Laya MLX Swift: run Laya inside a macOS app with no Python or helper process

# Laya MLX Swift

Run [Laya](https://github.com/NandhaKishorM/laya) directly inside a macOS app with Apple MLX.

- Native Swift library
- Runs locally on Apple silicon
- No Python, ONNX, helper process, or local server
- Downloads the model only when your app calls `prepare`

## Add the package

In Xcode, add:

```text
https://github.com/MstyAI/laya-mlx-swift
```

Then add `LayaMLX` to your target.

## Prepare once

Call this from a button or setup screen. The download is about 804 MiB.

```swift
import LayaMLX

let modelDirectory = try await LayaModel.prepare { progress in
    print("Downloaded \(Int(progress.fractionCompleted * 100))%")
}
```

`prepare` does nothing when the model is already present. Inference never starts a download.

## Make a decision

```swift
let laya = try await LayaAgent(modelDirectory: modelDirectory)

let result = try await laya.predict(LayaRequest(
    state: "The payment was declined twice and the customer is angry.",
    questions: [
        "route": .choice(
            instructions: "Choose the best destination.",
            options: [
                LayaOption("billing", description: "Payment and invoice problems"),
                LayaOption("technical", description: "Product bugs and outages"),
                LayaOption("sales", description: "Purchasing questions"),
            ]
        ),
        "urgent": .boolean(
            instructions: "This needs urgent human attention."
        ),
    ]
)))

print(result.answers["route"]?.choice ?? "unknown")
print(result.answers["urgent"]?.booleanProbability ?? 0)
```

Laya supports three question types:

| Type | Use it for | Result |
|---|---|---|
| `choice` | Routing or selecting an action | Winning label and probabilities |
| `boolean` | Checking whether a statement holds | Probabili