AI & Tensors

The current AI/tensor language and plugin direction, separated from broader “AI-native” marketing claims.

Aura’s current compact reference includes resource-like types such as Tensor and Model, and the workspace contains aura-plugin-ai, aura-ai-opt plus ONNX Runtime bridge assets/examples.

Shape-aware direction

Tensor shapes can be represented in type syntax such as:

text
Tensor<Elem, [d0, d1, ...]>

This gives the language and verifier a place to represent shape information rather than treating every tensor as an opaque runtime pointer.

ONNX integration

The repository contains an Aura Vision example tree plus ONNX Runtime bridge headers/source. That is evidence of a real integration path.

It does not automatically establish that every ONNX operation/model is verified, optimized or production-supported by Aura.

AI plugin

aura-plugin-ai is an optional domain plugin. Public documentation should describe concrete builtins/examples supported by current code, not elevate the presence of the plugin into a universal AI framework claim.

Resource semantics

Tensor and Model are included among resource-like types in the current reference’s move-oriented MVP rules. Broader ownership/linearity infrastructure exists in the compiler and verifier, but those guarantees remain pre-stable.