oai-swift/oAI/Providers/AppleFoundationProvider.swift

//
//  AppleFoundationProvider.swift
//  oAI
//
//  Apple Foundation Models provider (on-device Apple Intelligence)
//
// SPDX-License-Identifier: AGPL-3.0-or-later
// Copyright (C) 2026 Rune Olsen
//
// This file is part of oAI.
//
// oAI is free software: you can redistribute it and/or modify
// it under the terms of the GNU Affero General Public License as
// published by the Free Software Foundation, either version 3 of the
// License, or (at your option) any later version.
//
// oAI is distributed in the hope that it will be useful, but WITHOUT
// ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
// or FITNESS FOR A PARTICULAR PURPOSE. See the GNU Affero General
// Public License for more details.
//
// You should have received a copy of the GNU Affero General Public
// License along with oAI. If not, see <https://www.gnu.org/licenses/>.


import Foundation
import FoundationModels
import os

final class AppleFoundationProvider: AIProvider {
    let name = "Apple Intelligence"

    let capabilities = ProviderCapabilities(
        supportsStreaming: true,
        supportsVision: false,
        supportsTools: false,
        supportsOnlineSearch: false,
        maxContextLength: 4096
    )

    // MARK: - Models

    func listModels() async throws -> [ModelInfo] {
        [
            ModelInfo(
                id: "apple-on-device",
                name: "Apple On-Device",
                description: "On-device Apple Intelligence model. Private, free, and works offline. 4K context window.",
                contextLength: 4096,
                pricing: ModelInfo.Pricing(prompt: 0, completion: 0),
                capabilities: ModelInfo.ModelCapabilities(
                    vision: false,
                    tools: false,
                    online: false
                )
            )
        ]
    }

    func getModel(_ id: String) async throws -> ModelInfo? {
        try await listModels().first { $0.id == id }
    }

    func getCredits() async throws -> Credits? { nil }

    // MARK: - Streaming chat

    func streamChat(request: ChatRequest) -> AsyncThrowingStream<StreamChunk, Error> {
        AsyncThrowingStream { continuation in
            Task {
                do {
                    let session = try self.makeSession(for: request)
                    let prompt = self.lastUserMessage(from: request)

                    // streamResponse(to: String) → ResponseStream<String>
                    // Each snapshot.content is the full accumulated text so far (snapshot model).
                    // We compute deltas by comparing each snapshot to the previous.
                    let stream = session.streamResponse(to: prompt)
                    var lastContent = ""

                    for try await snapshot in stream {
                        let current = snapshot.content
                        if current.count > lastContent.count {
                            let delta = String(current.dropFirst(lastContent.count))
                            continuation.yield(StreamChunk(
                                id: UUID().uuidString,
                                model: request.model,
                                delta: StreamChunk.Delta(content: delta, role: "assistant"),
                                finishReason: nil,
                                usage: nil
                            ))
                            lastContent = current
                        }
                    }

                    continuation.yield(StreamChunk(
                        id: UUID().uuidString,
                        model: request.model,
                        delta: StreamChunk.Delta(content: nil, role: nil),
                        finishReason: "stop",
                        usage: nil
                    ))
                    continuation.finish()

                } catch let genError as LanguageModelSession.GenerationError {
                    continuation.finish(throwing: self.mapGenerationError(genError))
                } catch {
                    continuation.finish(throwing: error)
                }
            }
        }
    }

    // MARK: - Non-streaming chat

    func chat(request: ChatRequest) async throws -> ChatResponse {
        let session = try makeSession(for: request)
        let prompt = lastUserMessage(from: request)
        let response: LanguageModelSession.Response<String> = try await session.respond(to: prompt)
        return ChatResponse(
            id: UUID().uuidString,
            model: request.model,
            content: response.content,
            role: "assistant",
            finishReason: "stop",
            usage: nil,
            created: Date()
        )
    }

    // MARK: - Tool messages (not supported in Phase 1)

    func chatWithToolMessages(model: String, messages: [[String: Any]], tools: [Tool]?, maxTokens: Int?, temperature: Double?) async throws -> ChatResponse {
        throw ProviderError.unknown("Tool calling requires Apple Foundation Models Phase 3.")
    }

    // MARK: - Session construction

    private func makeSession(for request: ChatRequest) throws -> LanguageModelSession {
        guard case .available = SystemLanguageModel.default.availability else {
            throw availabilityError()
        }

        // Build instructions: system prompt + prior conversation turns as formatted text.
        // Foundation Models sessions don't accept a message array — we inject history inline.
        var instructions = request.systemPrompt ?? ""
        let priorMessages = request.messages.dropLast().filter { $0.role != .system }

        if !priorMessages.isEmpty {
            let history = priorMessages
                .map { m -> String in
                    let label = m.role == .user ? "User" : "Assistant"
                    return "\(label): \(m.content)"
                }
                .joined(separator: "\n")
            instructions += "\n\nConversation so far:\n\(history)\n\nContinue from here."
        }

        return instructions.isEmpty
            ? LanguageModelSession()
            : LanguageModelSession(instructions: instructions)
    }

    private func lastUserMessage(from request: ChatRequest) -> String {
        request.messages.last(where: { $0.role == .user })?.content ?? ""
    }

    // MARK: - Error mapping

    private func availabilityError() -> Error {
        switch SystemLanguageModel.default.availability {
        case .unavailable(.deviceNotEligible):
            return ProviderError.unknown("This Mac doesn't support Apple Intelligence. Apple Silicon is required.")
        case .unavailable(.appleIntelligenceNotEnabled):
            return ProviderError.unknown("Apple Intelligence is not enabled. Open System Settings → Apple Intelligence to turn it on.")
        case .unavailable(.modelNotReady):
            return ProviderError.unknown("Apple Intelligence model is still downloading. Please wait and try again.")
        default:
            return ProviderError.unknown("Apple Intelligence is not available on this device.")
        }
    }

    private func mapGenerationError(_ error: LanguageModelSession.GenerationError) -> Error {
        switch error {
        case .exceededContextWindowSize:
            return ProviderError.unknown("Apple Intelligence context limit exceeded (4,096 tokens). Start a new chat or enable Progressive Summarization in Settings → Advanced.")
        case .rateLimited:
            return ProviderError.rateLimitExceeded
        case .guardrailViolation:
            return ProviderError.unknown("Apple Intelligence declined to respond to this message.")
        default:
            return error
        }
    }
}