Complete programmatic API specifications for the Aura SDK Core engine.

The Aura SDK exposed programmatic API handles device enumeration, memory scheduling, and hardware enclave configuration.

GenieEngine

Initialize the Hexagon co-processor engine using the native Genie format for NPU execution:

use std::path::Path;
use aura_sdk::engines::GenieEngine;

fn initialize() -> Result<GenieEngine, Box<dyn std::error::Error>> {
    let config_path = Path::new("phi-config/genie_config.json");
    let engine = GenieEngine::new(config_path)?;
    Ok(engine)
}

Running Sync Queries

Executes inference synchronously on the NPU and processes output tokens using a callback closure:

engine.query_sync("Explain NPUs in one sentence.", 512, |token| {
    print!("{}", token);
})?;

AuraEngine

Initialize the ONNX Runtime execution engine utilizing QNN EP or CPU fallback:

use std::path::Path;
use aura_sdk::engines::AuraEngine;

fn initialize_ort() -> Result<AuraEngine, Box<dyn std::error::Error>> {
    let model_path = Path::new("models/model_q4.onnx");
    let engine = AuraEngine::new(model_path, "43")?;
    Ok(engine)
}

Running Queries

Executes inference on CPU fallback and processes output tokens using a callback closure:

engine.query("Explain NPUs in one sentence.", 512, |token| {
    print!("{}", token);
})?;