Mirai

Mirai is a technology company developing an on-device AI capability layer to make local inference more accessible for developers. Founded in 2024 by Dima Shvets and Alexey Moiseenkov, the San Francisco-based startup aims to address challenges in cloud-based AI, such as latency, privacy, and connectivity reliance. The founders are described as proven entrepreneurs who previously built and scaled consumer AI companies like Prisma and Reface. In February 2026, Mirai announced it had raised $10 million in a seed funding round led by Uncork Capital.

The company's core offering is a software development kit (SDK) and a proprietary inference engine optimized specifically for Apple Silicon. This technology enables developers to convert, optimize, distribute, and run AI models directly on Apple devices like Macs, iPhones, and iPads, with just a few lines of code. By focusing on the interaction between the AI model, runtime, and hardware, Mirai's platform is designed to deliver higher performance, including up to 37% faster generation and 59% faster prefill on certain models, compared to other runtimes. This on-device approach ensures data privacy by keeping information on the user's device, eliminates inference costs, and allows applications to function offline.

Mirai's business model targets developers and companies where low latency, cost-efficiency, and privacy are critical. The platform supports various AI architectures and offers tools for model integration, loading, execution, and optimization. It also allows for a hybrid model, where developers can route complex queries to the cloud while handling others on-device. The company's vision is to expand beyond Apple Silicon to support other platforms and workloads, including text, voice, and vision, with the goal of making on-device inference the default for AI applications.

Keywords: on-device AI, inference engine, AI SDK, Apple Silicon, developer tools, local inference, machine learning, LLM optimization, mobile AI, edge computing, AI runtime, model conversion, data privacy, low latency AI, Dima Shvets, Alexey Moiseenkov, Uncork Capital, AI hardware acceleration, private AI, offline AI