The SDK, slated for release in Q4 2026, incorporates LiteRT—formerly known as TensorFlow Lite—into the Linux-based firmware of upcoming cellular modules. This configuration allows developers to load standard .tflite files onto hardware without the typical overhead of proprietary rebuilds or complex software stacks. By enabling direct model transfers from PCs or development boards like Raspberry Pi, Telit Cinterion intends to simplify the deployment of intelligence in resource-constrained environments.

In testing, the runtime consumed a maximum of 17% of CPU resources during image classification and object detection, a threshold chosen to prevent thermal throttling that could otherwise disrupt cellular connectivity. Marco Argenton, senior vice president of product management, noted that the goal is to prevent industrial teams from needing to overhaul existing device architectures to integrate smart features. The SDK includes reference models and sample applications for tasks such as predictive maintenance on motors, acoustic monitoring for infrastructure security, and analog meter reading via computer vision. By keeping the logic on the module, system integrators gain a streamlined path from prototype to field deployment.