A machine learning toolkit designed for ease of use from prototyping to deployment, with memory safety and zero-cost abstractions.

What Hodu does:

- hodu-lib: Core ML library providing memory-safe tensor operations and model building through Rust's ownership system and zero-cost abstractions, with support for multiple backends (CPU, CUDA, Metal)

- hodu-cli: Command-line interface for model inference, format conversion, cross compile, and AOT compilation to native libraries, with built-in plugin management

- hodu-plugin-sdk: SDK for building custom plugins via JSON-RPC, enabling support for new model formats (ONNX, TensorFlow), tensor formats (NumPy, SafeTensors), and execution backends