Mirai is a London-based startup developing a high-performance inference engine, written in Rust, that allows large language and voice models to run faster on Apple Silicon chips. The company aims to offer developers a Stripe-like experience: an upcoming SDK that can be integrated with only a few lines of code to enable tasks such as summarization, classification, transcription and chat directly on phones and laptops. Internal tests show the engine can boost generation speeds by up to 37% without altering model weights, preserving output quality. Although the current focus is on text and voice, the roadmap includes support for vision models and an Android port. Mirai is also creating an orchestration layer that offloads requests the device cannot handle to the cloud, as well as publishing on-device benchmarks for model makers. Founded by Reface co-founder Dima Shvets and Prisma co-founder Alexey Moiseenkov, the company has assembled a 14-person technical team. It has raised a $10 million seed round to fund product development and partnerships with frontier model providers and chipmakers.

Total raised

$10M

Funding rounds

1

Latest round

Seed

Latest activity

Feb 2026