AI-Native Design

Auto is designed from the ground up for AI workloads, with first-class support for node-based dataflow and embedded model inference.

Node-Based Dataflow

Auto's dataflow system allows you to define computation graphs as code:

graph image_pipeline {
    input -> preprocess -> model -> postprocess -> output
}

Nodes can run on:

• CPU for general computation
• GPU via CUDA/Metal/Vulkan
• NPU for edge AI devices

Embedded Model Inference

Auto can compile and embed trained models directly into your application:

• ONNX runtime integration
• Custom operators for Auto-specific optimizations
• Quantization support for edge deployment

Actor-Based Training

Distributed training is naturally expressed using Auto's actor model:

• Each GPU worker is an actor
• Parameter servers communicate via messages
• Fault tolerance through actor supervision trees