swift-apple-intelligence-grpc/Sources/AppleIntelligenceGRPC/Services/AppleIntelligenceService.swift

import Foundation
import FoundationModels

/// Errors that can occur when using Apple Intelligence
enum AppleIntelligenceError: Error, CustomStringConvertible {
    case modelNotAvailable
    case generationFailed(String)
    case sessionCreationFailed

    var description: String {
        switch self {
        case .modelNotAvailable:
            return "Apple Intelligence model is not available on this device"
        case .generationFailed(let reason):
            return "Generation failed: \(reason)"
        case .sessionCreationFailed:
            return "Failed to create language model session"
        }
    }
}

/// Service wrapper for Apple Intelligence Foundation Models
actor AppleIntelligenceService {
    /// The language model session
    private var session: LanguageModelSession?

    /// Whether the model is available
    private(set) var isAvailable: Bool = false

    /// Initialize and check model availability
    init() async {
        await checkAvailability()
    }

    /// Check if Apple Intelligence is available
    private func checkAvailability() async {
        let availability = SystemLanguageModel.default.availability
        switch availability {
        case .available:
            isAvailable = true
            session = LanguageModelSession()
        case .unavailable:
            isAvailable = false
        @unknown default:
            isAvailable = false
        }
    }

    /// Get the current model status as a string
    func getModelStatus() -> String {
        let availability = SystemLanguageModel.default.availability
        switch availability {
        case .available:
            return "available"
        case .unavailable(let reason):
            return "unavailable: \(reason)"
        @unknown default:
            return "unknown"
        }
    }

    /// Generate a completion for the given prompt (non-streaming)
    func complete(prompt: String, temperature: Float?, maxTokens: Int?) async throws -> String {
        guard isAvailable, let session = session else {
            throw AppleIntelligenceError.modelNotAvailable
        }

        let response = try await session.respond(to: prompt)
        return response.content
    }

    /// Generate a streaming completion for the given prompt
    func streamComplete(
        prompt: String,
        temperature: Float?,
        maxTokens: Int?
    ) -> AsyncThrowingStream<String, Error> {
        AsyncThrowingStream { continuation in
            Task {
                guard self.isAvailable, let session = self.session else {
                    continuation.finish(throwing: AppleIntelligenceError.modelNotAvailable)
                    return
                }

                do {
                    let stream = session.streamResponse(to: prompt)
                    for try await partialResponse in stream {
                        continuation.yield(partialResponse.content)
                    }
                    continuation.finish()
                } catch {
                    continuation.finish(throwing: AppleIntelligenceError.generationFailed(error.localizedDescription))
                }
            }
        }
    }
}