feat: macOS architecture, cloud ASR/LLM providers, and 6-step iOS onboarding

- Add macOS menu-bar dictation app with local ASR models (SenseVoice/Qwen3), global Option hotkey, and bottom overlay - Add cloud ASR/LLM providers (Anthropic, Volcengine, Bailian, and more) with provider logos, model listing, and connection checks - Add shared 7-day usage stats UI (UsageStatsCluster / SevenDayUsageChart) - Add iOS onboarding step 6 for polish LLM setup; hide custom-language-model diagnostic toggle behind DEBUG - Unify iOS onboarding tagline with the macOS brand line ("开口即文字。") - Rewrite README (Chinese-first, product-oriented) and refresh GitHub Pages
2026-07-11 19:10:20 +08:00
parent cdf833935a
commit cc8dd1070a
116 changed files with 6659 additions and 2634 deletions
@@ -8,6 +8,11 @@
 import Foundation

 public struct AppGroupConfiguration: Sendable, Equatable {
+    /// Default polish LLM for fresh installs (local + cloud pickers).
+    public static let defaultPolishProviderId = "deepseek"
+    /// Default cloud ASR provider for fresh installs (independent from polish).
+    public static let defaultCloudASRProviderId = "volcengine"
+
    // MARK: - Keys

    public enum Keys {
@@ -31,6 +36,7 @@ public struct AppGroupConfiguration: Sendable, Equatable {
        public static let handednessPreference = "config.handednessPreference"
        public static let cursorDragNavigationEnabled = "config.cursorDragNavigationEnabled"
        public static let polishIntensity = "config.polishIntensity"
+        public static let llmThinkingEnabled = "config.llmThinkingEnabled"
        public static let detectedAppContext = "config.detectedAppContext"
        public static let detectedAppContextAt = "config.detectedAppContextAt"
        public static let personalDictionary = "config.personalDictionary.v1"
@@ -70,6 +76,8 @@ public struct AppGroupConfiguration: Sendable, Equatable {
    public var handednessPreference: HandednessPreference
    public var cursorDragNavigationEnabled: Bool
    public var polishIntensity: PolishIntensity
+    /// Enables provider-specific reasoning / thinking controls for polish LLM requests.
+    public var llmThinkingEnabled: Bool
    public var personalDictionary: PersonalDictionary
    /// Opt-in iCloud KVS sync for the personal dictionary (main app only).
    public var personalDictionaryICloudSyncEnabled: Bool
@@ -150,7 +158,9 @@ public struct AppGroupConfiguration: Sendable, Equatable {
        OpenAICompatibleClient(
            baseURL: baseURL,
            apiKey: apiKey,
-            model: model
+            model: model,
+            providerId: providerId,
+            thinkingEnabled: llmThinkingEnabled
        )
    }

@@ -193,8 +203,9 @@ public struct AppGroupConfiguration: Sendable, Equatable {

    /// Loads configuration from a known-available UserDefaults suite.
    public static func load(fromAvailable defaults: UserDefaults) -> AppGroupConfiguration {
+        let storedProviderId = defaults.string(forKey: Keys.providerId)
        var config = AppGroupConfiguration(
-            providerId: defaults.string(forKey: Keys.providerId) ?? "openai",
+            providerId: storedProviderId ?? defaultPolishProviderId,
            baseURL: "",
            model: "",
            asrProviderId: defaults.string(forKey: Keys.asrProviderId) ?? "",
@@ -228,6 +239,7 @@ public struct AppGroupConfiguration: Sendable, Equatable {
                return defaults.bool(forKey: Keys.cursorDragNavigationEnabled)
            }(),
            polishIntensity: resolvePolishIntensity(from: defaults),
+            llmThinkingEnabled: defaults.bool(forKey: Keys.llmThinkingEnabled),
            personalDictionary: decodePersonalDictionary(from: defaults),
            personalDictionaryICloudSyncEnabled: {
                if defaults.object(forKey: Keys.personalDictionaryICloudSyncEnabled) == nil {
@@ -267,7 +279,8 @@ public struct AppGroupConfiguration: Sendable, Equatable {
        }

        if config.asrProviderId.isEmpty {
-            config.asrProviderId = config.providerId
+            // Pre-split installs only stored `providerId`; copy it so ASR keeps working.
+            config.asrProviderId = storedProviderId ?? defaultCloudASRProviderId
            defaults.set(config.asrProviderId, forKey: Keys.asrProviderId)
        }
        let asrPreset = LLMProvider.provider(id: config.asrProviderId)
@@ -279,6 +292,17 @@ public struct AppGroupConfiguration: Sendable, Equatable {
                ?? CloudASRModelCatalog.defaultModel(for: config.asrProviderId)
        }

+        // Legacy qwen cloud ASR → bailian realtime (HTTP Flash path removed).
+        if config.asrProviderId == "qwen" {
+            let bailian = LLMProvider.provider(id: "bailian")
+            config.asrProviderId = "bailian"
+            config.asrBaseURL = bailian.defaultBaseURL
+            config.asrModel = CloudASRModelCatalog.alibabaFunASRRealtime
+            defaults.set(config.asrProviderId, forKey: Keys.asrProviderId)
+            defaults.set(config.asrBaseURL, forKey: Keys.asrBaseURL)
+            defaults.set(config.asrModel, forKey: Keys.asrModel)
+        }
+
        // One-shot legacy migration: plaintext apiKey in UserDefaults → Keychain.
        _ = resolveAPIKey(
            defaults: defaults,
@@ -311,26 +335,6 @@ public struct AppGroupConfiguration: Sendable, Equatable {
            config.modeId = "polish"
            defaults.set("polish", forKey: Keys.modeId)
        }
-        // DeepSeek is local-engine only — never a cloud picker choice.
-        if config.engineMode == "cloud", config.providerId == "deepseek" {
-            let openAI = LLMProvider.provider(id: "openai")
-            config.providerId = openAI.id
-            config.baseURL = openAI.defaultBaseURL
-            config.model = openAI.defaultModel
-            defaults.set(openAI.id, forKey: Keys.providerId)
-            defaults.set(openAI.defaultBaseURL, forKey: Keys.baseURL)
-            defaults.set(openAI.defaultModel, forKey: Keys.model)
-        }
-        if config.engineMode == "cloud", config.asrProviderId == "deepseek" {
-            let openAI = LLMProvider.provider(id: "openai")
-            config.asrProviderId = openAI.id
-            config.asrBaseURL = openAI.defaultBaseURL
-            config.asrModel = CloudASRModelCatalog.defaultModel(for: openAI.id)
-            defaults.set(openAI.id, forKey: Keys.asrProviderId)
-            defaults.set(openAI.defaultBaseURL, forKey: Keys.asrBaseURL)
-            defaults.set(openAI.defaultModel, forKey: Keys.asrModel)
-        }
-
        return config
    }

@@ -352,6 +356,7 @@ public struct AppGroupConfiguration: Sendable, Equatable {
        defaults.set(handednessPreference.rawValue, forKey: Keys.handednessPreference)
        defaults.set(cursorDragNavigationEnabled, forKey: Keys.cursorDragNavigationEnabled)
        defaults.set(polishIntensity.rawValue, forKey: Keys.polishIntensity)
+        defaults.set(llmThinkingEnabled, forKey: Keys.llmThinkingEnabled)
        defaults.set(flowSkipAppSwitch, forKey: Keys.flowSkipAppSwitch)
        defaults.set(flowInactivityDuration.rawValue, forKey: Keys.flowInactivityDuration)
        defaults.set(localASRCustomLanguageModelEnabled, forKey: Keys.localASRCustomLanguageModelEnabled)
@@ -10,10 +10,14 @@ import Foundation
 public enum CloudASRStrategy: String, Sendable, Equatable {
    /// 智谱 GLM-ASR — `hotwords` + optional `prompt`.
    case zhipuHotwords
-    /// 阿里百炼 Fun-ASR — managed `vocabulary_id` + context text.
-    case alibabaVocabulary
-    /// OpenAI / 小米 MiMo / 自定义端点 — `prompt` on transcription APIs.
+    /// 百炼 Fun-ASR Realtime — DashScope 经典 inference WebSocket 流式。
+    case bailianStreaming
+    /// OpenAI / Groq / 硅基流动 / Whisper 等 — `prompt` on transcription APIs.
    case prompt
+    /// OpenRouter `/audio/transcriptions` — JSON body + base64 WAV (not multipart).
+    case openRouterJson
+    /// 火山引擎 SAUC 大模型流式 ASR（WebSocket + binary frame）。
+    case volcengineStreaming
    /// Moonshot 托管 API 暂无音频转写；云端引擎回退端侧 ASR。
    case localFallback
 }
@@ -54,8 +58,23 @@ public enum CloudASRError: Error, LocalizedError, Sendable, Equatable {
 }

 public enum CloudASRModelCatalog {
+    /// Provider ids shown in the cloud ASR picker (explicit allowlist).
+    public static let selectableProviderIds: Set<String> = [
+        "openai",
+        "whisper",
+        "bailian",
+        "zhipu",
+        "groq",
+        "siliconflow",
+        "openrouter",
+        "mimo",
+        "volcengine",
+        "custom",
+    ]
+
    /// Sync Fun-ASR Flash — base64 upload, ≤ 5 min, supports context + vocabulary.
    public static let alibabaFunASRFlash = "fun-asr-flash-2026-06-15"
+    public static let alibabaFunASRRealtime = "fun-asr-realtime"
    /// Must match the ASR model used at recognition time.
    public static let alibabaVocabularyTargetModel = alibabaFunASRFlash

@@ -63,24 +82,38 @@ public enum CloudASRModelCatalog {
    public static let openAITranscribe = "gpt-4o-mini-transcribe"
    public static let openAIWhisper = "whisper-1"
    public static let mimoASR = "mimo-v2.5-asr"
+    public static let groqWhisper = "whisper-large-v3-turbo"
+    public static let siliconflowASR = "FunAudioLLM/SenseVoiceSmall"
+    public static let openrouterWhisper = "openai/whisper-large-v3-turbo"
+    public static let volcengineDefaultResourceID = "volc.seedasr.sauc.duration"
+    public static let volcengineEndpoint = "wss://openspeech.bytedance.com/api/v3/sauc/bigmodel_async"
+    public static let bailianDefaultEndpoint = "wss://dashscope.aliyuncs.com/api-ws/v1/inference/"

    public static let alibabaAPIBase = "https://dashscope.aliyuncs.com/api/v1"
    public static let alibabaCustomizationPath = "/services/audio/asr/customization"
    public static let alibabaMultimodalPath = "/services/aigc/multimodal-generation/generation"
    public static let zhipuTranscriptionPath = "/audio/transcriptions"

+    public static func supportsCloudASRSelection(providerId: String) -> Bool {
+        selectableProviderIds.contains(providerId)
+    }
+
    public static func strategy(for providerId: String) -> CloudASRStrategy {
        switch providerId {
        case "zhipu":
            return .zhipuHotwords
-        case "qwen":
-            return .alibabaVocabulary
+        case "bailian":
+            return .bailianStreaming
        case "moonshot":
            return .localFallback
-        case "openai", "mimo", "custom":
+        case "volcengine":
+            return .volcengineStreaming
+        case "openrouter":
+            return .openRouterJson
+        case "openai", "whisper", "mimo", "groq", "siliconflow", "custom":
            return .prompt
        default:
-            return .prompt
+            return .localFallback
        }
    }

@@ -88,16 +121,36 @@ public enum CloudASRModelCatalog {
        switch providerId {
        case "zhipu":
            return zhipuGLMASR
-        case "qwen":
-            return alibabaFunASRFlash
+        case "bailian":
+            return alibabaFunASRRealtime
+        case "whisper":
+            return openAIWhisper
        case "mimo":
            return mimoASR
+        case "groq":
+            return groqWhisper
+        case "siliconflow":
+            return siliconflowASR
+        case "openrouter":
+            return openrouterWhisper
+        case "volcengine":
+            return volcengineDefaultResourceID
        case "openai", "custom":
            return openAITranscribe
        default:
            return openAITranscribe
        }
    }
+
+    /// Whether the ASR settings card should expose a custom endpoint field.
+    public static func showsASREndpointField(for providerId: String) -> Bool {
+        switch strategy(for: providerId) {
+        case .prompt, .openRouterJson, .bailianStreaming:
+            return true
+        case .zhipuHotwords, .volcengineStreaming, .localFallback:
+            return false
+        }
+    }
 }

 extension LLMProvider {
@@ -112,9 +165,9 @@ extension LLMProvider {
    /// Official hotwords / vocabulary APIs during cloud ASR (not prompt-only bias).
    public var supportsPersonalDictionaryCloudASR: Bool {
        switch cloudASRStrategy {
-        case .zhipuHotwords, .alibabaVocabulary:
+        case .zhipuHotwords:
            return true
-        case .prompt, .localFallback:
+        case .bailianStreaming, .prompt, .openRouterJson, .volcengineStreaming, .localFallback:
            return false
        }
    }
@@ -0,0 +1,137 @@
+// FlowHandoffPolicy.swift
+// OSGKeyboard · Shared
+//
+// Pure decision helpers for keyboard → host handoff. Keeps "session still
+// alive, ready contract briefly missing" from being treated as a cold start.
+
+import Foundation
+
+/// Action the keyboard should take when the user presses the mic.
+public enum FlowMicPressAction: Equatable, Sendable {
+    case startRecording
+    /// Session is alive (or was very recently); poll for ready, then optionally record.
+    case waitForHostReady(recordWhenReady: Bool)
+    /// Host process is gone / no session — open `osgkeyboard://startflow`.
+    case openHostColdStart
+    case ignore
+}
+
+/// Whether the host app should show the cold-start overlay for a `startflow`.
+public enum FlowColdStartOverlayDecision: Equatable, Sendable {
+    /// Do not set handoff flags or show preparing/ready UI.
+    case silence
+    /// Show preparing and run the cold-start / recovery path.
+    case present
+}
+
+public enum FlowHandoffPolicy {
+    /// Proactive keyboard auto-launch of the host is intentionally disabled.
+    /// Opening the host must be driven by an explicit mic press (or Live Activity).
+    public static let allowsProactiveHostAutoLaunch = false
+
+    /// Samples of "host truly dead" required before a cold-start jump is allowed
+    /// from a non-press path. Mic press uses `shouldOpenHostColdStart` directly.
+    public static let coldStartDeadSampleThreshold = 2
+
+    /// True when the session contract still implies a living (or recoverable)
+    /// host — so a transient `ready=false` must wait, not jump.
+    public static func shouldTreatHostAsAlive(
+        sessionActive: Bool,
+        hostReachable: Bool,
+        hostStale: Bool,
+        withinReadyGrace: Bool
+    ) -> Bool {
+        if hostStale { return false }
+        guard sessionActive else { return false }
+        // Reachable heartbeat, or a recent ready sample, means the process is
+        // still ours — finalize races often look like hostNotReady for one frame.
+        if hostReachable || withinReadyGrace { return true }
+        // Session flag still valid and not past the zombie window: prefer wait.
+        return true
+    }
+
+    /// Whether `osgkeyboard://startflow` is justified for the current host state.
+    public static func shouldOpenHostColdStart(
+        sessionActive: Bool,
+        hostReachable: Bool,
+        hostStale: Bool,
+        withinReadyGrace: Bool
+    ) -> Bool {
+        !shouldTreatHostAsAlive(
+            sessionActive: sessionActive,
+            hostReachable: hostReachable,
+            hostStale: hostStale,
+            withinReadyGrace: withinReadyGrace
+        )
+    }
+
+    /// Mic-press routing shared by the keyboard coordinator and unit tests.
+    public static func micPressAction(
+        availability: MicVoiceAvailability,
+        sessionActive: Bool,
+        hostReachable: Bool,
+        hostStale: Bool,
+        withinReadyGrace: Bool
+    ) -> FlowMicPressAction {
+        switch availability {
+        case .ready:
+            return .startRecording
+        case .recording, .processing:
+            return .ignore
+        case .unavailable(.missingAPIKey),
+             .unavailable(.noFullAccess),
+             .unavailable(.appGroupUnavailable):
+            // Caller surfaces the specific error UI.
+            return .ignore
+        case .unavailable(.preparingSession):
+            // Session is warming — never cold-start; wait then record.
+            return .waitForHostReady(recordWhenReady: true)
+        case .unavailable(.hostNotReady):
+            if shouldTreatHostAsAlive(
+                sessionActive: sessionActive,
+                hostReachable: hostReachable,
+                hostStale: hostStale,
+                withinReadyGrace: withinReadyGrace
+            ) {
+                return .waitForHostReady(recordWhenReady: true)
+            }
+            return .openHostColdStart
+        }
+    }
+
+    /// Host-app gate: a `startflow` against an already-healthy (or busy) session
+    /// must not flash "Voice is ready".
+    public static func coldStartOverlayDecision(
+        sessionIsActive: Bool,
+        hostIsReady: Bool,
+        isUtteranceBusy: Bool
+    ) -> FlowColdStartOverlayDecision {
+        guard sessionIsActive else { return .present }
+        if hostIsReady || isUtteranceBusy { return .silence }
+        // Active but not ready and not busy — engine recovery may need UI.
+        return .present
+    }
+}
+
+/// Counts consecutive "host truly dead" observations to ignore single-frame races.
+public struct FlowColdStartDebouncer: Equatable, Sendable {
+    public private(set) var consecutiveDeadSamples: Int = 0
+
+    public init(consecutiveDeadSamples: Int = 0) {
+        self.consecutiveDeadSamples = consecutiveDeadSamples
+    }
+
+    /// Returns true once enough consecutive dead samples have been seen.
+    public mutating func observe(hostTrulyDead: Bool) -> Bool {
+        if hostTrulyDead {
+            consecutiveDeadSamples += 1
+        } else {
+            consecutiveDeadSamples = 0
+        }
+        return consecutiveDeadSamples >= FlowHandoffPolicy.coldStartDeadSampleThreshold
+    }
+
+    public mutating func reset() {
+        consecutiveDeadSamples = 0
+    }
+}
@@ -2,7 +2,7 @@
 // OSGKeyboard · Shared
 //
 // Which hand the user holds the phone with — controls bottom-row key order
-// on the keyboard (delete ↔ return swap for right-handed use).
+// on the keyboard (delete ↔ space swap for right-handed use).

 import Foundation

@@ -19,7 +19,7 @@ public enum HandednessPreference: String, CaseIterable, Identifiable, Sendable,
        }
    }

-    /// Right-handed preference places return on the left and delete on the right.
+    /// Right-handed preference places space on the left and delete on the right.
    public var swapsActionKeys: Bool { self == .right }

    public static func fromStored(_ raw: String?) -> HandednessPreference {
@@ -48,15 +48,21 @@ public struct LLMProvider: Identifiable, Codable, Hashable, Sendable {
            apiKeyURL: URL(string: "https://platform.openai.com/api-keys"),
            blurb: "GPT-4o mini · 多语言 · Multilingual"
        ),
+        .init(
+            id: "ark",
+            name: "火山方舟 Ark",
+            defaultBaseURL: "https://ark.cn-beijing.volces.com/api/v3",
+            defaultModel: "deepseek-v3-2-251201",
+            apiKeyURL: URL(string: "https://console.volcengine.com/ark/region:ark+cn-beijing/apiKey"),
+            blurb: "豆包 / DeepSeek · OpenAI 兼容 · OpenAI-compatible"
+        ),
        .init(
            id: "deepseek",
            name: "DeepSeek",
            defaultBaseURL: "https://api.deepseek.com/v1",
            defaultModel: "deepseek-v4-flash",
            apiKeyURL: URL(string: "https://platform.deepseek.com/api_keys"),
-            blurb: "deepseek-v4-flash · 本地引擎内置 · Local engine built-in",
-            // Local engine only — never shown in cloud-engine pickers.
-            isUserSelectable: false
+            blurb: "deepseek-v4-flash · 本地引擎可内置 · Local engine optional built-in"
        ),
        .init(
            id: "qwen",
@@ -82,6 +88,30 @@ public struct LLMProvider: Identifiable, Codable, Hashable, Sendable {
            apiKeyURL: URL(string: "https://platform.moonshot.cn/console/api-keys"),
            blurb: "Kimi · 长上下文 · Long context"
        ),
+        .init(
+            id: "siliconflow",
+            name: "硅基流动 SiliconFlow",
+            defaultBaseURL: "https://api.siliconflow.cn/v1",
+            defaultModel: "Qwen/Qwen2.5-7B-Instruct",
+            apiKeyURL: URL(string: "https://cloud.siliconflow.cn/account/ak"),
+            blurb: "多模型聚合 · OpenAI 兼容 · OpenAI-compatible"
+        ),
+        .init(
+            id: "groq",
+            name: "Groq",
+            defaultBaseURL: "https://api.groq.com/openai/v1",
+            defaultModel: "llama-3.3-70b-versatile",
+            apiKeyURL: URL(string: "https://console.groq.com/keys"),
+            blurb: "超低延迟 LPU · Ultra-low latency"
+        ),
+        .init(
+            id: "minimax",
+            name: "MiniMax",
+            defaultBaseURL: "https://api.minimaxi.com/v1",
+            defaultModel: "MiniMax-M2.5",
+            apiKeyURL: URL(string: "https://platform.minimaxi.com/user-center/basic-information"),
+            blurb: "MiniMax-M2.5 · 中文优化 · Chinese-optimized"
+        ),
        .init(
            id: "mimo",
            name: "小米 MiMo",
@@ -90,6 +120,106 @@ public struct LLMProvider: Identifiable, Codable, Hashable, Sendable {
            apiKeyURL: URL(string: "https://platform.xiaomimimo.com"),
            blurb: "mimo-v2.5 · 中文优化 · Chinese-optimized"
        ),
+        .init(
+            id: "openrouter",
+            name: "OpenRouter",
+            defaultBaseURL: "https://openrouter.ai/api/v1",
+            defaultModel: "qwen/qwen3-coder:free",
+            apiKeyURL: URL(string: "https://openrouter.ai/keys"),
+            blurb: "多模型路由 · Model routing · OpenAI-compatible"
+        ),
+        .init(
+            id: "gemini",
+            name: "Google Gemini",
+            defaultBaseURL: "https://generativelanguage.googleapis.com/v1beta/openai",
+            defaultModel: "gemini-2.5-flash",
+            apiKeyURL: URL(string: "https://aistudio.google.com/apikey"),
+            blurb: "Gemini 2.5 Flash · OpenAI 兼容端点"
+        ),
+        .init(
+            id: "anthropic",
+            name: "Anthropic Claude",
+            defaultBaseURL: "https://api.anthropic.com/v1",
+            defaultModel: "claude-sonnet-4-6",
+            apiKeyURL: URL(string: "https://console.anthropic.com/settings/keys"),
+            blurb: "Claude Sonnet · Messages API"
+        ),
+        .init(
+            id: "xai",
+            name: "xAI Grok",
+            defaultBaseURL: "https://api.x.ai/v1",
+            defaultModel: "grok-3-mini",
+            apiKeyURL: URL(string: "https://console.x.ai"),
+            blurb: "Grok · OpenAI 兼容 · OpenAI-compatible"
+        ),
+        .init(
+            id: "mistral",
+            name: "Mistral AI",
+            defaultBaseURL: "https://api.mistral.ai/v1",
+            defaultModel: "mistral-small-latest",
+            apiKeyURL: URL(string: "https://console.mistral.ai/api-keys"),
+            blurb: "Mistral Small · 欧洲托管 · EU-hosted"
+        ),
+        .init(
+            id: "cometapi",
+            name: "CometAPI",
+            defaultBaseURL: "https://api.cometapi.com/v1",
+            defaultModel: "gpt-4o",
+            apiKeyURL: URL(string: "https://api.cometapi.com"),
+            blurb: "多模型聚合 · OpenAI 兼容"
+        ),
+        .init(
+            id: "alibabaCoding",
+            name: "阿里 Coding Plan",
+            defaultBaseURL: "https://coding-intl.dashscope.aliyuncs.com/v1",
+            defaultModel: "qwen3-coder-plus",
+            apiKeyURL: URL(string: "https://dashscope.console.aliyun.com/apiKey"),
+            blurb: "通义 Coder · 代码润色 · Coding polish"
+        ),
+        .init(
+            id: "codingPlanX",
+            name: "CodingPlanX",
+            defaultBaseURL: "https://api.codingplanx.ai/v1",
+            defaultModel: "gpt-5-mini",
+            apiKeyURL: URL(string: "https://codingplanx.ai"),
+            blurb: "CodingPlanX · OpenAI 兼容"
+        ),
+        // MARK: - ASR-only presets (hidden from polish picker)
+        .init(
+            id: "volcengine",
+            name: "火山引擎 Volcengine",
+            defaultBaseURL: "wss://openspeech.bytedance.com/api/v3/sauc/bigmodel_async",
+            defaultModel: "volc.seedasr.sauc.duration",
+            apiKeyURL: URL(string: "https://console.volcengine.com/speech"),
+            blurb: "流式大模型 ASR · API Key 填 appId:accessToken[:resourceId]",
+            isUserSelectable: false
+        ),
+        .init(
+            id: "bailian",
+            name: "百炼实时 ASR",
+            defaultBaseURL: "wss://dashscope.aliyuncs.com/api-ws/v1/inference/",
+            defaultModel: "fun-asr-realtime",
+            apiKeyURL: URL(string: "https://dashscope.console.aliyun.com/apiKey"),
+            blurb: "Fun-ASR Realtime · 百炼词表",
+            isUserSelectable: false
+        ),
+        .init(
+            id: "whisper",
+            name: "Whisper (OpenAI)",
+            defaultBaseURL: "https://api.openai.com/v1",
+            defaultModel: "whisper-1",
+            apiKeyURL: URL(string: "https://platform.openai.com/api-keys"),
+            blurb: "whisper-1 · 经典 Whisper 端点",
+            isUserSelectable: false
+        ),
+        .init(
+            id: "codex_oauth",
+            name: "Codex OAuth",
+            defaultBaseURL: "",
+            defaultModel: "gpt-5.3-codex-spark",
+            blurb: "ChatGPT Codex OAuth · 暂不支持",
+            isUserSelectable: false
+        ),
        .init(
            id: "custom",
            name: "Custom · 自定义",
@@ -103,16 +233,13 @@ public struct LLMProvider: Identifiable, Codable, Hashable, Sendable {
        presets.first(where: { $0.id == id }) ?? .presets[0]
    }

-    /// Presets the user may pick in Settings / onboarding. DeepSeek is
-    /// excluded — it is wired exclusively to the local engine.
+    /// Presets the user may pick in Settings / onboarding.
    public static var userSelectablePresets: [LLMProvider] {
        presets.filter(\.isUserSelectable)
    }

-    /// Cloud ASR presets (excludes providers without a cloud transcription API).
+    /// Cloud ASR presets (explicit allowlist — polish-only providers excluded).
    public static var asrSelectablePresets: [LLMProvider] {
-        userSelectablePresets.filter {
-            CloudASRModelCatalog.strategy(for: $0.id) != .localFallback
-        }
+        presets.filter { CloudASRModelCatalog.supportsCloudASRSelection(providerId: $0.id) }
    }
 }
@@ -111,7 +111,6 @@ public final class ProviderConfig: ObservableObject, @unchecked Sendable {
        didSet {
            guard !isApplyingConfiguration, engineMode != configuration.engineMode else { return }
            configuration.engineMode = engineMode
-            applyEngineModeSideEffects()
            persistConfiguration(postConfigChanged: true)
        }
    }
@@ -179,7 +178,7 @@ public final class ProviderConfig: ObservableObject, @unchecked Sendable {
        }
    }
    /// Which hand the user holds the phone with — mirrors to the keyboard
-    /// extension so delete / return can swap on the bottom row.
+    /// extension so delete / space can swap on the bottom row.
    @Published public var handednessPreference: HandednessPreference {
        didSet {
            guard !isApplyingConfiguration,
@@ -218,6 +217,17 @@ public final class ProviderConfig: ObservableObject, @unchecked Sendable {
        }
    }

+    /// Enables provider-specific reasoning / thinking controls when the
+    /// selected polish LLM supports them.
+    @Published public var llmThinkingEnabled: Bool {
+        didSet {
+            guard !isApplyingConfiguration,
+                  llmThinkingEnabled != configuration.llmThinkingEnabled else { return }
+            configuration.llmThinkingEnabled = llmThinkingEnabled
+            persistConfiguration(postConfigChanged: true)
+        }
+    }
+
    /// When enabled, the host app tries to return to the source app after a cold-start handoff.
    @Published public var flowSkipAppSwitch: Bool {
        didSet {
@@ -335,6 +345,7 @@ public final class ProviderConfig: ObservableObject, @unchecked Sendable {
        handednessPreference = configuration.handednessPreference
        cursorDragNavigationEnabled = configuration.cursorDragNavigationEnabled
        polishIntensity = configuration.polishIntensity
+        llmThinkingEnabled = configuration.llmThinkingEnabled
        flowSkipAppSwitch = configuration.flowSkipAppSwitch
        flowInactivityDuration = configuration.flowInactivityDuration
        localASRCustomLanguageModelEnabled = configuration.localASRCustomLanguageModelEnabled
@@ -347,15 +358,34 @@ public final class ProviderConfig: ObservableObject, @unchecked Sendable {
        isApplyingConfiguration = false
    }

-    /// Keep cloud vs local provider choices isolated when the user
-    /// switches engines in Settings / onboarding.
-    private func applyEngineModeSideEffects() {
-        if engineMode == "cloud", providerId == "deepseek" {
-            apply(preset: LLMProvider.provider(id: "openai"))
-        }
-        if engineMode == "cloud", asrProviderId == "deepseek" {
-            applyAsr(preset: LLMProvider.provider(id: "openai"))
-        }
+    public func reset() {
+        isApplyingConfiguration = true
+        let polishPreset = LLMProvider.provider(id: AppGroupConfiguration.defaultPolishProviderId)
+        let asrPreset = LLMProvider.provider(id: AppGroupConfiguration.defaultCloudASRProviderId)
+        providerId = polishPreset.id
+        baseURL = polishPreset.defaultBaseURL
+        apiKey = ""
+        model = polishPreset.defaultModel
+        asrProviderId = asrPreset.id
+        asrBaseURL = asrPreset.defaultBaseURL
+        asrModel = CloudASRModelCatalog.defaultModel(for: asrPreset.id)
+        asrApiKey = ""
+        handednessPreference = .left
+        localASRCustomLanguageModelEnabled = true
+        llmThinkingEnabled = false
+        hasAcknowledgedCloudSharing = false
+        configuration.providerId = polishPreset.id
+        configuration.baseURL = polishPreset.defaultBaseURL
+        configuration.model = polishPreset.defaultModel
+        configuration.asrProviderId = asrPreset.id
+        configuration.asrBaseURL = asrPreset.defaultBaseURL
+        configuration.asrModel = CloudASRModelCatalog.defaultModel(for: asrPreset.id)
+        configuration.handednessPreference = .left
+        configuration.localASRCustomLanguageModelEnabled = true
+        configuration.llmThinkingEnabled = false
+        configuration.hasAcknowledgedCloudSharing = false
+        isApplyingConfiguration = false
+        persistConfiguration()
    }

    private func persistConfiguration(postConfigChanged: Bool = false) {
@@ -401,6 +431,7 @@ public final class ProviderConfig: ObservableObject, @unchecked Sendable {
        handednessPreference = fresh.handednessPreference
        cursorDragNavigationEnabled = fresh.cursorDragNavigationEnabled
        polishIntensity = fresh.polishIntensity
+        llmThinkingEnabled = fresh.llmThinkingEnabled
        flowSkipAppSwitch = fresh.flowSkipAppSwitch
        flowInactivityDuration = fresh.flowInactivityDuration
        localASRCustomLanguageModelEnabled = fresh.localASRCustomLanguageModelEnabled
@@ -455,31 +486,4 @@ public final class ProviderConfig: ObservableObject, @unchecked Sendable {
        isApplyingConfiguration = false
        persistConfiguration()
    }
-
-    public func reset() {
-        isApplyingConfiguration = true
-        let preset = LLMProvider.provider(id: "openai")
-        providerId = preset.id
-        baseURL = preset.defaultBaseURL
-        apiKey = ""
-        model = preset.defaultModel
-        asrProviderId = preset.id
-        asrBaseURL = preset.defaultBaseURL
-        asrModel = CloudASRModelCatalog.defaultModel(for: preset.id)
-        asrApiKey = ""
-        handednessPreference = .left
-        localASRCustomLanguageModelEnabled = true
-        hasAcknowledgedCloudSharing = false
-        configuration.providerId = preset.id
-        configuration.baseURL = preset.defaultBaseURL
-        configuration.model = preset.defaultModel
-        configuration.asrProviderId = preset.id
-        configuration.asrBaseURL = preset.defaultBaseURL
-        configuration.asrModel = CloudASRModelCatalog.defaultModel(for: preset.id)
-        configuration.handednessPreference = .left
-        configuration.localASRCustomLanguageModelEnabled = true
-        configuration.hasAcknowledgedCloudSharing = false
-        isApplyingConfiguration = false
-        persistConfiguration()
-    }
 }
@@ -10,12 +10,24 @@ public enum ProviderLogo {
    /// Asset name for the provider's logo, or `nil` when there is no bundled logo.
    public static func assetName(for providerId: String) -> String? {
        switch providerId {
-        case "openai": return "openai"
+        case "openai", "whisper": return "openai"
        case "deepseek": return "deepseek"
-        case "qwen": return "qwen"
+        case "qwen", "bailian", "alibabaCoding": return "qwen"
        case "moonshot": return "moonshot"
        case "zhipu": return "zhipu"
        case "mimo": return "mimo"
+        case "ark", "volcengine": return "ark"
+        case "siliconflow": return "siliconflow"
+        case "groq": return "groq"
+        case "minimax": return "minimax"
+        case "openrouter": return "openrouter"
+        case "gemini": return "gemini"
+        case "anthropic": return "anthropic"
+        case "xai": return "xai"
+        case "mistral": return "mistral"
+        case "cometapi": return "cometapi"
+        case "codingPlanX": return "codingplanx"
+        case "codex_oauth": return "openai"
        case "apple": return "apple"
        case "custom": return "custom"
        default: return nil
@@ -27,6 +27,7 @@ public struct SyncedAppSettingsV2: Codable, Equatable, Sendable {
    public var handednessPreference: SyncedField<HandednessPreference>
    public var cursorDragNavigationEnabled: SyncedField<Bool>
    public var polishIntensity: SyncedField<PolishIntensity>
+    public var llmThinkingEnabled: SyncedField<Bool>
    public var flowSkipAppSwitch: SyncedField<Bool>
    public var flowInactivityDuration: SyncedField<FlowInactivityDuration>

@@ -47,6 +48,7 @@ public struct SyncedAppSettingsV2: Codable, Equatable, Sendable {
        handednessPreference: SyncedField<HandednessPreference>,
        cursorDragNavigationEnabled: SyncedField<Bool>,
        polishIntensity: SyncedField<PolishIntensity>,
+        llmThinkingEnabled: SyncedField<Bool>,
        flowSkipAppSwitch: SyncedField<Bool>,
        flowInactivityDuration: SyncedField<FlowInactivityDuration>
    ) {
@@ -66,6 +68,7 @@ public struct SyncedAppSettingsV2: Codable, Equatable, Sendable {
        self.handednessPreference = handednessPreference
        self.cursorDragNavigationEnabled = cursorDragNavigationEnabled
        self.polishIntensity = polishIntensity
+        self.llmThinkingEnabled = llmThinkingEnabled
        self.flowSkipAppSwitch = flowSkipAppSwitch
        self.flowInactivityDuration = flowInactivityDuration
    }
@@ -87,6 +90,7 @@ public struct SyncedAppSettingsV2: Codable, Equatable, Sendable {
        case handednessPreference
        case cursorDragNavigationEnabled
        case polishIntensity
+        case llmThinkingEnabled
        case flowSkipAppSwitch
        case flowInactivityDuration
    }
@@ -115,6 +119,10 @@ public struct SyncedAppSettingsV2: Codable, Equatable, Sendable {
            forKey: .cursorDragNavigationEnabled
        )
        polishIntensity = try container.decode(SyncedField<PolishIntensity>.self, forKey: .polishIntensity)
+        llmThinkingEnabled = try container.decodeIfPresent(
+            SyncedField<Bool>.self,
+            forKey: .llmThinkingEnabled
+        ) ?? SyncedField(value: false, updatedAt: polishIntensity.updatedAt, deviceID: polishIntensity.deviceID)
        flowSkipAppSwitch = try container.decode(SyncedField<Bool>.self, forKey: .flowSkipAppSwitch)
        flowInactivityDuration = try container.decode(
            SyncedField<FlowInactivityDuration>.self,
@@ -161,6 +169,7 @@ public struct SyncedAppSettingsV2: Codable, Equatable, Sendable {
            handednessPreference.updatedAt,
            cursorDragNavigationEnabled.updatedAt,
            polishIntensity.updatedAt,
+            llmThinkingEnabled.updatedAt,
            flowSkipAppSwitch.updatedAt,
            flowInactivityDuration.updatedAt,
        ].max() ?? .distantPast
@@ -197,6 +206,7 @@ public extension SyncedAppSettingsV2 {
            handednessPreference: field(configuration.handednessPreference),
            cursorDragNavigationEnabled: field(configuration.cursorDragNavigationEnabled),
            polishIntensity: field(configuration.polishIntensity),
+            llmThinkingEnabled: field(configuration.llmThinkingEnabled),
            flowSkipAppSwitch: field(configuration.flowSkipAppSwitch),
            flowInactivityDuration: field(configuration.flowInactivityDuration)
        )
@@ -225,6 +235,7 @@ public extension SyncedAppSettingsV2 {
            handednessPreference: field(legacy.handednessPreference),
            cursorDragNavigationEnabled: field(legacy.cursorDragNavigationEnabled),
            polishIntensity: field(legacy.polishIntensity),
+            llmThinkingEnabled: field(false),
            flowSkipAppSwitch: field(legacy.flowSkipAppSwitch),
            flowInactivityDuration: field(legacy.flowInactivityDuration)
        )
@@ -256,6 +267,7 @@ public extension SyncedAppSettingsV2 {
                remote: remote.cursorDragNavigationEnabled
            ),
            polishIntensity: .merge(local: local.polishIntensity, remote: remote.polishIntensity),
+            llmThinkingEnabled: .merge(local: local.llmThinkingEnabled, remote: remote.llmThinkingEnabled),
            flowSkipAppSwitch: .merge(local: local.flowSkipAppSwitch, remote: remote.flowSkipAppSwitch),
            flowInactivityDuration: .merge(
                local: local.flowInactivityDuration,
@@ -280,6 +292,7 @@ public extension SyncedAppSettingsV2 {
        configuration.handednessPreference = handednessPreference.value
        configuration.cursorDragNavigationEnabled = cursorDragNavigationEnabled.value
        configuration.polishIntensity = polishIntensity.value
+        configuration.llmThinkingEnabled = llmThinkingEnabled.value
        configuration.flowSkipAppSwitch = flowSkipAppSwitch.value
        configuration.flowInactivityDuration = flowInactivityDuration.value
    }
@@ -306,6 +319,7 @@ public extension SyncedAppSettingsV2 {
        patch(&copy.handednessPreference, value: configuration.handednessPreference)
        patch(&copy.cursorDragNavigationEnabled, value: configuration.cursorDragNavigationEnabled)
        patch(&copy.polishIntensity, value: configuration.polishIntensity)
+        patch(&copy.llmThinkingEnabled, value: configuration.llmThinkingEnabled)
        patch(&copy.flowSkipAppSwitch, value: configuration.flowSkipAppSwitch)
        patch(&copy.flowInactivityDuration, value: configuration.flowInactivityDuration)
        return copy
@@ -335,6 +349,7 @@ public extension SyncedAppSettingsV2 {
        touch(&copy.handednessPreference, value: configuration.handednessPreference)
        touch(&copy.cursorDragNavigationEnabled, value: configuration.cursorDragNavigationEnabled)
        touch(&copy.polishIntensity, value: configuration.polishIntensity)
+        touch(&copy.llmThinkingEnabled, value: configuration.llmThinkingEnabled)
        touch(&copy.flowSkipAppSwitch, value: configuration.flowSkipAppSwitch)
        touch(&copy.flowInactivityDuration, value: configuration.flowInactivityDuration)
        return copy
@@ -5,30 +5,78 @@

 import Foundation

+/// Local-calendar day key (`yyyy-MM-dd`) for daily usage buckets. String keys
+/// sort lexicographically in chronological order, which keeps pruning and
+/// range queries index-free.
+public enum UsageStatisticsDayKey {
+    public static func key(for date: Date, calendar: Calendar = .current) -> String {
+        let c = calendar.dateComponents([.year, .month, .day], from: date)
+        return String(format: "%04d-%02d-%02d", c.year ?? 0, c.month ?? 0, c.day ?? 0)
+    }
+
+    /// Drops buckets older than `days` so the synced blob stays small even
+    /// after months of use (the chart only ever needs the last 7 days).
+    public static func prune(_ daily: inout [String: Int], keepingDays days: Int, now: Date = Date(), calendar: Calendar = .current) {
+        guard let cutoff = calendar.date(byAdding: .day, value: -days, to: now) else { return }
+        let cutoffKey = key(for: cutoff, calendar: calendar)
+        daily = daily.filter { $0.key >= cutoffKey }
+    }
+}
+
 public struct UsageStatisticsDeviceSlice: Codable, Equatable, Sendable {
    public var updatedAt: Date
    public var dictationDurationSeconds: TimeInterval
    public var dictationCharacterCount: Int
    public var translationCharacterCount: Int
+    /// Grow-only per-day dictation character counts, keyed by local `yyyy-MM-dd`.
+    /// Powers the home page's 7-day chart; merged per-key with `max` (each device
+    /// only ever grows its own days) and summed across devices when aggregated.
+    public var dailyDictationCharacters: [String: Int]

    public init(
        updatedAt: Date = Date(),
        dictationDurationSeconds: TimeInterval = 0,
        dictationCharacterCount: Int = 0,
-        translationCharacterCount: Int = 0
+        translationCharacterCount: Int = 0,
+        dailyDictationCharacters: [String: Int] = [:]
    ) {
        self.updatedAt = updatedAt
        self.dictationDurationSeconds = dictationDurationSeconds
        self.dictationCharacterCount = dictationCharacterCount
        self.translationCharacterCount = translationCharacterCount
+        self.dailyDictationCharacters = dailyDictationCharacters
+    }
+
+    private enum CodingKeys: String, CodingKey {
+        case updatedAt
+        case dictationDurationSeconds
+        case dictationCharacterCount
+        case translationCharacterCount
+        case dailyDictationCharacters
+    }
+
+    // Custom decode so slices written before the daily-buckets field still load
+    // (the missing key defaults to an empty map rather than failing the decode).
+    public init(from decoder: Decoder) throws {
+        let container = try decoder.container(keyedBy: CodingKeys.self)
+        updatedAt = try container.decode(Date.self, forKey: .updatedAt)
+        dictationDurationSeconds = try container.decode(TimeInterval.self, forKey: .dictationDurationSeconds)
+        dictationCharacterCount = try container.decode(Int.self, forKey: .dictationCharacterCount)
+        translationCharacterCount = try container.decode(Int.self, forKey: .translationCharacterCount)
+        dailyDictationCharacters = try container.decodeIfPresent([String: Int].self, forKey: .dailyDictationCharacters) ?? [:]
    }

    public static func merge(local: UsageStatisticsDeviceSlice, remote: UsageStatisticsDeviceSlice) -> UsageStatisticsDeviceSlice {
-        UsageStatisticsDeviceSlice(
+        var mergedDaily = local.dailyDictationCharacters
+        for (day, value) in remote.dailyDictationCharacters {
+            mergedDaily[day] = max(mergedDaily[day] ?? 0, value)
+        }
+        return UsageStatisticsDeviceSlice(
            updatedAt: max(local.updatedAt, remote.updatedAt),
            dictationDurationSeconds: max(local.dictationDurationSeconds, remote.dictationDurationSeconds),
            dictationCharacterCount: max(local.dictationCharacterCount, remote.dictationCharacterCount),
-            translationCharacterCount: max(local.translationCharacterCount, remote.translationCharacterCount)
+            translationCharacterCount: max(local.translationCharacterCount, remote.translationCharacterCount),
+            dailyDictationCharacters: mergedDaily
        )
    }

@@ -75,6 +123,17 @@ public struct SyncedUsageStatisticsV2: Codable, Equatable, Sendable {
        )
    }

+    /// Cross-device daily dictation characters (summed per `yyyy-MM-dd`).
+    public var aggregatedDailyDictationCharacters: [String: Int] {
+        var result: [String: Int] = [:]
+        for slice in devices.values {
+            for (day, value) in slice.dailyDictationCharacters {
+                result[day, default: 0] += value
+            }
+        }
+        return result
+    }
+
    public static func merge(local: SyncedUsageStatisticsV2, remote: SyncedUsageStatisticsV2) -> SyncedUsageStatisticsV2 {
        var mergedDevices = local.devices
        for (deviceID, remoteSlice) in remote.devices {
@@ -0,0 +1,73 @@
+// VolcengineASRFields.swift
+// OSGKeyboard · Shared
+//
+// Parse / encode Volcengine SAUC credentials stored in the ASR API key field.
+
+import Foundation
+
+public struct VolcengineASRFields: Sendable, Equatable {
+    public var appID: String
+    public var accessToken: String
+    public var resourceID: String
+
+    public init(
+        appID: String = "",
+        accessToken: String = "",
+        resourceID: String = CloudASRModelCatalog.defaultModel(for: "volcengine")
+    ) {
+        self.appID = appID
+        self.accessToken = accessToken
+        self.resourceID = resourceID
+    }
+
+    public var encodedAPIKey: String {
+        let object = [
+            "app_id": appID,
+            "access_token": accessToken,
+            "resource_id": resourceID,
+        ]
+        guard let data = try? JSONSerialization.data(withJSONObject: object),
+              let string = String(data: data, encoding: .utf8) else {
+            return [appID, accessToken, resourceID].joined(separator: ":")
+        }
+        return string
+    }
+
+    public static func parse(apiKey: String, resourceFallback: String) -> VolcengineASRFields {
+        let trimmed = apiKey.trimmingCharacters(in: .whitespacesAndNewlines)
+        var fields = VolcengineASRFields(
+            appID: "",
+            accessToken: "",
+            resourceID: resourceFallback.isEmpty
+                ? CloudASRModelCatalog.defaultModel(for: "volcengine")
+                : resourceFallback
+        )
+
+        if let data = trimmed.data(using: .utf8),
+           let json = try? JSONSerialization.jsonObject(with: data) as? [String: Any] {
+            fields.appID = string(json, keys: ["app_id", "appId", "appid"]) ?? ""
+            fields.accessToken = string(json, keys: ["access_token", "accessToken", "token"]) ?? ""
+            fields.resourceID = string(json, keys: ["resource_id", "resourceId", "resource"]) ?? fields.resourceID
+            return fields
+        }
+
+        let parts = trimmed
+            .components(separatedBy: CharacterSet(charactersIn: ":\n,"))
+            .map { $0.trimmingCharacters(in: .whitespacesAndNewlines) }
+            .filter { !$0.isEmpty }
+        if parts.indices.contains(0) { fields.appID = parts[0] }
+        if parts.indices.contains(1) { fields.accessToken = parts[1] }
+        if parts.indices.contains(2) { fields.resourceID = parts[2] }
+        return fields
+    }
+
+    private static func string(_ json: [String: Any], keys: [String]) -> String? {
+        for key in keys {
+            if let value = json[key] as? String {
+                let trimmed = value.trimmingCharacters(in: .whitespacesAndNewlines)
+                if !trimmed.isEmpty { return trimmed }
+            }
+        }
+        return nil
+    }
+}