KTAutoIntents

AI-powered App Intent generator for SwiftData models.
Add one protocol, run one command, get a complete suite of Siri & Shortcuts intents.

---

What is KTAutoIntents?

Writing App Intents by hand for every data model in your app is tedious and error-prone. KTAutoIntents automates this:

1. You conform your @Model class to AutoIntentable and describe what it represents.
2. The CLI reads your model, sends the schema to an LLM, and writes a complete, production-ready *Intents.swift file.
3. You add that file to Xcode. Done — your data is searchable via Siri, Spotlight, and the Shortcuts app.

┌──────────────────────┐        ┌─────────────┐        ┌─────────────────────────┐
│  Your @Model class   │──────▶│ LLM (Gemini │──────▶│  RecipeIntents.swift     │
│  Recipe.swift        │ schema │  or OpenAI) │ Swift  │  AppEntity, EntityQuery  │
│  + AutoIntentable    │        └─────────────┘  code  │  5 AppIntent subclasses  │
└──────────────────────┘                               │  AppShortcutsProvider    │
                                                       └─────────────────────────┘

The generated code is committed to your project — no LLM calls at build time or runtime.

---

Installation

Runtime Library (Swift Package Manager)

In Xcode: File → Add Package Dependencies… and paste:

https://github.com/ktaniguchi/KTAutoIntents

Or add to Package.swift:

.package(url: "https://github.com/ktaniguchi/KTAutoIntents", from: "1.0.0")

Then add KTAutoIntents to your target's dependencies.

CLI Generator

git clone https://github.com/ktaniguchi/KTAutoIntents
cd KTAutoIntents
swift build -c release --product KTAutoIntentsCLI
sudo cp .build/release/KTAutoIntentsCLI /usr/local/bin/

---

Quick Start

Step 1 — Conform your model

import SwiftData
import KTAutoIntents

@Model
final class Recipe: AutoIntentable {
    var name: String
    var cuisineType: String
    var preparationTime: Int   // minutes
    var isVegetarian: Bool
    var difficultyLevel: String

    // MARK: AutoIntentable metadata

    static var intentDisplayName: String { "Recipe" }
    static var intentDescription: String {
        "A cooking recipe with cuisine type and preparation time"
    }
    static var intentKeywords: [String] { ["food", "cooking", "meal"] }
    static var intentSearchablePropertyNames: [String] {
        ["name", "cuisineType", "isVegetarian", "preparationTime"]
    }
}

Step 2 — Run the generator

# Set your API key (Gemini is the default)
export GEMINI_API_KEY=your_key_here
# or for OpenAI:
# export OPENAI_API_KEY=your_key_here

KTAutoIntentsCLI generate \
    --path ./MyApp/Models \
    --output ./MyApp/GeneratedIntents

Output:

┌─────────────────────────────────────────────┐
│  KTAutoIntents — App Intent Generator        │
│  https://github.com/ktaniguchi/KTAutoIntents │
└─────────────────────────────────────────────┘

🔍 Scanning: ./MyApp/Models
   Found 2 model(s): Recipe, Ingredient

🤖 Provider: gemini / gemini-2.0-flash
📝 Output: ./MyApp/GeneratedIntents

   Generating intents for Recipe… ✅ RecipeIntents.swift
   Generating intents for Ingredient… ✅ IngredientIntents.swift

🎉 Done. 2/2 file(s) generated in ./MyApp/GeneratedIntents

Step 3 — Add to Xcode & configure

1. Drag the generated *Intents.swift files into your Xcode target.
2. Register the model container once in your app entry point:

import SwiftUI
import SwiftData
import KTAutoIntents

@main
struct RecipeApp: App {
    init() {
        IntentConfiguration.shared.register {
            try ModelContainer(for: Recipe.self)
        }
    }
    var body: some Scene {
        WindowGroup { ContentView() }
    }
}

That's it — build and run. Your recipes are now searchable via Siri ("Find quick recipes in Recipe App"), Spotlight, and the Shortcuts app.

---

CLI Reference

USAGE: KTAutoIntentsCLI generate [options]

OPTIONS:
  -p, --path <path>             Directory containing your @Model source files. (required)
  -o, --output <output>         Output directory for generated files. (default: ./GeneratedIntents)
      --provider <provider>     LLM provider: gemini (default) | openai
      --model <model>           LLM model name override (e.g. gemini-2.0-flash, gpt-4o)
      --dry-run                 Print the LLM prompts without calling the API.
      --skip-validation         Skip swiftc syntax validation of generated code.
  -h, --help                    Show help information.

API Keys

Provider	Environment Variable
Gemini (default)	GEMINI_API_KEY
OpenAI	OPENAI_API_KEY

---

AutoIntentable Protocol

public protocol AutoIntentable {
    /// Singular human-readable name (e.g. "Recipe")
    static var intentDisplayName: String { get }

    /// Brief description to help the LLM understand the entity
    static var intentDescription: String { get }

    /// Siri search keywords (default: [])
    static var intentKeywords: [String] { get }

    /// Properties the LLM should focus intents on (default: inferred)
    static var intentSearchablePropertyNames: [String] { get }
}

The generator infers searchable properties automatically from type (String, Bool, numeric) when intentSearchablePropertyNames is empty.

---

What Gets Generated?

For each @Model, the generator produces a single Swift file containing:

Component	Description
AppEntity extension	Conformance with display representation
EntityQuery	SwiftData-backed entity lookup for system
3–7 AppIntent subclasses	Natural-language intents (search, filter, range, etc.)
AppShortcutsProvider	Registers top intents with Siri phrase templates
ModelContainer helper	Private extension for container access

See Examples/RecipeApp/GeneratedIntents.swift for a real example.

---

Regenerating After Model Changes

Re-run the CLI after changing your model's properties. The existing *Intents.swift file will be overwritten with a fresh generation.

KTAutoIntentsCLI generate --path ./MyApp/Models --output ./MyApp/GeneratedIntents

Consider adding this as a Makefile target or a custom Xcode run script phase (disabled by default so it only runs on demand).

---

Contributing

Contributions are welcome! See CONTRIBUTING.md for guidelines.

Pull requests for new LLM providers, improved prompt engineering, and additional example models are especially appreciated.

---

License

MIT — see LICENSE.