Mirai

Mirai is focused on on-device AI for Apple Silicon and builds the full stack from model to runtime and quantization. The company designs architectures and software to run locally, aiming to remove latency and keep data on-device. It targets real-time interactions and complex tasks that previously relied on cloud services, with a philosophy centered on sovereignty over the entire stack.

The project emphasizes running AI with hardware specifics in mind, creating a tightly integrated set of components that work together on-device rather than as a cloud proxy. It presents a vision where a single interface can handle multiple tasks without launching separate apps or services, prioritizing speed, privacy, and control over the software stack.

The core focus is on on-device execution for Apple Silicon, with a stack that includes a Rust inference engine, model optimization tools, and a local API/CLI. Mirai highlights a co-design approach where models, runtime, and quantization are developed to fit the device constraints, aiming for responsive and private AI interactions.

Mirai is an on-device AI lab, describing itself as building the models, inference runtime, and quantization stack from the device up. It stresses sovereignty, performance, and memory-conscious design aimed at enabling locally executing AI across Apple devices.

Use cases

• On-device AI for interactive and ambient experiences