feat: 聊天会话底部显示token消耗

electron/main/ai/agent.ts

@@ -79,6 +79,15 @@ export interface AgentConfig {
   abortSignal?: AbortSignal
 }
 
+/**
+ * Token 使用量
+ */
+export interface TokenUsage {
+  promptTokens: number
+  completionTokens: number
+  totalTokens: number
+}
+
 /**
  * Agent 流式响应 chunk
  */
@@ -97,6 +106,8 @@ export interface AgentStreamChunk {
   error?: string
   /** 是否完成 */
   isFinished?: boolean
+  /** Token 使用量（type=done 时返回累计值） */
+  usage?: TokenUsage
 }
 
 /**
@@ -109,6 +120,8 @@ export interface AgentResult {
   toolsUsed: string[]
   /** 工具调用轮数 */
   toolRounds: number
+  /** 总 Token 使用量（累计所有 LLM 调用） */
+  totalUsage?: TokenUsage
 }
 
 // ==================== 提示词配置类型 ====================
@@ -235,6 +248,8 @@ export class Agent {
   private historyMessages: ChatMessage[] = []
   private chatType: 'group' | 'private' = 'group'
   private promptConfig?: PromptConfig
+  /** 累计 Token 使用量 */
+  private totalUsage: TokenUsage = { promptTokens: 0, completionTokens: 0, totalTokens: 0 }
 
   constructor(
     context: ToolContext,
@@ -261,6 +276,17 @@ export class Agent {
     return this.abortSignal?.aborted ?? false
   }
 
+  /**
+   * 累加 Token 使用量
+   */
+  private addUsage(usage?: { promptTokens: number; completionTokens: number; totalTokens: number }): void {
+    if (usage) {
+      this.totalUsage.promptTokens += usage.promptTokens
+      this.totalUsage.completionTokens += usage.completionTokens
+      this.totalUsage.totalTokens += usage.totalTokens
+    }
+  }
+
   /**
    * 执行对话（非流式）
    * @param userMessage 用户消息
@@ -270,7 +296,7 @@ export class Agent {
 
     // 检查是否已中止
     if (this.isAborted()) {
-      return { content: '', toolsUsed: [], toolRounds: 0 }
+      return { content: '', toolsUsed: [], toolRounds: 0, totalUsage: this.totalUsage }
     }
 
     // 初始化消息（包含历史记录）
@@ -294,6 +320,7 @@ export class Agent {
           content: '',
           toolsUsed: this.toolsUsed,
           toolRounds: this.toolRounds,
+          totalUsage: this.totalUsage,
         }
       }
 
@@ -303,6 +330,9 @@ export class Agent {
         abortSignal: this.abortSignal,
       })
 
+      // 累加 Token 使用量
+      this.addUsage(response.usage)
+
       let toolCallsToProcess = response.tool_calls
 
       // 如果没有标准 tool_calls，尝试 fallback 解析
@@ -323,6 +353,7 @@ export class Agent {
               content: cleanContent,
               toolsUsed: this.toolsUsed,
               toolRounds: this.toolRounds,
+              totalUsage: this.totalUsage,
             }
           }
         } else {
@@ -332,6 +363,7 @@ export class Agent {
             content: response.content,
             toolsUsed: this.toolsUsed,
             toolRounds: this.toolRounds,
+            totalUsage: this.totalUsage,
           }
         }
       }
@@ -349,10 +381,12 @@ export class Agent {
     })
 
     const finalResponse = await chat(this.messages, this.config.llmOptions)
+    this.addUsage(finalResponse.usage)
     return {
       content: finalResponse.content,
       toolsUsed: this.toolsUsed,
       toolRounds: this.toolRounds,
+      totalUsage: this.totalUsage,
     }
   }
 
@@ -366,8 +400,8 @@ export class Agent {
 
     // 检查是否已中止
     if (this.isAborted()) {
-      onChunk({ type: 'done', isFinished: true })
-      return { content: '', toolsUsed: [], toolRounds: 0 }
+      onChunk({ type: 'done', isFinished: true, usage: this.totalUsage })
+      return { content: '', toolsUsed: [], toolRounds: 0, totalUsage: this.totalUsage }
     }
 
     // 初始化消息（包含历史记录）
@@ -387,11 +421,12 @@ export class Agent {
     while (this.toolRounds < this.config.maxToolRounds!) {
       // 每轮开始时检查是否中止
       if (this.isAborted()) {
-        onChunk({ type: 'done', isFinished: true })
+        onChunk({ type: 'done', isFinished: true, usage: this.totalUsage })
         return {
           content: finalContent,
           toolsUsed: this.toolsUsed,
           toolRounds: this.toolRounds,
+          totalUsage: this.totalUsage,
         }
       }
 
@@ -408,11 +443,12 @@ export class Agent {
       })) {
         // 每个 chunk 时检查是否中止
         if (this.isAborted()) {
-          onChunk({ type: 'done', isFinished: true })
+          onChunk({ type: 'done', isFinished: true, usage: this.totalUsage })
           return {
             content: finalContent + accumulatedContent,
             toolsUsed: this.toolsUsed,
             toolRounds: this.toolRounds,
+            totalUsage: this.totalUsage,
           }
         }
         if (chunk.content) {
@@ -444,6 +480,11 @@ export class Agent {
           toolCalls = chunk.tool_calls
         }
 
+        // 累加 Token 使用量（流式响应在最后一个 chunk 返回 usage）
+        if (chunk.usage) {
+          this.addUsage(chunk.usage)
+        }
+
         if (chunk.isFinished) {
           // 如果没有标准 tool_calls，尝试 fallback 解析
           if (chunk.finishReason !== 'tool_calls' || !toolCalls) {
@@ -467,22 +508,24 @@ export class Agent {
                 }
                 finalContent = cleanContent
                 aiLogger.info('Agent', 'AI 回复', finalContent)
-                onChunk({ type: 'done', isFinished: true })
+                onChunk({ type: 'done', isFinished: true, usage: this.totalUsage })
                 return {
                   content: finalContent,
                   toolsUsed: this.toolsUsed,
                   toolRounds: this.toolRounds,
+                  totalUsage: this.totalUsage,
                 }
               }
             } else {
               // 没有 tool_call 标签，正常完成
               finalContent = extractThinkingContent(accumulatedContent).cleanContent
               aiLogger.info('Agent', 'AI 回复', finalContent)
-              onChunk({ type: 'done', isFinished: true })
+              onChunk({ type: 'done', isFinished: true, usage: this.totalUsage })
               return {
                 content: finalContent,
                 toolsUsed: this.toolsUsed,
                 toolRounds: this.toolRounds,
+                totalUsage: this.totalUsage,
               }
             }
           }
@@ -532,11 +575,12 @@ export class Agent {
 
     // 检查是否已中止
     if (this.isAborted()) {
-      onChunk({ type: 'done', isFinished: true })
+      onChunk({ type: 'done', isFinished: true, usage: this.totalUsage })
       return {
         content: finalContent,
         toolsUsed: this.toolsUsed,
         toolRounds: this.toolRounds,
+        totalUsage: this.totalUsage,
       }
     }
 
@@ -551,15 +595,19 @@ export class Agent {
       abortSignal: this.abortSignal,
     })) {
       if (this.isAborted()) {
-        onChunk({ type: 'done', isFinished: true })
+        onChunk({ type: 'done', isFinished: true, usage: this.totalUsage })
         break
       }
       if (chunk.content) {
         finalContent += chunk.content
         onChunk({ type: 'content', content: chunk.content })
       }
+      // 累加 Token 使用量
+      if (chunk.usage) {
+        this.addUsage(chunk.usage)
+      }
       if (chunk.isFinished) {
-        onChunk({ type: 'done', isFinished: true })
+        onChunk({ type: 'done', isFinished: true, usage: this.totalUsage })
       }
     }
 
@@ -567,6 +615,7 @@ export class Agent {
       content: finalContent,
       toolsUsed: this.toolsUsed,
       toolRounds: this.toolRounds,
+      totalUsage: this.totalUsage,
     }
   }
 

electron/main/ai/llm/deepseek.ts

@@ -151,6 +151,8 @@ export class DeepSeekService implements ILLMService {
       temperature: options?.temperature ?? 0.7,
       max_tokens: options?.maxTokens ?? 2048,
       stream: true,
+      // 启用流式响应中的 usage 统计
+      stream_options: { include_usage: true },
     }
 
     if (options?.tools && options.tools.length > 0) {
@@ -265,6 +267,15 @@ export class DeepSeekService implements ILLMService {
                 reason = 'tool_calls'
               }
 
+              // 解析 usage 信息
+              const usage = parsed.usage
+                ? {
+                    promptTokens: parsed.usage.prompt_tokens,
+                    completionTokens: parsed.usage.completion_tokens,
+                    totalTokens: parsed.usage.total_tokens,
+                  }
+                : undefined
+
               // 如果有 tool_calls，返回它们
               if (toolCallsAccumulator.size > 0) {
                 const toolCalls: ToolCall[] = Array.from(toolCallsAccumulator.values()).map((tc) => ({
@@ -275,9 +286,9 @@ export class DeepSeekService implements ILLMService {
                     arguments: tc.arguments,
                   },
                 }))
-                yield { content: '', isFinished: true, finishReason: reason, tool_calls: toolCalls }
+                yield { content: '', isFinished: true, finishReason: reason, tool_calls: toolCalls, usage }
               } else {
-                yield { content: '', isFinished: true, finishReason: reason }
+                yield { content: '', isFinished: true, finishReason: reason, usage }
               }
               return
             }

electron/main/ai/llm/gemini.ts

@@ -396,6 +396,8 @@ export class GeminiService implements ILLMService {
     let buffer = ''
     // 用于累积工具调用（可能跨多个 chunk）
     const toolCallsAccumulator: ToolCall[] = []
+    // 用于追踪最新的 usage 信息
+    let latestUsage: { promptTokens: number; completionTokens: number; totalTokens: number } | undefined
 
     try {
       while (true) {
@@ -445,19 +447,28 @@ export class GeminiService implements ILLMService {
               toolCallsAccumulator.push(...toolCalls)
             }
 
+            // 更新 usage 信息
+            if (parsed.usageMetadata) {
+              latestUsage = {
+                promptTokens: parsed.usageMetadata.promptTokenCount || 0,
+                completionTokens: parsed.usageMetadata.candidatesTokenCount || 0,
+                totalTokens: parsed.usageMetadata.totalTokenCount || 0,
+              }
+            }
+
             // 检查是否完成
             const finishReason = candidate?.finishReason
             if (finishReason) {
-              aiLogger.info('Gemini', '流式响应完成', { finishReason, toolCallsCount: toolCallsAccumulator.length })
+              aiLogger.info('Gemini', '流式响应完成', { finishReason, toolCallsCount: toolCallsAccumulator.length, usage: latestUsage })
 
               if (toolCallsAccumulator.length > 0) {
-                yield { content: '', isFinished: true, finishReason: 'tool_calls', tool_calls: toolCallsAccumulator }
+                yield { content: '', isFinished: true, finishReason: 'tool_calls', tool_calls: toolCallsAccumulator, usage: latestUsage }
               } else {
                 let reason: ChatStreamChunk['finishReason'] = 'stop'
                 if (finishReason === 'MAX_TOKENS') {
                   reason = 'length'
                 }
-                yield { content: '', isFinished: true, finishReason: reason }
+                yield { content: '', isFinished: true, finishReason: reason, usage: latestUsage }
               }
               return
             }
@@ -470,9 +481,9 @@ export class GeminiService implements ILLMService {
 
       // 如果循环正常结束，发送完成信号
       if (toolCallsAccumulator.length > 0) {
-        yield { content: '', isFinished: true, finishReason: 'tool_calls', tool_calls: toolCallsAccumulator }
+        yield { content: '', isFinished: true, finishReason: 'tool_calls', tool_calls: toolCallsAccumulator, usage: latestUsage }
       } else {
-        yield { content: '', isFinished: true, finishReason: 'stop' }
+        yield { content: '', isFinished: true, finishReason: 'stop', usage: latestUsage }
       }
     } catch (error) {
       // 如果是中止错误，正常返回

electron/main/ai/llm/openai-compatible.ts

@@ -162,6 +162,8 @@ export class OpenAICompatibleService implements ILLMService {
       temperature: options?.temperature ?? 0.7,
       max_tokens: options?.maxTokens ?? 2048,
       stream: true,
+      // 启用流式响应中的 usage 统计（OpenAI API 兼容）
+      stream_options: { include_usage: true },
     }
 
     if (options?.tools && options.tools.length > 0) {
@@ -290,6 +292,15 @@ export class OpenAICompatibleService implements ILLMService {
                 reason = 'tool_calls'
               }
 
+              // 解析 usage 信息
+              const usage = parsed.usage
+                ? {
+                    promptTokens: parsed.usage.prompt_tokens,
+                    completionTokens: parsed.usage.completion_tokens,
+                    totalTokens: parsed.usage.total_tokens,
+                  }
+                : undefined
+
               if (toolCallsAccumulator.size > 0) {
                 const toolCalls: ToolCall[] = Array.from(toolCallsAccumulator.values()).map((tc) => ({
                   id: tc.id,
@@ -299,9 +310,9 @@ export class OpenAICompatibleService implements ILLMService {
                     arguments: tc.arguments,
                   },
                 }))
-                yield { content: '', isFinished: true, finishReason: reason, tool_calls: toolCalls }
+                yield { content: '', isFinished: true, finishReason: reason, tool_calls: toolCalls, usage }
               } else {
-                yield { content: '', isFinished: true, finishReason: reason }
+                yield { content: '', isFinished: true, finishReason: reason, usage }
               }
               return
             }

electron/main/ai/llm/qwen.ts

@@ -149,6 +149,8 @@ export class QwenService implements ILLMService {
       temperature: options?.temperature ?? 0.7,
       max_tokens: options?.maxTokens ?? 2048,
       stream: true,
+      // 启用流式响应中的 usage 统计
+      stream_options: { include_usage: true },
     }
 
     if (options?.tools && options.tools.length > 0) {
@@ -247,15 +249,24 @@ export class QwenService implements ILLMService {
               else if (finishReason === 'length') reason = 'length'
               else if (finishReason === 'tool_calls') reason = 'tool_calls'
 
+              // 解析 usage 信息
+              const usage = parsed.usage
+                ? {
+                    promptTokens: parsed.usage.prompt_tokens,
+                    completionTokens: parsed.usage.completion_tokens,
+                    totalTokens: parsed.usage.total_tokens,
+                  }
+                : undefined
+
               if (toolCallsAccumulator.size > 0) {
                 const toolCalls: ToolCall[] = Array.from(toolCallsAccumulator.values()).map((tc) => ({
                   id: tc.id,
                   type: 'function' as const,
                   function: { name: tc.name, arguments: tc.arguments },
                 }))
-                yield { content: '', isFinished: true, finishReason: reason, tool_calls: toolCalls }
+                yield { content: '', isFinished: true, finishReason: reason, tool_calls: toolCalls, usage }
               } else {
-                yield { content: '', isFinished: true, finishReason: reason }
+                yield { content: '', isFinished: true, finishReason: reason, usage }
               }
               return
             }

electron/main/ai/llm/types.ts

@@ -67,6 +67,12 @@ export interface ChatStreamChunk {
   finishReason?: 'stop' | 'length' | 'error' | 'tool_calls'
   /** 流式过程中的 tool_calls（增量） */
   tool_calls?: ToolCall[]
+  /** Token 使用量（通常在最后一个 chunk 中返回） */
+  usage?: {
+    promptTokens: number
+    completionTokens: number
+    totalTokens: number
+  }
 }
 
 // ==================== Function Calling 相关类型 ====================

electron/main/ipc/ai.ts

@@ -445,6 +445,7 @@ export function registerAIHandlers({ win }: IpcContext): void {
                 content: result.content,
                 toolsUsed: result.toolsUsed,
                 toolRounds: result.toolRounds,
+                totalUsage: result.totalUsage,
               },
             })
 
@@ -452,6 +453,7 @@ export function registerAIHandlers({ win }: IpcContext): void {
               toolsUsed: result.toolsUsed,
               toolRounds: result.toolRounds,
               contentLength: result.content.length,
+              totalUsage: result.totalUsage,
             })
           } catch (error) {
             // 如果是中止错误，不报告为错误

electron/preload/index.d.ts

@@ -294,6 +294,13 @@ interface LlmApi {
   ) => Promise<{ success: boolean; error?: string }>
 }
 
+// Token 使用量类型
+interface TokenUsage {
+  promptTokens: number
+  completionTokens: number
+  totalTokens: number
+}
+
 // Agent 相关类型
 interface AgentStreamChunk {
   type: 'content' | 'tool_start' | 'tool_result' | 'done' | 'error'
@@ -303,12 +310,16 @@ interface AgentStreamChunk {
   toolResult?: unknown
   error?: string
   isFinished?: boolean
+  /** Token 使用量（type=done 时返回累计值） */
+  usage?: TokenUsage
 }
 
 interface AgentResult {
   content: string
   toolsUsed: string[]
   toolRounds: number
+  /** 总 Token 使用量（累计所有 LLM 调用） */
+  totalUsage?: TokenUsage
 }
 
 interface ToolContext {
@@ -399,6 +410,7 @@ export {
   AgentResult,
   ToolContext,
   PromptConfig,
+  TokenUsage,
   CacheDirectoryInfo,
   CacheInfo,
 }

src/assets/docs/agreement.md

@@ -20,7 +20,7 @@
   - **严禁**用于窃取、监控或分析未经授权的他人隐私。
   - **严禁**将分析结果用于骚扰、诈骗、人肉搜索或任何侵犯他人权益的行为。
   - **严禁**用于规避安全措施或网络渗透测试。
-- **禁止商用**：我们严禁任何个人或机构将本软件用于任何形式的商业盈利行为（如打包销售、提供有偿代分析服务等），因此产生的纠纷与损失由用户自行承担。
+- **禁止商用**：严禁任何个人或机构将本软件用于任何形式的商业盈利行为（如打包销售、提供有偿代分析服务等），因此产生的纠纷与损失由用户自行承担。
 
 ## 4. 风险警告
 

src/components/analysis/ai/ChatExplorer.vue

@@ -27,6 +27,7 @@ const {
   currentConversationId,
   currentToolStatus,
   toolsUsedInCurrentRound,
+  sessionTokenUsage,
   sendMessage,
   loadConversation,
   startNewConversation,
@@ -514,8 +515,18 @@ watch(
                 </template>
               </UPopover>
 
-              <!-- 右侧：配置状态指示 -->
+              <!-- 右侧：Token 使用量 + 配置状态指示 -->
               <div class="flex items-center gap-3">
+                <!-- Token 使用量 -->
+                <div
+                  v-if="sessionTokenUsage.totalTokens > 0"
+                  class="flex items-center gap-1.5 text-xs text-gray-400"
+                  title="本次会话累计 Token 使用量"
+                >
+                  <UIcon name="i-heroicons-chart-bar-square" class="h-3.5 w-3.5" />
+                  <span>{{ sessionTokenUsage.totalTokens.toLocaleString() }} tokens</span>
+                </div>
+
                 <div
                   v-if="!isCheckingConfig"
                   class="flex items-center gap-1.5 text-xs transition-colors"

src/composables/useAIChat.ts

@@ -67,6 +67,13 @@ export interface ToolStatus {
   result?: unknown
 }
 
+// Token 使用量类型
+export interface TokenUsage {
+  promptTokens: number
+  completionTokens: number
+  totalTokens: number
+}
+
 // 工具名称映射（英文 → 中文）
 const TOOL_DISPLAY_NAMES: Record<string, string> = {
   search_messages: '搜索聊天记录',
@@ -105,6 +112,9 @@ export function useAIChat(
   const currentToolStatus = ref<ToolStatus | null>(null)
   const toolsUsedInCurrentRound = ref<string[]>([])
 
+  // Token 使用量（当前会话累计）
+  const sessionTokenUsage = ref<TokenUsage>({ promptTokens: 0, completionTokens: 0, totalTokens: 0 })
+
   // 中止控制
   let isAborted = false
   // 当前请求 ID，用于区分不同请求的响应
@@ -339,9 +349,18 @@ export function useAIChat(
               break
 
             case 'done':
-              // 完成
-              console.log('[AI] Agent 完成')
+              // 完成 - 更新 Token 使用量
+              console.log('[AI] Agent 完成', chunk.usage)
               currentToolStatus.value = null
+              // 更新会话累计 Token（流式响应在最后一个 chunk 返回 usage）
+              if (chunk.usage) {
+                sessionTokenUsage.value = {
+                  promptTokens: sessionTokenUsage.value.promptTokens + chunk.usage.promptTokens,
+                  completionTokens: sessionTokenUsage.value.completionTokens + chunk.usage.completionTokens,
+                  totalTokens: sessionTokenUsage.value.totalTokens + chunk.usage.totalTokens,
+                }
+                console.log('[AI] Token 使用量更新:', sessionTokenUsage.value)
+              }
               break
 
             case 'error':
@@ -506,6 +525,8 @@ export function useAIChat(
     messages.value = []
     sourceMessages.value = []
     currentKeywords.value = []
+    // 重置 Token 计数
+    sessionTokenUsage.value = { promptTokens: 0, completionTokens: 0, totalTokens: 0 }
 
     if (welcomeMessage) {
       messages.value.push({
@@ -573,6 +594,7 @@ export function useAIChat(
     currentConversationId,
     currentToolStatus,
     toolsUsedInCurrentRound,
+    sessionTokenUsage,
 
     // 方法
     sendMessage,