feat(stats): merge token statistics onto latest main

app/build.gradle.kts

@@ -765,6 +765,10 @@ dependencies {
     androidTestImplementation(libs.androidx.espresso.core)
     androidTestImplementation(platform(libs.compose.bom))
 
+    // 单元测试中真实 org.json（Android 桩在 JVM 测试里会抛 Stub! 异常）；
+    // 统计 usage 归一化测试需要解析 JSONObject。
+    testImplementation("org.json:json:20240303")
+
     // Apache POI - for Document processing (DOC, DOCX, etc.)
     implementation(libs.poi)
     implementation(libs.poi.ooxml)

app/src/main/java/com/ai/assistance/operit/api/chat/EnhancedAIService.kt

@@ -1110,7 +1110,13 @@ class EnhancedAIService private constructor(private val context: Context) {
                                         currentRequestCachedInputTokenCount = cachedInput.coerceAtLeast(0)
                                         _perRequestTokenCounts.value = Pair(input, output)
                                     },
-                                    onNonFatalError = onNonFatalError
+                                    onNonFatalError = onNonFatalError,
+                                    statsCategory =
+                                        if (isSubTask) {
+                                            com.ai.assistance.operit.data.stats.TokenStatCategory.SUBAGENT
+                                        } else {
+                                            com.ai.assistance.operit.data.stats.TokenStatCategory.CHAT
+                                        }
                             )
                     val revisableStream = responseStream as? TextStreamEventCarrier
 
@@ -1211,10 +1217,6 @@ class EnhancedAIService private constructor(private val context: Context) {
                     currentRequestInputTokenCount = 0L
                     currentRequestOutputTokenCount = 0L
                     currentRequestCachedInputTokenCount = 0L
-                    apiPreferences.updateTokensForProviderModel(serviceForFunction.providerModel, inputTokens, outputTokens, cachedInputTokens)
-
-                    // Update request count
-                    apiPreferences.incrementRequestCountForProviderModel(serviceForFunction.providerModel)
 
                     AppLogger.d(
                             TAG,
@@ -2330,7 +2332,13 @@ class EnhancedAIService private constructor(private val context: Context) {
                                     currentRequestCachedInputTokenCount = cachedInput.coerceAtLeast(0)
                                     _perRequestTokenCounts.value = Pair(input, output)
                                 },
-                                onNonFatalError = onNonFatalError
+                                onNonFatalError = onNonFatalError,
+                                statsCategory =
+                                    if (isSubTask) {
+                                        com.ai.assistance.operit.data.stats.TokenStatCategory.SUBAGENT
+                                    } else {
+                                        com.ai.assistance.operit.data.stats.TokenStatCategory.CHAT
+                                    }
                         )
 
                 // 更新状态为接收中
@@ -2425,10 +2433,6 @@ class EnhancedAIService private constructor(private val context: Context) {
                 currentRequestInputTokenCount = 0L
                 currentRequestOutputTokenCount = 0L
                 currentRequestCachedInputTokenCount = 0L
-                apiPreferences.updateTokensForProviderModel(serviceForFunction.providerModel, inputTokens, outputTokens, cachedInputTokens)
-
-                // Update request count
-                apiPreferences.incrementRequestCountForProviderModel(serviceForFunction.providerModel)
 
                 AppLogger.d(
                         TAG,

app/src/main/java/com/ai/assistance/operit/api/chat/enhance/ConversationService.kt

@@ -269,7 +269,8 @@ class ConversationService(
                     summaryService.sendMessage(
                             context = context,
                             chatHistory = preparedHistory,
-                            modelParameters = modelParameters
+                            modelParameters = modelParameters,
+                            statsCategory = com.ai.assistance.operit.data.stats.TokenStatCategory.SUMMARY
                     )
 
             // 收集流中的所有内容
@@ -320,18 +321,7 @@ class ConversationService(
                 return "Conversation Summary: Unable to generate valid summary."
             }
 
-            // 将总结token计数添加到用户偏好分析的token统计中
-            try {
-                AppLogger.d(TAG, "总结生成使用了输入token: $inputTokens, 缓存token: $cachedInputTokens, 输出token: $outputTokens")
-                apiPreferences.updateTokensForProviderModel(summaryService.providerModel, inputTokens, outputTokens, cachedInputTokens)
-
-                // Update request count for summary generation
-                apiPreferences.incrementRequestCountForProviderModel(summaryService.providerModel)
-
-                AppLogger.d(TAG, "已将总结token统计添加到用户偏好分析token计数中")
-            } catch (e: Exception) {
-                AppLogger.e(TAG, "更新token统计失败", e)
-            }
+            AppLogger.d(TAG, "总结生成使用了输入token: $inputTokens, 缓存token: $cachedInputTokens, 输出token: $outputTokens")
 
             return summaryContent
         } catch (e: Exception) {
@@ -368,28 +358,18 @@ class ConversationService(
                 chatHistory = preparedHistory,
                 modelParameters = modelParameters,
                 stream = false,
-                enableRetry = false
+                enableRetry = false,
+                statsCategory = com.ai.assistance.operit.data.stats.TokenStatCategory.TITLE
             ).collect { content -> contentBuilder.append(content) }
 
             val title = sanitizeConversationTitle(
                 ChatUtils.removeThinkingContent(contentBuilder.toString().trim())
             )
 
-            try {
-                val inputTokens = titleService.inputTokenCount
-                val cachedInputTokens = titleService.cachedInputTokenCount
-                val outputTokens = titleService.outputTokenCount
-                apiPreferences.updateTokensForProviderModel(
-                    titleService.providerModel,
-                    inputTokens,
-                    outputTokens,
-                    cachedInputTokens
-                )
-                apiPreferences.incrementRequestCountForProviderModel(titleService.providerModel)
-                AppLogger.d(TAG, "标题生成使用了输入token: $inputTokens, 缓存token: $cachedInputTokens, 输出token: $outputTokens")
-            } catch (e: Exception) {
-                AppLogger.e(TAG, "更新标题生成token统计失败", e)
-            }
+            val inputTokens = titleService.inputTokenCount
+            val cachedInputTokens = titleService.cachedInputTokenCount
+            val outputTokens = titleService.outputTokenCount
+            AppLogger.d(TAG, "标题生成使用了输入token: $inputTokens, 缓存token: $cachedInputTokens, 输出token: $outputTokens")
 
             title
         } catch (e: Exception) {
@@ -1132,7 +1112,8 @@ ${FunctionalPrompts.translationUserPrompt(targetLanguage, text)}
             val stream = translationService.sendMessage(
                 context = context,
                 chatHistory = chatHistory + PromptTurn(kind = PromptTurnKind.USER, content = translationPrompt),
-                modelParameters = modelParameters
+                modelParameters = modelParameters,
+                statsCategory = com.ai.assistance.operit.data.stats.TokenStatCategory.OTHER
             )
 
             stream.collect { content ->
@@ -1191,7 +1172,8 @@ ${FunctionalPrompts.translationUserPrompt(targetLanguage, text)}
             val stream = summaryService.sendMessage(
                 context = context,
                 chatHistory = chatHistory + PromptTurn(kind = PromptTurnKind.USER, content = descriptionPrompt),
-                modelParameters = modelParameters
+                modelParameters = modelParameters,
+                statsCategory = com.ai.assistance.operit.data.stats.TokenStatCategory.OTHER
             )
 
             stream.collect { content ->
@@ -1249,7 +1231,8 @@ ${FunctionalPrompts.translationUserPrompt(targetLanguage, text)}
             service.sendMessage(
                 context = context,
                 chatHistory = listOf(PromptTurn(kind = PromptTurnKind.USER, content = prompt)),
-                modelParameters = modelParameters
+                modelParameters = modelParameters,
+                statsCategory = com.ai.assistance.operit.data.stats.TokenStatCategory.OTHER
             ).collect { chunk ->
                 result.append(chunk)
             }
@@ -1294,7 +1277,8 @@ ${FunctionalPrompts.translationUserPrompt(targetLanguage, text)}
             service.sendMessage(
                 context = context,
                 chatHistory = listOf(PromptTurn(kind = PromptTurnKind.USER, content = prompt)),
-                modelParameters = modelParameters
+                modelParameters = modelParameters,
+                statsCategory = com.ai.assistance.operit.data.stats.TokenStatCategory.OTHER
             ).collect { chunk ->
                 result.append(chunk)
             }
@@ -1337,7 +1321,8 @@ ${FunctionalPrompts.translationUserPrompt(targetLanguage, text)}
             service.sendMessage(
                 context = context,
                 chatHistory = listOf(PromptTurn(kind = PromptTurnKind.USER, content = prompt)),
-                modelParameters = modelParameters
+                modelParameters = modelParameters,
+                statsCategory = com.ai.assistance.operit.data.stats.TokenStatCategory.OTHER
             ).collect { chunk ->
                 result.append(chunk)
             }

app/src/main/java/com/ai/assistance/operit/api/chat/library/MemoryLibrary.kt

@@ -256,22 +256,12 @@ object MemoryLibrary {
             val stream =
                 aiService.sendMessage(
                     context = context,
-                    chatHistory = messages
+                    chatHistory = messages,
+                    statsCategory = com.ai.assistance.operit.data.stats.TokenStatCategory.MEMORY
                 )
             stream.collect { content -> result.append(content) }
         }
-
-        // 更新 token 统计
-        apiPreferences?.updateTokensForProviderModel(
-            aiService.providerModel,
-            aiService.inputTokenCount,
-            aiService.outputTokenCount,
-            aiService.cachedInputTokenCount
-        )
-
-        // Update request count
-        apiPreferences?.incrementRequestCountForProviderModel(aiService.providerModel)
-
+
         // 解析 AI 返回的 JSON 并更新记忆
         parseAndApplyCategorization(result.toString(), memories, repository)
     }
@@ -670,21 +660,12 @@ object MemoryLibrary {
                 val stream =
                     aiService.sendMessage(
                         context = context,
-                        chatHistory = messages
+                        chatHistory = messages,
+                        statsCategory = com.ai.assistance.operit.data.stats.TokenStatCategory.MEMORY
                     )
                 stream.collect { content -> result.append(content) }
             }
 
-            apiPreferences?.updateTokensForProviderModel(
-                    aiService.providerModel,
-                    aiService.inputTokenCount,
-                    aiService.outputTokenCount,
-                    aiService.cachedInputTokenCount
-            )
-
-            // Update request count
-            apiPreferences?.incrementRequestCountForProviderModel(aiService.providerModel)
-
             return parseAnalysisResult(ChatUtils.removeThinkingContent(result.toString()))
         } catch (e: CancellationException) {
             throw e

app/src/main/java/com/ai/assistance/operit/api/chat/llmprovider/AIService.kt

@@ -5,6 +5,8 @@ import com.ai.assistance.operit.core.chat.hooks.PromptTurn
 import com.ai.assistance.operit.data.model.ModelParameter
 import com.ai.assistance.operit.data.model.ModelOption
 import com.ai.assistance.operit.data.model.ToolPrompt
+import com.ai.assistance.operit.data.stats.ProviderUsageSnapshot
+import com.ai.assistance.operit.data.stats.TokenStatCategory
 import com.ai.assistance.operit.util.stream.Stream
 
 /** AI服务接口，定义与不同AI提供商进行交互的标准方法 */
@@ -44,8 +46,14 @@ interface AIService {
      * @param enableThinking 是否启用思考模式
      * @param stream 是否使用流式输出，true为流式，false为非流式（但返回值仍为Stream）
      * @param availableTools 可用工具列表(用于Tool Call API)，如果为null则使用系统提示词中的工具描述
-     * @param onTokensUpdated Token更新回调
+     * @param onTokensUpdated Token更新回调（UI 计数通道，可能携带估算值）
+     * @param onUsageReported 规范化 usage 上报回调（统计账本通道；只在解析到
+     *   provider 真实 usage/本地实测计数时回调，估算值不上报；可被多次调用，
+     *   第二次参数为 provider 内部尝试序号 attempt（从 1 开始，内部重试递增），
+     *   记录方按 attempt 聚合：同一 attempt 取最后一次，不同 attempt 累加）
      * @param onNonFatalError 非致命错误回调
+     * @param enableRetry 是否允许内部重试
+     * @param statsCategory 业务分类（统计账本）；null 表示调用方未声明（按 OTHER 记录）
      * @return 流式响应内容的Stream（无论stream参数如何，都返回Stream）
      */
     suspend fun sendMessage(
@@ -57,17 +65,24 @@ interface AIService {
             availableTools: List<ToolPrompt>? = null,
             preserveThinkInHistory: Boolean = false, // 新增参数，控制是否保留历史中的思考过程
             onTokensUpdated: suspend (input: Long, cachedInput: Long, output: Long) -> Unit = { _, _, _ -> },
+            onUsageReported: (suspend (ProviderUsageSnapshot, attempt: Int) -> Unit)? = null,
             onNonFatalError: suspend (error: String) -> Unit = {},
-            enableRetry: Boolean = true
+            enableRetry: Boolean = true,
+            statsCategory: TokenStatCategory? = null
     ): Stream<String>
 
     /**
      * 测试与AI服务的连接
      *
      * @param context Android Context
+     * @param onUsageReported 与 [sendMessage] 相同的 usage 上报回调；实现内部
+     *   通过 sendMessage 发起测试模型调用时必须透传，使探测用量进入统计账本。
      * @return 成功时返回成功信息，失败时返回包含错误的Result
      */
-    suspend fun testConnection(context: Context): Result<String>
+    suspend fun testConnection(
+        context: Context,
+        onUsageReported: (suspend (ProviderUsageSnapshot, attempt: Int) -> Unit)? = null
+    ): Result<String>
 
     /**
      * 精确计算下一次请求的输入Token数量

app/src/main/java/com/ai/assistance/operit/api/chat/llmprovider/AIServiceFactory.kt

@@ -255,7 +255,11 @@ object AIServiceFactory {
     }
 
     /**
-     * 创建AI服务实例
+     * 创建AI服务实例（统一统计记录边界）。
+     *
+     * 所有服务（包括连接测试器直接创建的探测服务）都在这里统一包装
+     * [TokenTrackingAIService]：任何 sendMessage/testConnection 调用都会落入
+     * 统计账本，业务分类由调用方通过 sendMessage 的 statsCategory 声明。
      *
      * @param config 模型配置数据
      * @param modelConfigManager 模型配置管理器，用于多API Key模式
@@ -266,6 +270,19 @@ object AIServiceFactory {
         config: ModelConfigData,
         modelConfigManager: ModelConfigManager,
         context: Context
+    ): AIService {
+        val rawService = buildService(config, modelConfigManager, context)
+        return TokenTrackingAIService(
+            delegate = rawService,
+            context = context,
+            configId = config.id,
+        )
+    }
+
+    private fun buildService(
+        config: ModelConfigData,
+        modelConfigManager: ModelConfigManager,
+        context: Context
     ): AIService {
         val providerTypeId = config.apiProviderTypeId.trim()
         ToolPkgAiProviderRegistry.get(providerTypeId)?.let { provider ->
@@ -300,7 +317,21 @@ object AIServiceFactory {
 
         return when (providerType) {
             // OpenAI格式，支持原生和兼容OpenAI API的服务
-            ApiProviderType.OPENAI,
+            ApiProviderType.OPENAI ->
+                OpenAIProvider(
+                    apiEndpoint = config.apiEndpoint,
+                    apiKeyProvider = apiKeyProvider,
+                    modelName = config.modelName,
+                    client = httpClient,
+                    customHeaders = customHeaders,
+                    providerType = providerType,
+                    supportsVision = supportsVision,
+                    supportsAudio = supportsAudio,
+                    supportsVideo = supportsVideo,
+                    enableToolCall = enableToolCall,
+                    includeUsageInStream = true,
+                )
+
             ApiProviderType.OPENAI_GENERIC,
             ApiProviderType.OPENAI_LOCAL ->
                 OpenAIProvider(