The current AI landscape is dominated by discussions about expanding cloud infrastructure and constructing massive data centers to support large models. However, companies such as Apple and Qualcomm are beginning to explore ways to make on-device artificial intelligence more practical and powerful. In this context, the 14-person technical team at London-based Mirai is dedicated to enhancing how AI models operate directly on smartphones and laptops.

Mirai was established last year by founders Dima Shvets and Alexey Moiseenkov, supported by a $10 million seed investment led by Uncork Capital. Both founders bring significant experience in creating scalable consumer applications. Shvets previously co-founded the face-swapping app Reface, which received backing from a16z, and later served as a scout for that venture firm. Moiseenkov was the CEO and co-founder of Prisma, the viral AI photo filter app from the previous decade.

As developers focused on consumer technology, both founders had been considering the potential of on-device AI and machine learning well before the generative AI boom gained momentum. "When we met together in London, we started to chat about technology, and we realized that within the hype of gen AI and more AI adoption, everybody speaks about cloud, about servers, about AGI coming," Shvets explained. Their shared vision was to use AI to build a pipeline capable of handling complex tasks directly on mobile devices, which inspired them to launch Mirai. Conversations with other consumer app developers revealed a widespread desire for improved cost optimization and better margins per token usage.

Today, Mirai is creating a framework designed to improve model performance on devices. The company has developed an inference engine optimized for Apple Silicon to boost on-device processing throughput. According to the company, its upcoming software development kit will allow developers to integrate this runtime into their applications with just a few lines of code.

"One of the visions why we started the company was that we wanted to give developers, like this Stripe-like, eight lines of code [integration] experience…you basically go to our platform, integrate the key, and start working with summarization, classification, or whatever your use case is," Shvets stated. This engine was built using the Rust programming language, which the company claims can increase a model's generation speed by up to 37%. Importantly, Mirai emphasizes that when tuning a model for a specific platform, it does not alter the model's weights, ensuring no degradation in output quality.

Currently, Mirai's technology stack is focused on improving text and voice capabilities on devices, with plans to add vision support in the future. The team has begun collaborating with leading model providers to optimize their models for edge computing and is in discussions with various chip manufacturers. Future plans include expanding the engine to support Android devices. Additionally, Mirai intends to release standardized on-device benchmarks, allowing model creators to test and compare performance directly on hardware.

Shvets acknowledges that not all AI tasks can be processed on-device. To address this, the team is building an orchestration layer that can seamlessly route requests too complex for local processing up to the cloud, enabling a hybrid operational mode. Although the startup is not yet working directly with application developers, its technology has the potential to power on-device assistants, transcription services, translation tools, and chat applications.

Andy McLoughlin, Managing Partner at Uncork Capital, noted that he previously invested in an edge machine learning company about a decade ago, which was ultimately acquired by Spotify. He believes the current environment is fundamentally different. "Given the cost of cloud inference, something has to change… For now, VCs are happy to continue funding the rocketship companies, spending inordinate sums on cloud inference. But that won’t last—at some point, people will focus on the underlying economics of these businesses and realize that something has to change," he said. "It feels like every model maker will want to run part of their inference workloads at the edge, and Mirai feels very well positioned to capture this demand."

Mirai's seed funding round also included participation from several notable individual investors. These include Dreamer CEO David Singleton, Y Combinator Partner Francois Chaubard, Snowflake co-founder Marcin Żukowski, ElevenLabs co-founder Mati Staniszewski, former Google AdSense product manager and Coinbase board member Gokul Rajaram, Groq investor Scooter Braun, Turing.com CTO Vijay Krishnan, Theory Forge Ventures partners Ben Parr and Matt Schlicht, and former Netflix technical leader Aditya Jami.