Reface and Prisma Founders Launch Mirai for On-Device AI
▼ Summary
– Mirai is a London-based startup developing a framework to optimize AI model performance on consumer devices like phones and laptops, contrasting with the industry’s focus on cloud-based AI.
– The company was founded by Dima Shvets and Alexey Moiseenkov, who have backgrounds in scalable consumer apps like Reface and Prisma, and is backed by a $10 million seed round led by Uncork Capital.
– Mirai has built an inference engine in Rust for Apple Silicon that increases generation speed by up to 37% without altering model weights, and plans an easy-to-integrate SDK for developers.
– The startup’s technology currently supports text and voice AI tasks with future plans for vision, and it is building an orchestration layer to manage workloads between devices and the cloud.
– Investors believe the high cost of cloud inference will drive demand for on-device AI solutions, positioning Mirai to capture this market shift as model makers seek edge computing options.
While much of the current AI discussion centers on massive cloud infrastructure, a new startup is focusing its efforts on bringing powerful intelligence directly to your pocket. Mirai, founded by the creators of the popular apps Reface and Prisma, is developing technology to make AI models run faster and more efficiently on personal devices like smartphones and laptops. This shift toward on-device AI promises to reduce costs, improve privacy, and unlock new capabilities for consumer applications without constant reliance on distant data centers.
The company was launched last year by Dima Shvets and Alexey Moiseenkov, who pooled their experience from building viral consumer apps. Their collaboration began with a shared realization that the industry’s intense focus on cloud-based artificial general intelligence overlooked a critical opportunity. They identified a growing need among developers for tools that could handle complex AI tasks locally on a phone, optimizing both performance and operational costs. This insight led directly to the creation of Mirai, which recently secured $10 million in seed funding led by Uncork Capital.
Mirai’s core product is a framework designed to enhance how AI models operate on consumer hardware. The team has constructed an inference engine specifically optimized for Apple Silicon, claiming it can boost a model’s generation speed by as much as 37 percent. A key feature is that the technology tunes models for specific platforms without altering their core weights, ensuring the final output maintains its original quality. The company plans to release a software development kit that allows other developers to integrate this runtime into their own apps with just a few lines of code.
“We envisioned giving developers a Stripe-like integration experience,” Shvets explained. “You come to our platform, integrate a key, and immediately start working on tasks like summarization or classification for your specific use case.”
Currently, the startup’s stack supports text and voice-based AI functions, with plans to add computer vision capabilities in the future. The team is actively collaborating with leading model providers to adapt their offerings for edge computing and is engaging in discussions with various chipmakers. While the initial focus is on Apple’s ecosystem, there are concrete plans to expand the engine to the Android platform. Mirai also intends to publish a set of on-device performance benchmarks, giving model creators standardized tools to test and improve their edge offerings.
The founders acknowledge that not every AI task can be processed locally. To address this, they are building an orchestration layer that can intelligently route requests. If a task is too complex for the device, the system will seamlessly offload it to the cloud, enabling a flexible, hybrid mode of operation. This technology could soon power a new generation of on-device assistants, transcription services, translation tools, and chat applications.
Investor Andy McLoughlin of Uncork Capital sees a compelling economic argument for Mirai’s approach. He notes that the soaring costs of cloud-based AI inference are unsustainable for many businesses in the long term. As companies begin scrutinizing their underlying economics, demand for efficient edge computing solutions is poised to grow significantly.
“Every model maker will likely want to run part of their inference workloads at the edge,” McLoughlin stated. “Mirai feels very well positioned to capture this emerging demand.”
The seed funding round also attracted investment from a notable group of individual backers, including executives from companies like Dream, ElevenLabs, Snowflake, and former leaders from Google and Netflix. This broad support underscores the industry’s growing interest in moving AI capabilities closer to the end-user, signaling a potential transformation in how intelligent applications are built and experienced.
(Source: TechCrunch)
