feat(chat): 前端显示上下文真实token消耗，/usage命令结果与cli格式同步

packages/client/src/api/hermes/chat.ts

@@ -85,6 +85,7 @@ export interface ResumeSessionPayload {
   inputTokens?: number
   outputTokens?: number
   contextTokens?: number
+  bridgeUsage?: any
   queueLength?: number
   queueMessages?: RunEvent['queued_messages']
 }

packages/client/src/components/hermes/chat/ChatInput.vue

@@ -252,6 +252,15 @@ watch(
 )
 
 const totalTokens = computed(() => {
+  // When upstream API data is available, prefer last_prompt_tokens
+  // (the single-call prompt cost) over the cumulative session accumulator,
+  // which double-counts the shared context across multiple tool calls.
+  const api = chatStore.activeSession?.apiUsage
+  if (api) {
+    if (api.lastPromptTokens && api.lastPromptTokens > 0) return api.lastPromptTokens
+    const promptTokens = (api.inputTokens || 0) + (api.cacheReadTokens || 0) + (api.cacheWriteTokens || 0)
+    if (promptTokens > 0) return promptTokens
+  }
   const context = chatStore.activeSession?.contextTokens
   if (typeof context === 'number' && Number.isFinite(context) && context > 0) return context
   const input = chatStore.activeSession?.inputTokens ?? 0

packages/client/src/stores/hermes/chat.ts

@@ -87,6 +87,32 @@ export interface Session {
   endedAt?: number | null
   lastActiveAt?: number
   workspace?: string | null
+  apiUsage?: {
+    inputTokens: number
+    outputTokens: number
+    cacheReadTokens?: number
+    cacheWriteTokens?: number
+    reasoningTokens?: number
+    totalTokens?: number
+    lastPromptTokens?: number
+    model?: string
+    costStatus?: string
+    actualCostUsd?: number
+    estimatedCostUsd?: number
+  }
+}
+
+/** Apply token usage from a run.* event, preferring upstream API data */
+function applyApiUsage(target: Session, evt: any): void {
+  const apiUsage = evt.apiUsage
+  if (apiUsage) {
+    target.inputTokens = apiUsage.inputTokens
+    target.outputTokens = apiUsage.outputTokens
+    target.apiUsage = apiUsage
+  } else if (evt.inputTokens != null) {
+    target.inputTokens = evt.inputTokens
+    target.outputTokens = evt.outputTokens
+  }
 }
 
 interface CompressionState {
@@ -479,11 +505,13 @@ export const useChatStore = defineStore('chat', () => {
       const runtimeByIdBefore = new Map(sessions.value.map(s => [s.id, {
         messages: s.messages,
         contextTokens: s.contextTokens,
+        apiUsage: s.apiUsage,
       }]))
       for (const s of fresh) {
         const prev = runtimeByIdBefore.get(s.id)
         if (prev?.messages?.length) s.messages = prev.messages
         if (prev?.contextTokens != null) s.contextTokens = prev.contextTokens
+        if (prev?.apiUsage) s.apiUsage = prev.apiUsage
       }
       sessions.value = fresh
 
@@ -630,6 +658,25 @@ export const useChatStore = defineStore('chat', () => {
           if (data.inputTokens != null) target.inputTokens = data.inputTokens
           if (data.outputTokens != null) target.outputTokens = data.outputTokens
           if ((data as any).contextTokens != null) target.contextTokens = (data as any).contextTokens
+          // Rebuild apiUsage from bridge data sent by server on resume.
+          // Only when lastPromptTokens is available — stale bridgeUsage
+          // without it carries cumulative session_* values.
+          const bu = (data as any).bridgeUsage
+          if (bu && bu.lastPromptTokens != null) {
+            target.apiUsage = {
+              inputTokens: bu.inputTokens,
+              outputTokens: bu.outputTokens,
+              cacheReadTokens: bu.cacheReadTokens,
+              cacheWriteTokens: bu.cacheWriteTokens,
+              reasoningTokens: bu.reasoningTokens,
+              totalTokens: bu.totalTokens,
+              lastPromptTokens: bu.lastPromptTokens,
+              model: bu.model,
+              costStatus: bu.costStatus,
+              actualCostUsd: bu.actualCostUsd,
+              estimatedCostUsd: bu.estimatedCostUsd,
+            }
+          }
           if (data.messages?.length) {
             target.messages = mapHermesMessages(data.messages as any[])
             target.loadedMessageCount = data.messageLoadedCount ?? data.messages.length
@@ -980,9 +1027,33 @@ export const useChatStore = defineStore('chat', () => {
     }
 
     if (action === 'usage' && target) {
-      target.inputTokens = (evt as any).inputTokens
-      target.outputTokens = (evt as any).outputTokens
-      if ((evt as any).contextTokens != null) target.contextTokens = (evt as any).contextTokens
+      // Only apply values from live bridge data (signaled by lastPromptTokens).
+      // DB fallback carries cumulative/inflated session_* counters — not safe
+      // for the progress bar or context display.
+      const isLive = (evt as any).lastPromptTokens != null
+      if (isLive) {
+        target.inputTokens = (evt as any).inputTokens
+        target.outputTokens = (evt as any).outputTokens
+        if ((evt as any).contextTokens != null) target.contextTokens = (evt as any).contextTokens
+      }
+      // Only build apiUsage from live bridge data (signaled by lastPromptTokens).
+      // DB fallback lacks lastPromptTokens and carries cumulative/inflated values
+      // from the previous turn's session_* counters — not safe for progress bar.
+      if (isLive) {
+        target.apiUsage = {
+          inputTokens: (evt as any).inputTokens,
+          outputTokens: (evt as any).outputTokens,
+          cacheReadTokens: (evt as any).cacheReadTokens,
+          cacheWriteTokens: (evt as any).cacheWriteTokens,
+          reasoningTokens: (evt as any).reasoningTokens,
+          totalTokens: (evt as any).totalTokens,
+          lastPromptTokens: (evt as any).lastPromptTokens,
+          model: (evt as any).model,
+          costStatus: (evt as any).costStatus,
+          actualCostUsd: (evt as any).actualCostUsd,
+          estimatedCostUsd: (evt as any).estimatedCostUsd,
+        }
+      }
     }
 
     if (action === 'destroy') {
@@ -1481,6 +1552,23 @@ export const useChatStore = defineStore('chat', () => {
         if (data.inputTokens != null) target.inputTokens = data.inputTokens
         if (data.outputTokens != null) target.outputTokens = data.outputTokens
         if (data.contextTokens != null) target.contextTokens = data.contextTokens
+        // Rebuild apiUsage from bridge data on reconnect resume
+        const bu = (data as any).bridgeUsage
+        if (bu && bu.lastPromptTokens != null) {
+          target.apiUsage = {
+            inputTokens: bu.inputTokens,
+            outputTokens: bu.outputTokens,
+            cacheReadTokens: bu.cacheReadTokens,
+            cacheWriteTokens: bu.cacheWriteTokens,
+            reasoningTokens: bu.reasoningTokens,
+            totalTokens: bu.totalTokens,
+            lastPromptTokens: bu.lastPromptTokens,
+            model: bu.model,
+            costStatus: bu.costStatus,
+            actualCostUsd: bu.actualCostUsd,
+            estimatedCostUsd: bu.estimatedCostUsd,
+          }
+        }
 
         if (Array.isArray(data.messages)) {
           target.messages = mapHermesMessages(data.messages as any[])
@@ -1834,11 +1922,10 @@ export const useChatStore = defineStore('chat', () => {
                 updateMessage(sid, lastMsg.id, { isStreaming: false })
               }
               // Server-computed usage (local countTokens, snapshot-aware)
-              if ((evt as any).inputTokens != null) {
+              if ((evt as any).inputTokens != null || (evt as any).apiUsage) {
                 const target = sessions.value.find(s => s.id === sid)
                 if (target) {
-                  target.inputTokens = (evt as any).inputTokens
-                  target.outputTokens = (evt as any).outputTokens
+                  applyApiUsage(target, evt)
                   if ((evt as any).contextTokens != null) target.contextTokens = (evt as any).contextTokens
                 }
               }
@@ -1930,11 +2017,10 @@ export const useChatStore = defineStore('chat', () => {
 
             case 'run.failed': {
               clearAgentEventMessages(sid)
-              if ((evt as any).inputTokens != null) {
+              if ((evt as any).inputTokens != null || (evt as any).apiUsage) {
                 const target = sessions.value.find(s => s.id === sid)
                 if (target) {
-                  target.inputTokens = (evt as any).inputTokens
-                  target.outputTokens = (evt as any).outputTokens
+                  applyApiUsage(target, evt)
                   if ((evt as any).contextTokens != null) target.contextTokens = (evt as any).contextTokens
                 }
               }
@@ -2313,11 +2399,10 @@ export const useChatStore = defineStore('chat', () => {
             updateMessage(sid, lastMsg.id, { isStreaming: false })
           }
           // Server-computed usage (local countTokens, snapshot-aware)
-          if ((evt as any).inputTokens != null) {
+          if ((evt as any).inputTokens != null || (evt as any).apiUsage) {
             const target = sessions.value.find(s => s.id === sid)
             if (target) {
-              target.inputTokens = (evt as any).inputTokens
-              target.outputTokens = (evt as any).outputTokens
+              applyApiUsage(target, evt)
               if ((evt as any).contextTokens != null) target.contextTokens = (evt as any).contextTokens
             }
           }
@@ -2389,11 +2474,10 @@ export const useChatStore = defineStore('chat', () => {
 
         case 'run.failed': {
           clearAgentEventMessages(sid)
-          if ((evt as any).inputTokens != null) {
+          if ((evt as any).inputTokens != null || (evt as any).apiUsage) {
             const target = sessions.value.find(s => s.id === sid)
             if (target) {
-              target.inputTokens = (evt as any).inputTokens
-              target.outputTokens = (evt as any).outputTokens
+              applyApiUsage(target, evt)
               if ((evt as any).contextTokens != null) target.contextTokens = (evt as any).contextTokens
             }
           }

packages/server/src/db/hermes/schemas.ts

@@ -50,6 +50,7 @@ export const SESSIONS_SCHEMA: Record<string, string> = {
   estimated_cost_usd: 'REAL NOT NULL DEFAULT 0',
   actual_cost_usd: 'REAL',
   cost_status: 'TEXT NOT NULL DEFAULT \'\'',
+  last_prompt_tokens: 'INTEGER NOT NULL DEFAULT 0',
   preview: 'TEXT NOT NULL DEFAULT \'\'',
   last_active: 'INTEGER NOT NULL',
   workspace: 'TEXT',

packages/server/src/db/hermes/session-store.ts

@@ -29,6 +29,7 @@ export interface HermesSessionRow {
   estimated_cost_usd: number
   actual_cost_usd: number | null
   cost_status: string
+  last_prompt_tokens: number
   preview: string
   last_active: number
   workspace: string | null
@@ -103,6 +104,7 @@ function mapSessionRow(row: Record<string, unknown>): HermesSessionRow {
     estimated_cost_usd: Number(row.estimated_cost_usd || 0),
     actual_cost_usd: row.actual_cost_usd != null ? Number(row.actual_cost_usd) : null,
     cost_status: String(row.cost_status || ''),
+    last_prompt_tokens: Number(row.last_prompt_tokens || 0),
     preview: String(row.preview || ''),
     last_active: Number(row.last_active || 0),
     workspace: row.workspace != null ? String(row.workspace) : null,
@@ -148,7 +150,7 @@ export function createSession(data: {
       message_count: 0, tool_call_count: 0,
       input_tokens: 0, output_tokens: 0, cache_read_tokens: 0, cache_write_tokens: 0, reasoning_tokens: 0,
       billing_provider: null, estimated_cost_usd: 0, actual_cost_usd: null,
-      cost_status: '', preview: '', last_active: now, workspace: data.workspace || null,
+      cost_status: '', last_prompt_tokens: 0, preview: '', last_active: now, workspace: data.workspace || null,
     }
   }
   const db = getDb()!
@@ -177,6 +179,7 @@ export function updateSession(id: string, data: Partial<Omit<HermesSessionRow, '
     if (key === 'id' || key === 'profile') continue
     // Skip last_active and ended_at - handle them separately below
     if (key === 'last_active' || key === 'ended_at') continue
+    if (val === undefined) continue
     fields.push(`"${key}" = ?`)
     values.push(val)
   }

packages/server/src/services/hermes/run-chat/handle-bridge-run.ts

@@ -26,7 +26,7 @@ import {
   recordBridgeToolCompleted,
 } from './bridge-message'
 import { summarizeToolArguments } from './response-utils'
-import type { ContentBlock, QueuedRun, SessionState } from './types'
+import type { ContentBlock, QueuedRun, SessionState, BridgeUsageState } from './types'
 import type { ChatMessage } from '../../../lib/context-compressor'
 import { resolveBridgeRunModelConfig, type RunModelGroup } from './model-config'
 import { filterBridgeToolCallMarkupDelta, flushPendingToolCallMarkup } from './bridge-delta'
@@ -106,6 +106,30 @@ function finiteToken(value: unknown): number | undefined {
     : undefined
 }
 
+function extractBridgeUsage(result: unknown): BridgeUsageState | undefined {
+  if (!result || typeof result !== 'object' || Array.isArray(result)) return undefined
+  const r = result as Record<string, unknown>
+  const inputTokens = finiteToken(r.input_tokens)
+  if (inputTokens == null) return undefined
+  return {
+    inputTokens,
+    outputTokens: finiteToken(r.output_tokens) ?? 0,
+    cacheReadTokens: finiteToken(r.cache_read_tokens) ?? 0,
+    cacheWriteTokens: finiteToken(r.cache_write_tokens) ?? 0,
+    reasoningTokens: finiteToken(r.reasoning_tokens) ?? 0,
+    promptTokens: finiteToken(r.prompt_tokens) ?? 0,
+    completionTokens: finiteToken(r.completion_tokens) ?? 0,
+    totalTokens: finiteToken(r.total_tokens) ?? 0,
+    apiCalls: finiteToken(r.api_calls) ?? 0,
+    lastPromptTokens: finiteToken(r.last_prompt_tokens),
+    model: typeof r.model === 'string' ? r.model : undefined,
+    estimatedCostUsd: typeof r.estimated_cost_usd === 'number' ? r.estimated_cost_usd : undefined,
+    actualCostUsd: typeof r.actual_cost_usd === 'number' ? r.actual_cost_usd : undefined,
+    costStatus: typeof r.cost_status === 'string' ? r.cost_status : undefined,
+    costSource: typeof r.cost_source === 'string' ? r.cost_source : undefined,
+  }
+}
+
 function cacheBridgeContext(state: SessionState, data: Record<string, unknown> | AgentBridgeContextEstimate) {
   const fixedContextTokens = finiteToken(data.fixed_context_tokens)
   if (fixedContextTokens == null) return
@@ -248,6 +272,7 @@ export async function handleBridgeRun(
   state.bridgePendingAssistantContent = ''
   state.bridgePendingReasoningContent = ''
   state.bridgePendingToolCallMarkup = ''
+  state.bridgeUsage = undefined
   state.bridgeToolCounter = 0
   state.bridgePendingTools = []
   state.responseRun = undefined
@@ -453,6 +478,7 @@ export async function handleBridgeRun(
       inputTokens: errUsage.inputTokens,
       outputTokens: errUsage.outputTokens,
       contextTokens: errContextTokens,
+      apiUsage: state.bridgeUsage,
       queue_remaining: queueLen,
     })
     if (queueLen > 0) dequeueNextQueuedRun(socket, session_id)
@@ -904,6 +930,25 @@ async function applyBridgeChunkAsync(
     outputTokens: usage.outputTokens,
     profile: state.profile,
   })
+  // Extract upstream API token usage from bridge result (hermes agent's
+  // conversation_loop.py already includes input_tokens, output_tokens,
+  // cache_*, reasoning_*, total_tokens, api_calls, and cost fields).
+  state.bridgeUsage = extractBridgeUsage(chunk.result)
+  // Persist upstream API token values in the Web UI session store so
+  // they survive page reloads and the frontend shows correct numbers.
+  if (state.bridgeUsage) {
+    updateSession(sessionId, {
+      input_tokens: state.bridgeUsage.inputTokens,
+      output_tokens: state.bridgeUsage.outputTokens,
+      cache_read_tokens: state.bridgeUsage.cacheReadTokens,
+      cache_write_tokens: state.bridgeUsage.cacheWriteTokens,
+      reasoning_tokens: state.bridgeUsage.reasoningTokens,
+      estimated_cost_usd: state.bridgeUsage.estimatedCostUsd,
+      actual_cost_usd: state.bridgeUsage.actualCostUsd,
+      cost_status: state.bridgeUsage.costStatus,
+      last_prompt_tokens: state.bridgeUsage.lastPromptTokens,
+    })
+  }
   const terminalError = bridgeTerminalError(chunk)
   const hadQueuedRunBeforeGoalEvaluation = state.queue.length > 0
   state.isWorking = hadQueuedRunBeforeGoalEvaluation
@@ -923,6 +968,7 @@ async function applyBridgeChunkAsync(
     inputTokens: usage.inputTokens,
     outputTokens: usage.outputTokens,
     contextTokens,
+    apiUsage: state.bridgeUsage,
     queue_remaining: state.queue.length,
   }
   emit(eventName, payload)

packages/server/src/services/hermes/run-chat/index.ts

@@ -344,6 +344,7 @@ export class ChatRunSocket {
       inputTokens: state.inputTokens,
       outputTokens: state.outputTokens,
       contextTokens: state.contextTokens,
+      bridgeUsage: state.bridgeUsage,
       queueLength: state.queue?.length || 0,
       queueMessages: this.serializeQueuedMessages(state.queue || []),
     })