What library did Google release in December 2025, designed to run AI models on resource-constrained platforms like browsers and microcontrollers?

Google unveiled LiteRT in December 2025, a crucial library engineered to bring the power of artificial intelligence to platforms with limited resources, such as web browsers and microcontrollers. This development marks a significant step in making AI more pervasive and accessible, moving beyond large data centers to devices we interact with daily. LiteRT is designed to efficiently execute AI models directly on these "edge" devices, enabling intelligent functionalities without constant reliance on cloud connectivity.

The need for a solution like LiteRT stems from the inherent limitations of traditional cloud-based AI, which can suffer from latency, privacy concerns, and a lack of reliability due to network (Review) dependencies. By running AI models on-device, LiteRT ensures faster response times, enhanced user privacy as data remains local, and robust performance even without an internet connection. It supports a wide array of platforms including Android, iOS, embedded Linux systems, and microcontrollers, with APIs for languages like Java, Kotlin, Swift, Embedded C, and C++.

LiteRT represents the evolution of Google's efforts in on-device machine learning, building upon the foundation (Review) of its predecessor, TensorFlow Lite. Its focus is on optimizing AI for deployment on devices with severely restricted memory, processing power, and energy budgets. This includes everything from smart appliances and industrial sensors to wearable devices. By simplifying the process of deploying AI on these resource-constrained platforms, LiteRT aims to accelerate the adoption of edge AI, much like TensorFlow propelled general AI development. It offers improved GPU and NPU acceleration and supports models from various popular frameworks, including PyTorch, JAX, and Keras.