APIs

Python and C/C++ interfaces for building applications that run inference on Metis hardware.

Python​

API	What it covers
InferenceStream API	The recommended Python API for inference applications. create_inference_stream, FrameResult, metadata types (detection, tracking, classification, etc.), display, and tracers.
axelera.runtime API	The low-level Python API for direct AIPU access. Context, Connection, Model, tensor management. Use this when you need fine-grained control over device selection, buffer management, or multi-instance execution.
Pipeline Builder API	The Pythonic API for composable ML pipelines using axelera.runtime.op. Build pipelines with operators for transforms, inference, postprocessing, tracking, and more. Includes quickstart, model compilation guide, and full operator reference.

C/C++​

API	What it covers
AxInferenceNet C++ Reference	The recommended C++ API for inference applications. An asynchronous pipeline handling multiple streams and models, with preprocessing, dequantization and decoding plugins. Usable standalone or through the axinferencenet GStreamer element.
AxRuntime C/C++ API	The low-level C API for direct AIPU access, callable from C or C++. The same object model as axelera.runtime: context, connection, model, model instance, tensor and device management.

For most applications, start with the InferenceStream API and the Run Inference in Python tutorial. For C++ applications, start with AxInferenceNet and the AxInferenceNet C++ Example.