Laya Mlx

About Laya Mlx

Laya-MLX is a Python package that runs the open-weight Laya decision models on Apple Silicon through Apple's MLX framework, without PyTorch, Transformers or a cloud API. Laya answers constrained questions instead of generating text: pick one of several options, score against a rubric, or judge whether a statement is true. Each answer comes from one forward pass with calibrated probabilities, so there is no generated JSON to parse. You install it with pip (Apple Silicon, macOS 14 or later, Python 3.11 or later), load a pre-converted English or multilingual checkpoint from Hugging Face, and call predict with a schema of questions. Extras add a CLI, an HTTP server, an MCP server and integrations for LangChain, LlamaIndex, CrewAI and Agno. The README reports a median of about 13 ms for a short English decision on an M3 Max. Apache-2.0 licensed.

• Open Source