SwiftFM

Super-simple API for Apple Foundation Models in Swift (iOS 26+, macOS 26+).
A lightweight façade designed for beginners: generate text or strongly-typed results in one line.

---

Features

• ✅ Plain text generation — pass a prompt, get back a string.
• ✅ Guided typed generation — decode directly into your own Decodable & Generable structs.
• ✅ Streaming Text Generation — get incremental text chunks as they’re produced
• ✅ Beginner-friendly API — minimal surface, sensible defaults.
• ✅ On-device — powered by Apple’s Foundation Models framework, no server required.
• ✅ Future-proof — extend with streaming, tools, and advanced options later.

---

Availability

SwiftFM requires the Apple Intelligence Foundation Models, which are only available on:

• iOS 26 or later
• iPadOS 26 or later
• macOS 26 or later

---

Swift 6 Readiness

SwiftFM is fully Swift 6 ready.

• All types conform to Sendable where appropriate.
• Concurrency is handled safely with actor isolation.
• Package settings enable Strict Concurrency checking for maximum safety.

---

Installation

Add SwiftFM to your project using Swift Package Manager:

1. In Xcode: File → Add Packages…
2. Enter the repository URL: https://github.com/ricky-stone/SwiftFM
3. Select main branch or a tagged version (e.g. 0.1.0).

---

Usage

1. Plain Text Generation

import SwiftFM


let fm = SwiftFM()

Task {
    do {
        let response = try await fm.generateText(
            for: "Explain a century break in snooker in one sentence."
        )
        print(response)
        // "A century break is when a player scores 100 points or more in a single visit."
    } catch {
        print("Error:", error)
    }
}

2. Strongly-Typed Guided Generation

To use guided generation, your type must conform to Decodable & Sendable & Generable.

@Generable
struct MatchPrediction: Decodable, Sendable {
    @Guide(description: "First player’s name")
    let player: String

    @Guide(description: "Opponent’s name")
    let opponent: String

    @Guide(description: "Who is predicted to win")
    let predictedWinner: String

    @Guide(description: "Confidence 0.0–1.0")
    let confidence: Double
}

let fm = SwiftFM()

Task {
    do {
        let prediction: MatchPrediction = try await fm.generateJSON(
            for: """
            Imagine a snooker match.
            Choose two well-known players and predict the winner.
            Return {player, opponent, predictedWinner, confidence}.
            """,
            as: MatchPrediction.self
        )

        print("Upcoming match: \(prediction.player) vs \(prediction.opponent)")
        print("Predicted winner: \(prediction.predictedWinner) (\(Int(prediction.confidence * 100))%)")
    } catch {
        print("Error:", error)
    }
}

3. Using a System Prompt

import SwiftFM

// Configure SwiftFM with a system role
let fm = SwiftFM(config: .init(system: "You are a professional snooker coach."))

Task {
    do {
        let tip = try await fm.generateText(
            for: "Give me a practice drill to improve cue ball control."
        )
        print(tip)
        // e.g. "Set up three reds in a line and practice stopping the cue ball dead after each pot."
    } catch {
        print("Error:", error)
    }
}

4. Streaming Text Generation

let fm = SwiftFM()

Task {
    do {
        for try await chunk in await fm.streamText(for: "Explain the rules of snooker step by step.") {
            print(chunk, terminator: "")
        }
    } catch {
        print("Stream failed:", error)
    }
}

SwiftUI Example

@State private var text = ""

Task {
    do {
        for try await chunk in await fm.streamText(for: "Explain the rules of snooker step by step.") {
            text = chunk
        }
    } catch {
        print("Stream failed:", error)
    }
}

Check availability at runtime:

if SwiftFM.isModelAvailable {
    print("Foundation Model ready to use.")
} else {
    print("Not available:", SwiftFM.modelAvailability)
}

Testing

SwiftFM includes Swift Testing style tests to ensure everything works as expected on devices where Foundation Models are available.

License

SwiftFM is released under the MIT License.

This means: • You are free to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of this software. • The only requirements are that you include the original license and copyright notice in copies or substantial portions of the software. • The software is provided “as is”, without warranty of any kind.

Attribution

SwiftFM was created by Ricky Stone.
If you use or modify this library, please consider including a reference such as:
“Based on Ricky Stone’s SwiftFM.”

This is not a requirement of the MIT License, but an appreciated courtesy.