minimizing RAM consumption. The release is targeted at developers creating automated decision-making systems where minimal response latency is critical. Refusal to use heavy dependencies makes the engine suitable for embedding in lightweight desktop applications that work in real-time on macOS.