oai-swift/oAI/Services/EmbeddingService.swift

//
//  EmbeddingService.swift
//  oAI
//
//  Embedding generation and semantic search
//  Supports multiple providers: OpenAI, OpenRouter, Google
//
// SPDX-License-Identifier: PolyForm-Noncommercial-1.0.0
// Copyright (C) 2026 Rune Olsen
//
// This file is part of oAI.
//
// oAI is licensed under the PolyForm Noncommercial License 1.0.0.
// You may use, study, modify, and share it for any noncommercial
// purpose. Commercial use — including selling oAI or any part of
// it, standalone or bundled into another product or service —
// requires a separate commercial license from the copyright holder.
//
// See the LICENSE file or
// <https://polyformproject.org/licenses/noncommercial/1.0.0> for
// the full license text. For commercial licensing, contact Rune
// Olsen via <https://oai.pm>.


import Foundation
import os

// MARK: - Embedding Provider

enum EmbeddingProvider {
    case openai(model: String)
    case openrouter(model: String)
    case google(model: String)

    var defaultModel: String {
        switch self {
        case .openai: return "text-embedding-3-small"
        case .openrouter: return "openai/text-embedding-3-small"
        case .google: return "text-embedding-004"
        }
    }

    var dimension: Int {
        switch self {
        case .openai(let model):
            return model == "text-embedding-3-large" ? 3072 : 1536
        case .openrouter(let model):
            if model.contains("text-embedding-3-large") {
                return 3072
            } else if model.contains("qwen3-embedding-8b") {
                return 8192
            } else {
                return 1536 // Default for most models
            }
        case .google:
            return 768
        }
    }

    var displayName: String {
        switch self {
        case .openai: return "OpenAI"
        case .openrouter: return "OpenRouter"
        case .google: return "Google"
        }
    }
}

// MARK: - Embedding Service

final class EmbeddingService {
    nonisolated static let shared = EmbeddingService()

    private let settings = SettingsService.shared

    /// Dedicated session for embedding requests — keeps embedding traffic isolated
    /// from the chat API sessions and self-limits concurrent connections.
    private let session: URLSession = {
        let config = URLSessionConfiguration.default
        config.httpMaximumConnectionsPerHost = 2  // max 2 concurrent embedding requests
        config.timeoutIntervalForRequest = 60
        config.timeoutIntervalForResource = 120
        return URLSession(configuration: config)
    }()

    private init() {}

    // MARK: - Provider Detection

    /// Get the embedding provider based on user's selection in settings
    func getSelectedProvider() -> EmbeddingProvider? {
        let selectedModel = settings.embeddingProvider

        // Map user's selection to provider
        switch selectedModel {
        case "openai-small":
            guard let key = settings.openaiAPIKey, !key.isEmpty else { return nil }
            return .openai(model: "text-embedding-3-small")
        case "openai-large":
            guard let key = settings.openaiAPIKey, !key.isEmpty else { return nil }
            return .openai(model: "text-embedding-3-large")
        case "openrouter-openai-small":
            guard let key = settings.openrouterAPIKey, !key.isEmpty else { return nil }
            return .openrouter(model: "openai/text-embedding-3-small")
        case "openrouter-openai-large":
            guard let key = settings.openrouterAPIKey, !key.isEmpty else { return nil }
            return .openrouter(model: "openai/text-embedding-3-large")
        case "openrouter-qwen":
            guard let key = settings.openrouterAPIKey, !key.isEmpty else { return nil }
            return .openrouter(model: "qwen/qwen3-embedding-8b")
        case "google-gemini":
            guard let key = settings.googleAPIKey, !key.isEmpty else { return nil }
            return .google(model: "text-embedding-004")
        default:
            // Fall back to best available
            return getBestAvailableProvider()
        }
    }

    /// Get the best available embedding provider based on user's API keys (priority: OpenAI → OpenRouter → Google)
    func getBestAvailableProvider() -> EmbeddingProvider? {
        // Priority: OpenAI → OpenRouter → Google
        if let key = settings.openaiAPIKey, !key.isEmpty {
            return .openai(model: "text-embedding-3-small")
        }

        if let key = settings.openrouterAPIKey, !key.isEmpty {
            return .openrouter(model: "openai/text-embedding-3-small")
        }

        if let key = settings.googleAPIKey, !key.isEmpty {
            return .google(model: "text-embedding-004")
        }

        return nil
    }

    /// Check if embeddings are available (user has at least one compatible provider)
    var isAvailable: Bool {
        return getBestAvailableProvider() != nil
    }

    // MARK: - Embedding Generation

    /// Generate embedding for text using the configured provider
    func generateEmbedding(text: String, provider: EmbeddingProvider) async throws -> [Float] {
        switch provider {
        case .openai(let model):
            return try await generateOpenAIEmbedding(text: text, model: model)
        case .openrouter(let model):
            return try await generateOpenRouterEmbedding(text: text, model: model)
        case .google(let model):
            return try await generateGoogleEmbedding(text: text, model: model)
        }
    }

    /// Generate OpenAI embedding
    private func generateOpenAIEmbedding(text: String, model: String) async throws -> [Float] {
        guard let apiKey = settings.openaiAPIKey, !apiKey.isEmpty else {
            throw EmbeddingError.missingAPIKey("OpenAI")
        }

        let url = URL(string: "https://api.openai.com/v1/embeddings")!
        var request = URLRequest(url: url)
        request.httpMethod = "POST"
        request.setValue("Bearer \(apiKey)", forHTTPHeaderField: "Authorization")
        request.setValue("application/json", forHTTPHeaderField: "Content-Type")

        let body: [String: Any] = [
            "input": text,
            "model": model
        ]
        request.httpBody = try JSONSerialization.data(withJSONObject: body)

        let (data, response) = try await session.data(for: request)

        guard let httpResponse = response as? HTTPURLResponse else {
            throw EmbeddingError.invalidResponse
        }

        guard httpResponse.statusCode == 200 else {
            let errorMessage = String(data: data, encoding: .utf8) ?? "Unknown error"
            Log.api.error("OpenAI embedding error (\(httpResponse.statusCode)): \(errorMessage)")
            throw EmbeddingError.apiError(httpResponse.statusCode, errorMessage)
        }

        let json = try JSONSerialization.jsonObject(with: data) as? [String: Any]
        guard let dataArray = json?["data"] as? [[String: Any]],
              let first = dataArray.first,
              let embedding = first["embedding"] as? [Double] else {
            throw EmbeddingError.invalidResponse
        }

        return embedding.map { Float($0) }
    }

    /// Generate OpenRouter embedding (OpenAI-compatible API)
    private func generateOpenRouterEmbedding(text: String, model: String) async throws -> [Float] {
        guard let apiKey = settings.openrouterAPIKey, !apiKey.isEmpty else {
            throw EmbeddingError.missingAPIKey("OpenRouter")
        }

        let url = URL(string: "https://openrouter.ai/api/v1/embeddings")!
        var request = URLRequest(url: url)
        request.httpMethod = "POST"
        request.setValue("Bearer \(apiKey)", forHTTPHeaderField: "Authorization")
        request.setValue("application/json", forHTTPHeaderField: "Content-Type")
        request.setValue("https://github.com/yourusername/oAI", forHTTPHeaderField: "HTTP-Referer")

        let body: [String: Any] = [
            "input": text,
            "model": model
        ]
        request.httpBody = try JSONSerialization.data(withJSONObject: body)

        let (data, response) = try await session.data(for: request)

        guard let httpResponse = response as? HTTPURLResponse else {
            throw EmbeddingError.invalidResponse
        }

        guard httpResponse.statusCode == 200 else {
            let errorMessage = String(data: data, encoding: .utf8) ?? "Unknown error"
            Log.api.error("OpenRouter embedding error (\(httpResponse.statusCode)): \(errorMessage)")
            throw EmbeddingError.apiError(httpResponse.statusCode, errorMessage)
        }

        let json = try JSONSerialization.jsonObject(with: data) as? [String: Any]
        guard let dataArray = json?["data"] as? [[String: Any]],
              let first = dataArray.first,
              let embedding = first["embedding"] as? [Double] else {
            throw EmbeddingError.invalidResponse
        }

        return embedding.map { Float($0) }
    }

    /// Generate Google embedding
    private func generateGoogleEmbedding(text: String, model: String) async throws -> [Float] {
        guard let apiKey = settings.googleAPIKey, !apiKey.isEmpty else {
            throw EmbeddingError.missingAPIKey("Google")
        }

        let url = URL(string: "https://generativelanguage.googleapis.com/v1beta/models/\(model):embedContent?key=\(apiKey)")!
        var request = URLRequest(url: url)
        request.httpMethod = "POST"
        request.setValue("application/json", forHTTPHeaderField: "Content-Type")

        let body: [String: Any] = [
            "content": [
                "parts": [
                    ["text": text]
                ]
            ]
        ]
        request.httpBody = try JSONSerialization.data(withJSONObject: body)

        let (data, response) = try await session.data(for: request)

        guard let httpResponse = response as? HTTPURLResponse else {
            throw EmbeddingError.invalidResponse
        }

        guard httpResponse.statusCode == 200 else {
            let errorMessage = String(data: data, encoding: .utf8) ?? "Unknown error"
            Log.api.error("Google embedding error (\(httpResponse.statusCode)): \(errorMessage)")
            throw EmbeddingError.apiError(httpResponse.statusCode, errorMessage)
        }

        let json = try JSONSerialization.jsonObject(with: data) as? [String: Any]
        guard let embedding = json?["embedding"] as? [String: Any],
              let values = embedding["values"] as? [Double] else {
            throw EmbeddingError.invalidResponse
        }

        return values.map { Float($0) }
    }

    // MARK: - Similarity Calculation

    /// Calculate cosine similarity between two embeddings
    nonisolated func cosineSimilarity(_ a: [Float], _ b: [Float]) -> Float {
        guard a.count == b.count else {
            Log.api.error("Embedding dimension mismatch: \(a.count) vs \(b.count)")
            return 0.0
        }

        var dotProduct: Float = 0.0
        var magnitudeA: Float = 0.0
        var magnitudeB: Float = 0.0

        for i in 0..<a.count {
            dotProduct += a[i] * b[i]
            magnitudeA += a[i] * a[i]
            magnitudeB += b[i] * b[i]
        }

        magnitudeA = sqrt(magnitudeA)
        magnitudeB = sqrt(magnitudeB)

        guard magnitudeA > 0 && magnitudeB > 0 else {
            return 0.0
        }

        return dotProduct / (magnitudeA * magnitudeB)
    }

    // MARK: - Database Operations

    /// Save message embedding to database
    func saveMessageEmbedding(messageId: UUID, embedding: [Float], model: String) throws {
        let data = serializeEmbedding(embedding)
        try DatabaseService.shared.saveMessageEmbedding(
            messageId: messageId,
            embedding: data,
            model: model,
            dimension: embedding.count
        )
    }

    /// Get message embedding from database
    func getMessageEmbedding(messageId: UUID) throws -> [Float]? {
        guard let data = try DatabaseService.shared.getMessageEmbedding(messageId: messageId) else {
            return nil
        }
        return deserializeEmbedding(data)
    }

    /// Save conversation embedding to database
    func saveConversationEmbedding(conversationId: UUID, embedding: [Float], model: String) throws {
        let data = serializeEmbedding(embedding)
        try DatabaseService.shared.saveConversationEmbedding(
            conversationId: conversationId,
            embedding: data,
            model: model,
            dimension: embedding.count
        )
    }

    /// Get conversation embedding from database
    func getConversationEmbedding(conversationId: UUID) throws -> [Float]? {
        guard let data = try DatabaseService.shared.getConversationEmbedding(conversationId: conversationId) else {
            return nil
        }
        return deserializeEmbedding(data)
    }

    // MARK: - Serialization

    /// Serialize embedding to binary data (4 bytes per float, little-endian)
    private func serializeEmbedding(_ embedding: [Float]) -> Data {
        var data = Data(capacity: embedding.count * 4)
        for value in embedding {
            var littleEndian = value.bitPattern.littleEndian
            withUnsafeBytes(of: &littleEndian) { bytes in
                data.append(contentsOf: bytes)
            }
        }
        return data
    }

    /// Deserialize embedding from binary data
    private func deserializeEmbedding(_ data: Data) -> [Float] {
        var embedding: [Float] = []
        embedding.reserveCapacity(data.count / 4)

        for offset in stride(from: 0, to: data.count, by: 4) {
            let bytes = data.subdata(in: offset..<(offset + 4))
            let bitPattern = bytes.withUnsafeBytes { $0.load(as: UInt32.self) }
            let value = Float(bitPattern: UInt32(littleEndian: bitPattern))
            embedding.append(value)
        }

        return embedding
    }

    // MARK: - Conversation Embedding Generation

    /// Generate embedding for an entire conversation (aggregate of messages)
    func generateConversationEmbedding(conversationId: UUID) async throws {
        // Use user's selected provider, or fall back to best available
        guard let provider = getSelectedProvider() else {
            throw EmbeddingError.noProvidersAvailable
        }

        // Load conversation messages
        guard let (_, messages) = try? DatabaseService.shared.loadConversation(id: conversationId) else {
            throw EmbeddingError.conversationNotFound
        }

        // Combine all message content
        let chatMessages = messages.filter { $0.role == .user || $0.role == .assistant }
        let combinedText = chatMessages.map { $0.content }.joined(separator: "\n\n")

        // Truncate if too long (8191 tokens max for most embedding models)
        let truncated = String(combinedText.prefix(30000)) // ~7500 tokens

        // Generate embedding
        let embedding = try await generateEmbedding(text: truncated, provider: provider)

        // Save to database
        try saveConversationEmbedding(conversationId: conversationId, embedding: embedding, model: provider.defaultModel)

        Log.api.info("Generated conversation embedding for \(conversationId) using \(provider.displayName) (\(embedding.count) dimensions)")
    }
}

// MARK: - Errors

enum EmbeddingError: LocalizedError {
    case missingAPIKey(String)
    case invalidResponse
    case apiError(Int, String)
    case providerNotImplemented(String)
    case conversationNotFound
    case noProvidersAvailable

    var errorDescription: String? {
        switch self {
        case .missingAPIKey(let provider):
            return "\(provider) API key not configured"
        case .invalidResponse:
            return "Invalid response from embedding API"
        case .apiError(let code, let message):
            return "Embedding API error (\(code)): \(message)"
        case .providerNotImplemented(let message):
            return message
        case .conversationNotFound:
            return "Conversation not found"
        case .noProvidersAvailable:
            return "No embedding providers available. Please configure an API key for OpenAI, OpenRouter, or Google."
        }
    }
}