Getting Started
heliaRT keeps the familiar LiteRT programming model and adds Ambiq-tuned kernels for Apollo platforms. Pick the integration path that matches your project.
Pick Your Path
-
Zephyr Module
Source module or prebuilt bundle via
west. Switch backend with a single Kconfig.Best for: product integration, Zephyr-based applications.
-
CMSIS-Pack
Planned package-manager flow for Arm ecosystem projects.
Best for: Keil MDK, CMSIS-Toolbox users.
-
neuralSPOT
Profile and deploy a
.tflitemodel withns_autodeployin minutes.Best for: fast model evaluation on Ambiq EVBs.
-
Source / CMake
Full control over target, toolchain, and build type. Link the
.ainto any project.Best for: custom build systems, source-level debugging.
Core Concepts
If you've used LiteRT for Micro before, you already know the model:
| Concept | Same in heliaRT? |
|---|---|
.tflite flatbuffer models |
✓ |
MicroMutableOpResolver |
✓ |
MicroInterpreter |
✓ |
| Statically-allocated tensor arenas | ✓ |
| Embedded logging and profiling | ✓ |
The key additions are three kernel backends (Reference, CMSIS-NN, HELIA), two build variants (SPEED, SIZE), and three toolchain options (GCC, armclang, ATfE).
Recommended Order
- Evaluating a model? Start with neuralSPOT and
ns_autodeploy. - Building a product? Move to Zephyr for module-based integration.
- Need full control? Use source builds with custom toolchain and target flags.
Related Pages
- Why heliaRT — the pitch
- Features — capabilities overview
- Examples — working integration patterns
- Upgrading from upstream LiteRT — step-by-step swap guide