Skip to content

Getting Started

Welcome to the heliaRT getting-started guide. heliaRT keeps the familiar TensorFlow Lite for Microcontrollers programming model and adds Ambiq-focused runtime and kernel optimizations for Apollo platforms.

Start Here

Choose the setup path that matches how you want to evaluate or integrate heliaRT:

  • Zephyr setup: integrate heliaRT into a Zephyr west workspace using either the raw module or the prebuilt release bundle.
  • neuralSPOT setup: profile and deploy a .tflite model with ns_autodeploy using a fast Ambiq-oriented workflow.
  • Source builds: build heliaRT directly when you need a custom environment or tighter control over the build.

Core Concepts

If you have already used TFLM, the high-level model is the same:

  • .tflite flatbuffer models
  • operator resolvers
  • tensor arenas
  • MicroInterpreter-based inference
  • embedded-friendly logging and profiling

The main differences are in packaging, supported integration paths, and Ambiq-optimized kernels.

Setup Paths at a Glance

Path Best for Notes
Zephyr raw module Source-visible integration and custom builds Public-safe source path uses Reference or open CMSIS-NN; HELIA requires a separate Ambiq-provided module
Zephyr prebuilt bundle Fast-start Zephyr integration Ambiq-optimized kernels embedded in the archive
neuralSPOT with ns_autodeploy Quick profiling and deployment Good first step when evaluating a model on hardware
Source build Custom build systems and low-level integration Most flexible, but most manual
  1. Start with neuralSPOT setup if your first goal is profiling and basic model validation.
  2. Move to Zephyr setup when you are integrating heliaRT into a product or application workspace.
  3. Use source builds when you need direct control over toolchains, archives, or custom packaging.