TensorFlow Lite Micro

Deploy machine learning models on microcontrollers and digital signal processors using a C++ library requiring only 16 KB of memory.

TensorFlow Lite Micro (TFLM) enables on-device inference for bare-metal hardware without operating system dependencies or standard C libraries. It targets ultra-low-power chips (such as the ARM Cortex-M series, ESP32, and Xtensa) to perform complex tasks like keyword spotting and person detection. The framework uses a specialized interpreter to execute models stored in the FlatBuffers format, ensuring minimal memory overhead. By processing sensor data locally on devices like the Ambiq Apollo3, TFLM eliminates the need for cloud connectivity: this reduces latency, enhances privacy, and extends battery life for edge AI applications.

https://www.tensorflow.org/lite/microcontrollers