Getting started

heliaRT is Ambiq’s optimized fork of LiteRT for Microcontrollers. It keeps the model, interpreter and operator-resolver programming model while integrating HELIA kernel adapters.

Start with your integration

Choose your path to connect heliaRT to Zephyr, neuralSPOT-X, a CMSIS-Pack project or your own firmware build.

Zephyr

For applications built with west and Kconfig.

neuralSPOT-X

For applications built with Ambiq’s development SDK.

CMSIS-Pack

For projects managed with CMSIS tooling.

Build from source

For control over compiler options and kernel selection.

Prebuilt archive

For firmware that links a released runtime library.

Have a runtime library integrated already? Follow First inference for the sequence from model loading to output handling.

What you need

A .tflite model whose operators and tensor types are supported by your selected runtime backend.
A toolchain and runtime library built for the same target and compatible compiler options as your application.
Application-owned memory for the model, resolver and tensor arena.
A way to feed model inputs and inspect its outputs on your target.

Source builds and prebuilt libraries have different configuration boundaries. Choose the distribution before setting kernel and floating-point options.

Learn the runtime contract

Runtime concepts explains model lifetime, arena ownership, operator registration and backend selection. The runtime API links each core type to its header.