Today’s AI discussion is dominated by building cloud capacity and massive data centers to power models. However, companies like Apple and Qualcomm are beginning to explore making on-device AI more practical. In this landscape, the 14-person technical team at London-based Mirai is focused on improving how AI models run directly on phones and laptops.
Mirai was founded last year by Dima Shvets and Alexey Moiseenkov, backed by a 10 million dollar seed round led by Uncork Capital. Both founders bring experience in building scalable consumer apps. Shvets co-founded the popular face-swapping app Reface and later served as a scout for a venture firm. Moiseenkov was the CEO and co-founder of the viral AI photo filter app Prisma.
As consumer app developers, both founders had been considering AI and machine learning on devices even before generative AI became mainstream. Shvets explained that when they met in London, they discussed technology and realized a key piece was missing amid the hype around cloud servers and artificial general intelligence. That missing piece was capable on-device AI for consumer hardware.
Shvets and Moiseenkov wanted to use AI to create a pipeline enabling complex tasks directly on phones, which inspired them to start Mirai. Conversations with other consumer app developers revealed a shared desire for better cost optimization and efficiency per token usage.
Currently, Mirai is developing a framework to help models perform better on devices. The company has built an inference engine for Apple Silicon that optimizes on-device processing speed. With its upcoming software development kit, developers will be able to integrate the runtime into their apps with just a few lines of code.
The founders envisioned providing developers with a simple integration experience, similar to adding a payment processor. A developer could go to their platform, integrate a key, and immediately start working on tasks like summarization or classification for their specific use case.
The startup built this engine using the Rust programming language, claiming it can increase a model’s generation speed by up to 37 percent. While tuning the model for a specific platform, the company states it does not alter the model’s weights, ensuring no loss in output quality.
Mirai’s current technology stack focuses on improving text and voice capabilities on devices, with plans to support vision in the future. The team has started collaborating with leading model providers to tune their models for edge device use and is in discussions with various chipmakers. Later, it plans to bring its engine to the Android ecosystem.
Additionally, Mirai aims to release on-device benchmarks so model makers can test performance directly on hardware. Shvets acknowledges that not all AI work can be done on-device. To enable a hybrid approach, the team is building an orchestration layer to seamlessly send requests that cannot be fulfilled on the device up to the cloud.
While the startup is not yet working directly with app developers, its engine could eventually power on-device assistants, transcribers, translators, and chat applications.
Andy McLoughlin, managing partner at Uncork Capital, noted his prior investment in an edge machine learning company a decade ago, which was ahead of its time and eventually sold to Spotify. He believes the situation is different today. Given the high cost of cloud-based AI processing, a shift is inevitable. He suggests that while investors currently fund companies spending heavily on cloud inference, this will not last as the underlying economics come into focus. He believes every model maker will want to run part of their workload at the edge, and Mirai is well-positioned to meet this demand.
Mirai’s seed round also included participation from numerous individuals, such as Dreamer CEO David Singleton, Y Combinator Partner Francois Chaubard, Snowflake co-founder Marcin Żukowski, ElevenLabs co-founder Mati Staniszewski, former Google and Coinbase executive Gokul Rajaram, investor Scooter Braun, Turing.com CTO Vijay Krishnan, Theory Forge Ventures’ Ben Parr and Matt Schlicht, and former Netflix technical leader Aditya Jami.