Mirai has secured $10 million in Seed funding to accelerate the development of its on-device AI execution layer, aiming to make local inference practical and accessible for developers.
The round was led by Uncork Capital.
Lowering The Barrier To Local AI Execution
Although modern laptops and smartphones are equipped with dedicated AI chips, most AI applications still rely heavily on cloud processing. Running models directly on devices typically demands deep technical expertise in areas such as memory optimization, hardware-level tuning, and performance engineering.
Mirai is building an abstraction layer that removes this complexity.
The company has developed a proprietary inference engine tailored for Apple Silicon, allowing developers to deploy models locally without engaging with low-level systems work. Applications can maintain their existing cloud architecture while optionally routing certain workloads to the device for hybrid execution.
In benchmark scenarios across selected model–device configurations, Mirai reports measurable improvements in both response speed and model initialization compared to common open-source runtimes.
Advertisement
Optimizing The Interaction Between Model And Hardware
Mirai’s approach centers on the interaction between three core components: the model, the runtime environment, and the hardware.
While many AI teams focus primarily on improving model performance, Mirai concentrates on optimizing how models are executed on specific hardware architectures. By introducing a hardware-aware runtime layer, the company aims to unlock better performance from smaller models, reduce latency, and shift the cost dynamics of AI applications.
In this framework, inference is not just a backend service but a foundational execution layer embedded directly into software systems.
A Move Toward Device-Native Intelligence
Mirai sees AI evolving toward a more device-centric model. Rather than depending exclusively on cloud infrastructure, applications are beginning to leverage compact, efficient models that run continuously on personal devices.
These models may trade breadth of knowledge for responsiveness and deep system integration. Benefits include lower interaction latency, direct access to local data and context, improved privacy controls, and the ability to function offline.
As this shift accelerates, inference becomes part of the operating layer of applications rather than an external dependency.
Next Steps
With the new funding, Mirai plans to extend its runtime support beyond Apple Silicon and expand into additional modalities including voice, vision, and multimodal AI workloads.
The company’s long-term objective is to make local inference the standard execution path for AI-driven applications.
About Mirai
Mirai develops an optimized inference runtime designed to simplify and accelerate on-device AI deployment. By bridging the gap between models and hardware, the company enables developers to run AI workloads locally with improved efficiency, performance, and integration.
