Mirai's AI Runs Faster on Your Phone

Mirai, a London-based startup, is focusing on enhancing artificial intelligence performance directly on consumer devices like phones and laptops. The company, founded by experienced developers Dima Shvets and Alexey Moiseenkov, secured $10 million in seed funding to advance its mission.

Their core technology involves an inference engine optimized for platforms like Apple Silicon, built using Rust. Mirai claims this engine can improve AI generation speed by up to 37% without any loss in output quality. The startup aims to simplify integration for developers, offering a streamlined experience.

Mirai's current efforts concentrate on text and voice modalities, with future plans to include vision support. They are also in discussions with chipmakers and frontier model providers to tune models for edge use and intend to expand their engine's availability to Android devices. This move positions Mirai to capture growing demand for efficient on-device AI solutions, addressing the high costs associated with cloud inference.

The company is also developing an orchestration layer to manage mixed-mode operations, sending complex requests to the cloud when necessary. This approach enables a seamless blend of on-device and cloud-based AI processing, catering to a wide range of applications from assistants to translation tools.