feat: Antigravity 端点签名迁移至 gemini:chat，流式 usage 提取重构与请求详情自动刷新

- Antigravity 端点签名从 gemini:cli 统一为 gemini:chat，保留向后兼容，含 DB 迁移
- 流式处理中 usage/completion/text 提取抽离为 _update_ctx_from_provider_event，
  支持 envelope 解包后提取，避免格式转换流程中重复计数
- 新增 has_format_conversion 属性，区分真正的格式转换与 envelope rewrite，
  修正 usage 展示层的格式转换标记
- 请求详情抽屉对未完成请求支持自动轮询刷新，关闭时自动停止
- 按次计费样式调整；实际成本仅在倍率非 1.0 时显示；缓存日志降级为 trace

alembic/versions/20260206_2345_e1b2c3d4f5a6_antigravity_api_format_to_gemini_chat.py

@@ -0,0 +1,116 @@
+"""Update Antigravity endpoint signature to gemini:chat
+
+Revision ID: e1b2c3d4f5a6
+Revises: b5c6d7e8f9a0
+Create Date: 2026-02-06 23:45:00.000000
+
+"""
+
+from __future__ import annotations
+
+from collections.abc import Sequence
+
+from sqlalchemy import text
+
+from alembic import op
+
+# revision identifiers, used by Alembic.
+revision: str = "e1b2c3d4f5a6"
+down_revision: str | None = "b5c6d7e8f9a0"
+branch_labels: str | Sequence[str] | None = None
+depends_on: str | Sequence[str] | None = None
+
+
+def upgrade() -> None:
+    conn = op.get_bind()
+
+    # --- provider_endpoints ---
+    # Update only when there is no conflicting gemini:chat endpoint for the same provider
+    # (provider_endpoints has a unique constraint on (provider_id, api_format)).
+    conn.execute(text("""
+            UPDATE provider_endpoints pe
+            SET
+                api_format = 'gemini:chat',
+                api_family = 'gemini',
+                endpoint_kind = 'chat'
+            WHERE pe.api_format = 'gemini:cli'
+              AND pe.provider_id IN (
+                SELECT id FROM providers WHERE lower(provider_type) = 'antigravity'
+              )
+              AND NOT EXISTS (
+                SELECT 1 FROM provider_endpoints pe2
+                WHERE pe2.provider_id = pe.provider_id
+                  AND pe2.api_format = 'gemini:chat'
+              )
+            """))
+
+    # Best-effort normalization for already-existing Antigravity gemini:chat endpoints.
+    conn.execute(text("""
+            UPDATE provider_endpoints pe
+            SET
+                api_family = 'gemini',
+                endpoint_kind = 'chat'
+            WHERE pe.api_format = 'gemini:chat'
+              AND pe.provider_id IN (
+                SELECT id FROM providers WHERE lower(provider_type) = 'antigravity'
+              )
+            """))
+
+    # --- provider_api_keys.api_formats (JSON array) ---
+    # Replace "gemini:cli" with "gemini:chat" in the JSON array for Antigravity keys.
+    # Uses text-level replace on the serialized JSON — safe because the value is a
+    # simple string with no special characters that could cause ambiguous replacements.
+    conn.execute(text("""
+            UPDATE provider_api_keys pak
+            SET api_formats = replace(pak.api_formats::text, '"gemini:cli"', '"gemini:chat"')::json
+            WHERE pak.provider_id IN (
+                SELECT id FROM providers WHERE lower(provider_type) = 'antigravity'
+            )
+              AND pak.api_formats IS NOT NULL
+              AND pak.api_formats::text LIKE '%"gemini:cli"%'
+            """))
+
+
+def downgrade() -> None:
+    conn = op.get_bind()
+
+    # --- provider_endpoints ---
+    conn.execute(text("""
+            UPDATE provider_endpoints pe
+            SET
+                api_format = 'gemini:cli',
+                api_family = 'gemini',
+                endpoint_kind = 'cli'
+            WHERE pe.api_format = 'gemini:chat'
+              AND pe.provider_id IN (
+                SELECT id FROM providers WHERE lower(provider_type) = 'antigravity'
+              )
+              AND NOT EXISTS (
+                SELECT 1 FROM provider_endpoints pe2
+                WHERE pe2.provider_id = pe.provider_id
+                  AND pe2.api_format = 'gemini:cli'
+              )
+            """))
+
+    # Best-effort normalization for already-existing Antigravity gemini:cli endpoints.
+    conn.execute(text("""
+            UPDATE provider_endpoints pe
+            SET
+                api_family = 'gemini',
+                endpoint_kind = 'cli'
+            WHERE pe.api_format = 'gemini:cli'
+              AND pe.provider_id IN (
+                SELECT id FROM providers WHERE lower(provider_type) = 'antigravity'
+              )
+            """))
+
+    # --- provider_api_keys.api_formats (JSON array) ---
+    conn.execute(text("""
+            UPDATE provider_api_keys pak
+            SET api_formats = replace(pak.api_formats::text, '"gemini:chat"', '"gemini:cli"')::json
+            WHERE pak.provider_id IN (
+                SELECT id FROM providers WHERE lower(provider_type) = 'antigravity'
+            )
+              AND pak.api_formats IS NOT NULL
+              AND pak.api_formats::text LIKE '%"gemini:chat"%'
+            """))

frontend/src/api/dashboard.ts

@@ -161,6 +161,7 @@ export interface RequestDetail {
   request_type: string
   is_stream: boolean
   status_code: number
+  status?: string  // pending, streaming, completed, failed, cancelled
   error_message?: string
   response_time_ms: number
   created_at: string

frontend/src/features/providers/components/AntigravityQuotaDialog.vue

@@ -184,7 +184,7 @@ async function handleTestModel(modelName: string) {
       provider_id: props.providerId,
       model_name: modelName,
       api_key_id: props.keyId,
-      api_format: 'gemini:cli',
+      api_format: 'gemini:chat',
       message: 'hello',
     })
 

frontend/src/features/usage/components/HorizontalRequestTimeline.vue

@@ -642,19 +642,28 @@ const navigateGroup = (direction: number) => {
 }
 
 // 加载请求追踪数据
-const loadTrace = async () => {
+const isSilentRefresh = ref(false)
+const loadTrace = async (silent = false) => {
   if (!props.requestId) return
 
-  loading.value = true
+  isSilentRefresh.value = silent
+
+  if (!silent) {
+    loading.value = true
+  }
   error.value = null
 
   try {
     trace.value = await requestTraceApi.getRequestTrace(props.requestId)
   } catch (err: any) {
-    error.value = err.response?.data?.detail || err.message || '加载失败'
+    if (!silent) {
+      error.value = err.response?.data?.detail || err.message || '加载失败'
+    }
     log.error('加载请求追踪失败:', err)
   } finally {
-    loading.value = false
+    if (!silent) {
+      loading.value = false
+    }
   }
 }
 
@@ -662,6 +671,12 @@ const loadTrace = async () => {
 watch(groupedTimeline, (newGroups) => {
   if (!newGroups || newGroups.length === 0) return
 
+  // 静默刷新时不重置选中状态
+  if (isSilentRefresh.value) {
+    isSilentRefresh.value = false
+    return
+  }
+
   // 查找成功的组
   const successIdx = newGroups.findIndex(g => g.primaryStatus === 'success')
   if (successIdx >= 0) {
@@ -712,6 +727,8 @@ watch(() => props.requestId, () => {
   loadTrace()
 }, { immediate: true })
 
+defineExpose({ refresh: () => loadTrace(true) })
+
 // 格式化时间（详细）
 const formatTime = (dateStr: string) => {
   const date = new Date(dateStr)

frontend/src/features/usage/components/RequestDetailDrawer.vue

@@ -66,13 +66,13 @@
                   variant="ghost"
                   size="icon"
                   class="h-8 w-8"
-                  :disabled="loading"
+                  :disabled="loading && !autoRefreshing"
-                  title="刷新"
+                  :title="autoRefreshing ? '停止自动刷新' : '刷新'"
                   @click="refreshDetail"
                 >
                   <RefreshCw
                     class="w-4 h-4"
-                    :class="{ 'animate-spin': loading }"
+                    :class="{ 'animate-spin': loading || autoRefreshing, 'text-primary': autoRefreshing }"
                   />
                 </Button>
                 <Button
@@ -275,25 +275,27 @@
                     </div>
                   </div>
 
-                  <!-- ========== 3. 按次计费（独立隔离） ========== -->
+                  <!-- ========== 3. 按次计费 ========== -->
                   <div
                     v-if="perRequestCost > 0 && !detail.video_billing"
-                    class="rounded-lg p-3 bg-amber-500/5 border border-amber-500/30 mb-3"
+                    class="space-y-2 mb-3"
                   >
-                    <div class="flex items-center justify-between text-xs mb-2">
+                    <div class="flex items-center justify-between text-xs">
-                      <span class="font-medium text-amber-600 dark:text-amber-400">按次计费</span>
+                      <span class="font-medium text-foreground">按次计费</span>
-                      <span
-                        v-if="detail.price_per_request"
-                        class="text-muted-foreground"
-                      >
-                        单价 ${{ detail.price_per_request.toFixed(6) }}/次
-                      </span>
                     </div>
-                    <div class="flex items-center">
+                    <div class="rounded-lg p-3 bg-primary/5 border border-primary/30 space-y-2">
-                      <div class="flex items-center flex-1">
+                      <div
-                        <span class="text-xs text-muted-foreground w-[56px]">请求次数</span>
+                        v-if="detail.price_per_request"
-                        <span class="text-sm font-semibold font-mono flex-1 text-center">1</span>
+                        class="flex items-center justify-end text-xs"
-                        <span class="text-xs font-mono font-medium">${{ perRequestCost.toFixed(6) }}</span>
+                      >
+                        <span class="text-muted-foreground">${{ detail.price_per_request.toFixed(6) }}/次</span>
+                      </div>
+                      <div class="flex items-center">
+                        <div class="flex items-center flex-1">
+                          <span class="text-xs text-muted-foreground w-[56px]">请求次数</span>
+                          <span class="text-sm font-semibold font-mono flex-1 text-center">1</span>
+                          <span class="text-xs font-mono font-medium">${{ perRequestCost.toFixed(6) }}</span>
+                        </div>
                       </div>
                     </div>
                   </div>
@@ -361,6 +363,7 @@
               <!-- 请求链路追踪卡片 -->
               <div v-if="detail.request_id || detail.id">
                 <HorizontalRequestTimeline
+                  ref="timelineRef"
                   :request-id="detail.request_id || detail.id"
                   :override-status-code="detail.status_code"
                 />
@@ -590,7 +593,7 @@
 </template>
 
 <script setup lang="ts">
-import { ref, watch, computed } from 'vue'
+import { ref, watch, computed, onBeforeUnmount } from 'vue'
 import Button from '@/components/ui/button.vue'
 import { useEscapeKey } from '@/composables/useEscapeKey'
 import { useClipboard } from '@/composables/useClipboard'
@@ -631,6 +634,7 @@ const emit = defineEmits<{
 const loading = ref(false)
 const error = ref<string | null>(null)
 const detail = ref<RequestDetail | null>(null)
+const timelineRef = ref<InstanceType<typeof HorizontalRequestTimeline> | null>(null)
 const activeTab = ref('request-body')
 const copiedStates = ref<Record<string, boolean>>({})
 const viewMode = ref<'compare' | 'formatted' | 'raw'>('formatted')
@@ -645,6 +649,8 @@ const historicalPricing = ref<{
   cache_read_price: string
   request_price: string
 } | null>(null)
+const autoRefreshTimer = ref<ReturnType<typeof setInterval> | null>(null)
+const autoRefreshing = ref(false)
 
 // 监听标签页切换
 watch(activeTab, (newTab) => {
@@ -870,26 +876,38 @@ watch(() => props.requestId, async (newId) => {
 watch(() => props.isOpen, async (isOpen) => {
   if (isOpen && props.requestId) {
     await loadDetail(props.requestId)
+  } else if (!isOpen) {
+    stopAutoRefresh()
   }
 })
 
-async function loadDetail(id: string) {
+async function loadDetail(id: string, silent = false) {
-  loading.value = true
+  if (!silent) {
+    loading.value = true
+    historicalPricing.value = null
+  }
   error.value = null
-  historicalPricing.value = null
   try {
     detail.value = await dashboardApi.getRequestDetail(id)
 
-    // 默认显示有内容的第一个可见 tab
+    // 首次加载时选择默认 tab
-    const visibleTabNames = visibleTabs.value.map(t => t.name)
+    if (!silent) {
-    if (detail.value.request_body && visibleTabNames.includes('request-body')) {
+      const visibleTabNames = visibleTabs.value.map(t => t.name)
-      activeTab.value = 'request-body'
+      if (detail.value.request_body && visibleTabNames.includes('request-body')) {
-    } else if (detail.value.response_body && visibleTabNames.includes('response-body')) {
+        activeTab.value = 'request-body'
-      activeTab.value = 'response-body'
+      } else if (detail.value.response_body && visibleTabNames.includes('response-body')) {
-    } else if (visibleTabNames.length > 0) {
+        activeTab.value = 'response-body'
-      activeTab.value = visibleTabNames[0]
+      } else if (visibleTabNames.length > 0) {
+        activeTab.value = visibleTabNames[0]
+      }
     }
 
+    // 根据数据可用性自动选择请求头数据源
+    // provider_request_headers 在 streaming 完成后才写入，pending/streaming 期间为空
+    const hasProviderReqHeaders = detail.value.provider_request_headers &&
+      Object.keys(detail.value.provider_request_headers).length > 0
+    dataSource.value = hasProviderReqHeaders ? 'provider' : 'client'
+
     // 使用请求记录中保存的历史价格
     if (detail.value.input_price_per_1m || detail.value.output_price_per_1m || detail.value.price_per_request) {
       historicalPricing.value = {
@@ -900,24 +918,81 @@ async function loadDetail(id: string) {
         request_price: detail.value.price_per_request ? detail.value.price_per_request.toFixed(4) : 'N/A'
       }
     }
+
+    // 静默刷新时同步刷新链路追踪
+    if (silent) {
+      timelineRef.value?.refresh()
+    }
   } catch (err) {
     log.error('Failed to load request detail:', err)
-    error.value = '加载请求详情失败'
+    if (!silent) {
+      error.value = '加载请求详情失败'
+    }
   } finally {
-    loading.value = false
+    if (!silent) {
+      loading.value = false
+    }
   }
 }
 
 function handleClose() {
+  stopAutoRefresh()
   emit('close')
 }
 
-async function refreshDetail() {
+function isRequestCompleted(): boolean {
-  if (props.requestId) {
+  if (!detail.value?.status) return true
-    await loadDetail(props.requestId)
+  return !['pending', 'streaming'].includes(detail.value.status)
-  }
 }
 
+function stopAutoRefresh() {
+  if (autoRefreshTimer.value) {
+    clearInterval(autoRefreshTimer.value)
+    autoRefreshTimer.value = null
+  }
+  autoRefreshing.value = false
+}
+
+async function refreshDetail() {
+  if (!props.requestId) return
+
+  // 已完成：单次静默刷新
+  if (isRequestCompleted()) {
+    await loadDetail(props.requestId, true)
+    return
+  }
+
+  // 未完成：如果已在自动刷新则停止，否则启动
+  if (autoRefreshing.value) {
+    stopAutoRefresh()
+    return
+  }
+
+  autoRefreshing.value = true
+  await loadDetail(props.requestId, true)
+
+  // 加载后可能已经完成了
+  if (isRequestCompleted()) {
+    stopAutoRefresh()
+    return
+  }
+
+  autoRefreshTimer.value = setInterval(async () => {
+    if (!props.requestId || !props.isOpen) {
+      stopAutoRefresh()
+      return
+    }
+    await loadDetail(props.requestId, true)
+    if (isRequestCompleted()) {
+      stopAutoRefresh()
+    }
+  }, 1000)
+}
+
+onBeforeUnmount(() => {
+  stopAutoRefresh()
+})
+
 function formatDateTime(dateStr: string | null | undefined): string {
   if (!dateStr) return 'N/A'
   const date = new Date(dateStr)

frontend/src/features/usage/components/UsageRecordsTable.vue

@@ -193,7 +193,7 @@
           <div class="flex flex-col items-end flex-shrink-0">
             <span class="text-xs text-primary font-medium">{{ formatCurrency(record.cost || 0) }}</span>
             <span
-              v-if="showActualCost && record.actual_cost !== undefined"
+              v-if="showActualCost && record.actual_cost !== undefined && record.rate_multiplier && record.rate_multiplier !== 1.0"
               class="text-[10px] text-muted-foreground"
             >{{ formatCurrency(record.actual_cost) }}</span>
           </div>
@@ -562,7 +562,7 @@
             <div class="flex flex-col items-end text-xs gap-0.5">
               <span class="text-primary font-medium">{{ formatCurrency(record.cost || 0) }}</span>
               <span
-                v-if="showActualCost && record.actual_cost !== undefined"
+                v-if="showActualCost && record.actual_cost !== undefined && record.rate_multiplier && record.rate_multiplier !== 1.0"
                 class="text-muted-foreground"
               >
                 {{ formatCurrency(record.actual_cost) }}

src/api/admin/endpoints/keys.py

@@ -1009,7 +1009,7 @@ class AdminRefreshProviderQuotaAdapter(AdminApiAdapter):
 
         # 获取端点：
         # - Codex: openai:cli
-        # - Antigravity: gemini:cli（用于触发 oauth 刷新 + 提供 auth_config.project_id）
+        # - Antigravity: gemini:chat（用于触发 oauth 刷新 + 提供 auth_config.project_id）
         endpoint = None
         if provider_type == ProviderType.CODEX:
             for ep in provider.endpoints:
@@ -1019,12 +1019,16 @@ class AdminRefreshProviderQuotaAdapter(AdminApiAdapter):
             if not endpoint:
                 raise InvalidRequestException("找不到有效的 openai:cli 端点")
         else:
-            for ep in provider.endpoints:
+            # Prefer the new signature, but keep backward-compat with existing DB rows.
-                if ep.api_format == "gemini:cli" and ep.is_active:
+            for sig in ("gemini:chat", "gemini:cli"):
-                    endpoint = ep
+                for ep in provider.endpoints:
+                    if ep.api_format == sig and ep.is_active:
+                        endpoint = ep
+                        break
+                if endpoint is not None:
                     break
             if not endpoint:
-                raise InvalidRequestException("找不到有效的 gemini:cli 端点")
+                raise InvalidRequestException("找不到有效的 gemini:chat/gemini:cli 端点")
 
         results: list[dict] = []
         success_count = 0

src/api/admin/provider_query.py

@@ -70,7 +70,7 @@ async def _resolve_key_auth(
     auth_config: dict[str, Any] | None = None
 
     if auth_type == "oauth":
-        endpoint_api_format = "gemini:cli" if provider_type == ProviderType.ANTIGRAVITY else None
+        endpoint_api_format = "gemini:chat" if provider_type == ProviderType.ANTIGRAVITY else None
         try:
             resolved = await resolve_oauth_access_token(
                 key_id=str(api_key.id),

src/api/admin/usage/routes.py

@@ -1181,6 +1181,7 @@ class AdminUsageDetailAdapter(AdminApiAdapter):
             "is_stream": usage_record.is_stream,
             "status_code": usage_record.status_code,
             "error_message": usage_record.error_message,
+            "status": usage_record.status,
             "response_time_ms": usage_record.response_time_ms,
             "first_byte_time_ms": usage_record.first_byte_time_ms,  # 首字时间 (TTFB)
             "created_at": usage_record.created_at.isoformat() if usage_record.created_at else None,

src/api/handlers/base/base_handler.py

@@ -571,7 +571,7 @@ class BaseMessageHandler:
         api_format = ctx.api_format
         # 格式转换追踪
         endpoint_api_format = ctx.provider_api_format or None
-        has_format_conversion = ctx.needs_conversion
+        has_format_conversion = ctx.has_format_conversion
 
         # 如果 provider 为空，记录警告（不应该发生，但用于调试）
         if not provider:

src/api/handlers/base/chat_handler_base.py

@@ -40,7 +40,7 @@ from src.api.handlers.base.base_handler import (
 from src.api.handlers.base.parsers import get_parser_for_format
 from src.api.handlers.base.request_builder import PassthroughRequestBuilder, get_provider_auth
 from src.api.handlers.base.response_parser import ResponseParser
-from src.api.handlers.base.stream_context import StreamContext
+from src.api.handlers.base.stream_context import StreamContext, is_format_converted
 from src.api.handlers.base.stream_processor import StreamProcessor
 from src.api.handlers.base.stream_telemetry import StreamTelemetryRecorder
 from src.api.handlers.base.upstream_stream_bridge import (
@@ -1322,7 +1322,7 @@ class ChatHandlerBase(BaseMessageHandler, ABC):
             client_response_headers=client_response_headers,
             # 格式转换追踪
             endpoint_api_format=ctx.provider_api_format or None,
-            has_format_conversion=ctx.needs_conversion,
+            has_format_conversion=ctx.has_format_conversion,
             target_model=ctx.mapped_model,
         )
 
@@ -1367,7 +1367,7 @@ class ChatHandlerBase(BaseMessageHandler, ABC):
         provider_request_body: dict[str, Any] | None = None
         provider_api_format_for_error: str | None = None
         client_api_format_for_error: str | None = None
-        needs_conversion_for_error: bool = False
+        needs_conversion_for_error: bool = False  # 用于构建错误 payload（含 envelope rewrite）
         provider_id: str | None = None  # Provider ID（用于失败记录）
         endpoint_id: str | None = None  # Endpoint ID（用于失败记录）
         key_id: str | None = None  # Key ID（用于失败记录）
@@ -1795,7 +1795,9 @@ class ChatHandlerBase(BaseMessageHandler, ABC):
                 api_format=api_format,
                 # 格式转换追踪
                 endpoint_api_format=provider_api_format_for_error or None,
-                has_format_conversion=needs_conversion_for_error,
+                has_format_conversion=is_format_converted(
+                    provider_api_format_for_error, client_api_format_for_error
+                ),
                 provider_id=provider_id,
                 provider_endpoint_id=endpoint_id,
                 provider_api_key_id=key_id,
@@ -1865,7 +1867,9 @@ class ChatHandlerBase(BaseMessageHandler, ABC):
                 client_response_headers={"content-type": "application/json"},
                 # 格式转换追踪
                 endpoint_api_format=provider_api_format_for_error or None,
-                has_format_conversion=needs_conversion_for_error,
+                has_format_conversion=is_format_converted(
+                    provider_api_format_for_error, client_api_format_for_error
+                ),
                 target_model=mapped_model_result,
                 request_metadata=request_metadata,
             )
@@ -1916,7 +1920,9 @@ class ChatHandlerBase(BaseMessageHandler, ABC):
                 client_response_headers={"content-type": "application/json"},
                 # 格式转换追踪
                 endpoint_api_format=provider_api_format_for_error or None,
-                has_format_conversion=needs_conversion_for_error,
+                has_format_conversion=is_format_converted(
+                    provider_api_format_for_error, client_api_format_for_error
+                ),
                 # 模型映射信息
                 target_model=mapped_model_result,
                 request_metadata=request_metadata,

src/api/handlers/base/cli_handler_base.py

@@ -43,7 +43,7 @@ from src.api.handlers.base.request_builder import PassthroughRequestBuilder, get
 from src.api.handlers.base.response_parser import (
     ResponseParser,
 )
-from src.api.handlers.base.stream_context import StreamContext
+from src.api.handlers.base.stream_context import StreamContext, is_format_converted
 from src.api.handlers.base.upstream_stream_bridge import (
     aggregate_upstream_stream_to_internal_response,
 )
@@ -2459,7 +2459,7 @@ class CliMessageHandlerBase(BaseMessageHandler):
                             response_headers=ctx.response_headers,
                             client_response_headers=client_response_headers,
                             endpoint_api_format=ctx.provider_api_format or None,
-                            has_format_conversion=ctx.needs_conversion,
+                            has_format_conversion=ctx.has_format_conversion,
                             target_model=ctx.mapped_model,
                             request_metadata=request_metadata,
                         )
@@ -2492,7 +2492,7 @@ class CliMessageHandlerBase(BaseMessageHandler):
                             client_response_headers=client_response_headers,
                             # 格式转换追踪
                             endpoint_api_format=ctx.provider_api_format or None,
-                            has_format_conversion=ctx.needs_conversion,
+                            has_format_conversion=ctx.has_format_conversion,
                             # 模型映射信息
                             target_model=ctx.mapped_model,
                             request_metadata=request_metadata,
@@ -2552,7 +2552,7 @@ class CliMessageHandlerBase(BaseMessageHandler):
                         api_format=ctx.api_format,
                         # 格式转换追踪
                         endpoint_api_format=ctx.provider_api_format or None,
-                        has_format_conversion=ctx.needs_conversion,
+                        has_format_conversion=ctx.has_format_conversion,
                         # Provider 侧追踪信息（用于记录真实成本）
                         provider_id=ctx.provider_id,
                         provider_endpoint_id=ctx.endpoint_id,
@@ -2709,7 +2709,7 @@ class CliMessageHandlerBase(BaseMessageHandler):
             client_response_headers=client_response_headers,
             # 格式转换追踪
             endpoint_api_format=ctx.provider_api_format or None,
-            has_format_conversion=ctx.needs_conversion,
+            has_format_conversion=ctx.has_format_conversion,
             # 模型映射信息
             target_model=ctx.mapped_model,
             request_metadata=request_metadata,
@@ -3170,7 +3170,7 @@ class CliMessageHandlerBase(BaseMessageHandler):
                 api_format=api_format,
                 # 格式转换追踪
                 endpoint_api_format=provider_api_format or None,
-                has_format_conversion=needs_conversion,
+                has_format_conversion=is_format_converted(provider_api_format, str(api_format)),
                 # Provider 侧追踪信息（用于记录真实成本）
                 provider_id=provider_id,
                 provider_endpoint_id=endpoint_id,
@@ -3249,7 +3249,7 @@ class CliMessageHandlerBase(BaseMessageHandler):
                 client_response_headers={"content-type": "application/json"},
                 # 格式转换追踪
                 endpoint_api_format=provider_api_format or None,
-                has_format_conversion=needs_conversion,
+                has_format_conversion=is_format_converted(provider_api_format, str(api_format)),
                 # 模型映射信息
                 target_model=mapped_model_result,
                 request_metadata=request_metadata,
@@ -3364,7 +3364,7 @@ class CliMessageHandlerBase(BaseMessageHandler):
                         first_byte_time_ms=ctx.first_byte_time_ms,
                         api_format=ctx.api_format,
                         endpoint_api_format=ctx.provider_api_format or None,
-                        has_format_conversion=ctx.needs_conversion,
+                        has_format_conversion=ctx.has_format_conversion,
                     )
                 except Exception as e:
                     logger.warning(f"[{self.request_id}] 同步更新 streaming 状态失败: {e}")

src/api/handlers/base/stream_context.py

@@ -18,6 +18,18 @@ if TYPE_CHECKING:
     from src.core.api_format.conversion.stream_state import StreamState
 
 
+def is_format_converted(
+    provider_api_format: str | None,
+    client_api_format: str | None,
+) -> bool:
+    """client 与 provider 的 api_format 是否真正不同（用于 usage 展示层）"""
+    return bool(
+        provider_api_format
+        and client_api_format
+        and provider_api_format.strip().lower() != client_api_format.strip().lower()
+    )
+
+
 @dataclass
 class StreamContext:
     """
@@ -234,6 +246,15 @@ class StreamContext:
         if self.first_byte_time_ms is None:
             self.first_byte_time_ms = int((time.time() - start_time) * 1000)
 
+    @property
+    def has_format_conversion(self) -> bool:
+        """是否发生了真正的格式转换（client 和 provider 的 api_format 不同）
+
+        区别于 needs_conversion：后者包含 envelope rewrite（如 Antigravity v1internal），
+        不代表客户端与上游的数据格式真正不同。此属性用于 usage 展示层。
+        """
+        return is_format_converted(self.provider_api_format, self.client_api_format)
+
     def is_success(self) -> bool:
         """检查请求是否成功"""
         return self.status_code < 400

src/api/handlers/base/stream_processor.py

@@ -108,6 +108,115 @@ class StreamProcessor:
                 pass
         return self.default_parser
 
+    @staticmethod
+    def _maybe_mark_gemini_completion(ctx: StreamContext, data: dict[str, Any]) -> None:
+        """Gemini: mark completion based on candidates[].finishReason."""
+        candidates = data.get("candidates")
+        if not isinstance(candidates, list) or not candidates:
+            return
+        for candidate in candidates:
+            if not isinstance(candidate, dict):
+                continue
+            finish_reason = candidate.get("finishReason")
+            if finish_reason is None:
+                continue
+            # UNSPECIFIED is the only clear "not done" sentinel across Gemini variants.
+            if str(finish_reason) != "FINISH_REASON_UNSPECIFIED":
+                ctx.has_completion = True
+                return
+
+    @staticmethod
+    def _extract_antigravity_usage_from_gemini_event(data: dict[str, Any]) -> dict[str, int] | None:
+        """Antigravity: lenient Gemini usage extraction (totalTokenCount may be missing)."""
+        usage_metadata = data.get("usageMetadata", {})
+        if not isinstance(usage_metadata, dict) or not usage_metadata:
+            return None
+
+        def _as_int(v: Any) -> int:
+            try:
+                return int(v or 0)
+            except Exception:
+                return 0
+
+        prompt = _as_int(usage_metadata.get("promptTokenCount"))
+        cached = _as_int(usage_metadata.get("cachedContentTokenCount"))
+        candidates = _as_int(usage_metadata.get("candidatesTokenCount"))
+        thoughts = _as_int(usage_metadata.get("thoughtsTokenCount"))
+
+        # Align with Gemini billing convention: input_tokens includes cached content.
+        return {
+            "input_tokens": max(0, prompt),
+            "output_tokens": max(0, candidates + thoughts),
+            "cache_creation_tokens": 0,
+            "cache_read_tokens": max(0, cached),
+        }
+
+    def _unwrap_provider_envelope(self, ctx: StreamContext, data: dict[str, Any]) -> dict[str, Any]:
+        behavior = get_provider_behavior(
+            provider_type=str(getattr(ctx, "provider_type", "") or ""),
+            endpoint_sig=str(getattr(ctx, "provider_api_format", "") or ""),
+        )
+        envelope = behavior.envelope
+        if not envelope:
+            return data
+
+        try:
+            unwrapped = envelope.unwrap_response(data)
+            envelope.postprocess_unwrapped_response(
+                model=str(getattr(ctx, "model", "") or ""),
+                data=unwrapped,
+            )
+            return unwrapped if isinstance(unwrapped, dict) else data
+        except Exception:
+            return data
+
+    def _update_ctx_from_provider_event(
+        self,
+        ctx: StreamContext,
+        data: dict[str, Any],
+        *,
+        already_unwrapped: bool = False,
+    ) -> None:
+        # Unwrap provider-specific envelopes (e.g. Antigravity v1internal wrapper)
+        if not already_unwrapped:
+            data = self._unwrap_provider_envelope(ctx, data)
+
+        parser = self.get_parser_for_provider(ctx)
+
+        # Provider usage extraction (best-effort)
+        provider_type = str(getattr(ctx, "provider_type", "") or "").lower()
+        provider_format = str(getattr(ctx, "provider_api_format", "") or "").strip().lower()
+
+        usage: dict[str, int] | None = None
+        if provider_type == "antigravity" and provider_format.startswith("gemini:"):
+            usage = self._extract_antigravity_usage_from_gemini_event(data)
+        if usage is None:
+            try:
+                usage = parser.extract_usage_from_response(data)
+            except Exception:
+                usage = None
+
+        if usage:
+            ctx.update_usage(
+                input_tokens=usage.get("input_tokens"),
+                output_tokens=usage.get("output_tokens"),
+                cached_tokens=usage.get("cache_read_tokens"),
+                cache_creation_tokens=usage.get("cache_creation_tokens"),
+            )
+
+        # Provider completion detection (Gemini doesn't emit response.completed).
+        if provider_format.startswith("gemini:"):
+            self._maybe_mark_gemini_completion(ctx, data)
+
+        # Provider text extraction (optional)
+        if self.collect_text:
+            try:
+                text = parser.extract_text_content(data)
+            except Exception:
+                text = ""
+            if text:
+                ctx.append_text(text)
+
     def handle_sse_event(
         self,
         ctx: StreamContext,
@@ -115,6 +224,7 @@ class StreamProcessor:
         data_str: str,
         *,
         skip_record: bool = False,
+        skip_ctx_update: bool = False,
     ) -> None:
         """
         处理单个 SSE 事件
@@ -126,6 +236,7 @@ class StreamProcessor:
             event_name: 事件名称
             data_str: 事件数据字符串
             skip_record: 是否跳过记录到 parsed_chunks（当需要格式转换时应为 True）
+            skip_ctx_update: 跳过 usage/completion/text 提取（由调用方统一处理时使用）
         """
         if not data_str:
             return
@@ -142,31 +253,17 @@ class StreamProcessor:
         if not isinstance(data, dict):
             return
 
+        # Update usage/completion/text from provider event (envelope-aware).
+        # 在 needs_conversion 正常流中由 _emit_converted_line 统一处理，此处跳过以避免重复。
+        if not skip_ctx_update:
+            self._update_ctx_from_provider_event(ctx, data)
+
         # 统计数据事件数量（当需要格式转换时跳过，由 _emit_converted_line 统计/记录转换后的数据）
         if not skip_record:
             ctx.data_count += 1
             if ctx.record_parsed_chunks:
                 ctx.parsed_chunks.append(data)
 
-        # 根据 Provider 格式选择解析器
-        parser = self.get_parser_for_provider(ctx)
-
-        # 使用解析器提取 usage
-        usage = parser.extract_usage_from_response(data)
-        if usage:
-            ctx.update_usage(
-                input_tokens=usage.get("input_tokens"),
-                output_tokens=usage.get("output_tokens"),
-                cached_tokens=usage.get("cache_read_tokens"),
-                cache_creation_tokens=usage.get("cache_creation_tokens"),
-            )
-
-        # 提取文本
-        if self.collect_text:
-            text = parser.extract_text_content(data)
-            if text:
-                ctx.append_text(text)
-
         # 检查完成
         event_type = event_name or data.get("type", "")
         if event_type in ("response.completed", "message_stop"):
@@ -476,14 +573,31 @@ class StreamProcessor:
                     self.on_streaming_start()
                     streaming_started = True
 
-            def _process_line_with_perf(line: str, *, skip_record: bool = False) -> None:
+            def _process_line_with_perf(
+                line: str,
+                *,
+                skip_record: bool = False,
+                skip_ctx_update: bool = False,
+            ) -> None:
                 nonlocal parse_time
                 if perf_capture:
                     t0 = time.perf_counter()
-                    self._process_line(ctx, sse_parser, line, skip_record=skip_record)
+                    self._process_line(
+                        ctx,
+                        sse_parser,
+                        line,
+                        skip_record=skip_record,
+                        skip_ctx_update=skip_ctx_update,
+                    )
                     parse_time += time.perf_counter() - t0
                     return
-                self._process_line(ctx, sse_parser, line, skip_record=skip_record)
+                self._process_line(
+                    ctx,
+                    sse_parser,
+                    line,
+                    skip_record=skip_record,
+                    skip_ctx_update=skip_ctx_update,
+                )
 
             def _build_stream_error_payload(message: str) -> dict:
                 if client_family == "openai":
@@ -589,6 +703,14 @@ class StreamProcessor:
                             data=data_obj,
                         )
 
+                    # Update usage/completion/text based on the unwrapped provider event.
+                    if isinstance(data_obj, dict):
+                        self._update_ctx_from_provider_event(
+                            ctx,
+                            data_obj,
+                            already_unwrapped=True,
+                        )
+
                     try:
                         converted_events = registry.convert_stream_chunk(
                             data_obj,
@@ -648,7 +770,9 @@ class StreamProcessor:
 
                             if line:
                                 # 需要格式转换时，跳过记录原始数据（由 _emit_converted_line 记录转换后的数据）
-                                _process_line_with_perf(line, skip_record=True)
+                                _process_line_with_perf(
+                                    line, skip_record=True, skip_ctx_update=True
+                                )
                             normalized_line = line.rstrip("\r\n") if line else ""
                             out_chunks = _emit_converted_line(normalized_line)
                             if not out_chunks:
@@ -878,6 +1002,7 @@ class StreamProcessor:
         line: str,
         *,
         skip_record: bool = False,
+        skip_ctx_update: bool = False,
     ) -> None:
         """
         处理单行数据
@@ -887,6 +1012,7 @@ class StreamProcessor:
             sse_parser: SSE 解析器
             line: 原始行数据
             skip_record: 是否跳过记录到 parsed_chunks（当需要格式转换时应为 True）
+            skip_ctx_update: 跳过 usage/completion/text 提取（由调用方统一处理时使用）
         """
         # SSEEventParser 以"去掉换行符"的单行文本作为输入；这里统一剔除 CR/LF，
         # 避免把空行误判成 "\n" 并导致事件边界解析错误。
@@ -898,7 +1024,11 @@ class StreamProcessor:
 
         for event in events:
             self.handle_sse_event(
-                ctx, event.get("event"), event.get("data") or "", skip_record=skip_record
+                ctx,
+                event.get("event"),
+                event.get("data") or "",
+                skip_record=skip_record,
+                skip_ctx_update=skip_ctx_update,
             )
 
     async def create_monitored_stream(

src/api/handlers/base/stream_telemetry.py

@@ -227,7 +227,7 @@ class StreamTelemetryRecorder:
             request_type="chat",
             metadata=metadata,
             endpoint_api_format=ctx.provider_api_format,
-            has_format_conversion=ctx.needs_conversion,
+            has_format_conversion=ctx.has_format_conversion,
         )
 
         logger.debug(f"{self.format_id} 流式响应完成")
@@ -273,7 +273,7 @@ class StreamTelemetryRecorder:
             request_type="chat",
             metadata=metadata,
             endpoint_api_format=ctx.provider_api_format,
-            has_format_conversion=ctx.needs_conversion,
+            has_format_conversion=ctx.has_format_conversion,
         )
 
         logger.debug(f"{self.format_id} 流式响应中断")
@@ -320,7 +320,7 @@ class StreamTelemetryRecorder:
             request_type="chat",
             metadata=metadata,
             endpoint_api_format=ctx.provider_api_format,
-            has_format_conversion=ctx.needs_conversion,
+            has_format_conversion=ctx.has_format_conversion,
         )
 
         logger.debug(f"{self.format_id} 流式响应被客户端取消")

src/core/provider_templates/fixed_providers.py

@@ -106,7 +106,7 @@ FIXED_PROVIDERS: dict[ProviderType, FixedProviderTemplate] = {
         provider_type=ProviderType.ANTIGRAVITY,
         display_name="Antigravity",
         api_base_url=ANTIGRAVITY_PROD_URL,
-        endpoint_signatures=["gemini:cli"],
+        endpoint_signatures=["gemini:chat"],
         oauth=FixedProviderOAuth(
             authorize_url="https://accounts.google.com/o/oauth2/v2/auth",
             token_url="https://oauth2.googleapis.com/token",

src/services/model/fetch_scheduler.py

@@ -303,7 +303,9 @@ class ModelFetchScheduler:
             # Use request_builder's lazy refresh logic and persist refreshed token back to DB.
             # Endpoint signature is only used for tracing/debug; auth logic doesn't depend on it.
             endpoint_api_format = (
-                "gemini:cli" if prepared.provider_type.lower() == ProviderType.ANTIGRAVITY else None
+                "gemini:chat"
+                if prepared.provider_type.lower() == ProviderType.ANTIGRAVITY
+                else None
             )
             try:
                 resolved = await resolve_oauth_access_token(

src/services/provider/adapters/antigravity/envelope.py

@@ -1,6 +1,6 @@
 """Antigravity v1internal request/response envelope helpers.
 
-Antigravity reuses the `gemini:cli` endpoint signature but wraps the actual
+Antigravity reuses the `gemini:chat` endpoint signature but wraps the actual
 wire format:
 - Request: V1InternalRequest (top-level metadata + nested GeminiRequest)
 - Response: V1InternalResponse (top-level responseId + nested GeminiResponse)

src/services/provider/adapters/antigravity/plugin.py

@@ -249,7 +249,7 @@ async def fetch_models_antigravity(
                 "id": model_id,
                 "owned_by": "antigravity",
                 "display_name": display_name,
-                "api_format": "gemini:cli",
+                "api_format": "gemini:chat",
             }
         )
 
@@ -324,10 +324,14 @@ def register_all() -> None:
     from src.services.provider.transport import register_transport_hook
 
     # Envelope
+    register_envelope("antigravity", "gemini:chat", antigravity_v1internal_envelope)
+    # Backward compat: allow existing endpoints that still use the old signature.
     register_envelope("antigravity", "gemini:cli", antigravity_v1internal_envelope)
     register_envelope("antigravity", "", antigravity_v1internal_envelope)
 
     # Transport
+    register_transport_hook("antigravity", "gemini:chat", build_antigravity_url)
+    # Backward compat: allow existing endpoints that still use the old signature.
     register_transport_hook("antigravity", "gemini:cli", build_antigravity_url)
 
     # Auth

src/utils/cache_decorator.py

@@ -106,7 +106,7 @@ def cache_result(
                 if cached:
                     try:
                         result = json.loads(cached)
-                        logger.debug(f"缓存命中: {cache_key}")
+                        logger.trace(f"缓存命中: {cache_key}")
                         return result
                     except json.JSONDecodeError as e:
                         logger.warning(f"缓存解析失败，删除损坏缓存: {cache_key}, 错误: {e}")
@@ -123,7 +123,7 @@ def cache_result(
                     await redis_client.setex(
                         cache_key, ttl, json.dumps(result, ensure_ascii=False, default=str)
                     )
-                    logger.debug(f"缓存已保存: {cache_key}, TTL: {ttl}s")
+                    logger.trace(f"缓存已保存: {cache_key}, TTL: {ttl}s")
                 except Exception as e:
                     logger.warning(f"保存缓存失败: {e}")
 

tests/api/handlers/base/test_antigravity_v1internal.py

@@ -62,7 +62,7 @@ def test_convert_sse_line_unwraps_before_convert_stream_chunk() -> None:
 
     ctx = StreamContext(model="gemini-2.0-flash", api_format="gemini:cli")
     ctx.provider_type = "antigravity"
-    ctx.provider_api_format = "gemini:cli"
+    ctx.provider_api_format = "gemini:chat"
     ctx.client_api_format = "gemini:cli"
 
     v1_line = (

tests/services/antigravity/test_thinking_signature_conversion.py

@@ -32,7 +32,7 @@ def test_antigravity_converts_claude_thinking_block_to_gemini_thought_part_prefe
     }
 
     registry = get_format_converter_registry()
-    out = registry.convert_request(req, "claude:chat", "gemini:cli", target_variant="antigravity")
+    out = registry.convert_request(req, "claude:chat", "gemini:chat", target_variant="antigravity")
 
     assert isinstance(out.get("contents"), list)
     model_turn = out["contents"][1]
@@ -63,7 +63,7 @@ def test_antigravity_uses_dummy_signature_for_gemini_when_missing() -> None:
     }
 
     registry = get_format_converter_registry()
-    out = registry.convert_request(req, "claude:chat", "gemini:cli", target_variant="antigravity")
+    out = registry.convert_request(req, "claude:chat", "gemini:chat", target_variant="antigravity")
 
     parts = out["contents"][1]["parts"]
     assert parts[0]["thought"] is True
@@ -91,7 +91,7 @@ def test_antigravity_drops_unsigned_thinking_for_non_gemini_models() -> None:
     }
 
     registry = get_format_converter_registry()
-    out = registry.convert_request(req, "claude:chat", "gemini:cli", target_variant="antigravity")
+    out = registry.convert_request(req, "claude:chat", "gemini:chat", target_variant="antigravity")
 
     parts = out["contents"][1]["parts"]
     assert all(p.get("thought") is not True for p in parts)
@@ -111,7 +111,7 @@ def test_antigravity_inserts_dummy_thought_for_last_assistant_when_thinking_enab
     }
 
     registry = get_format_converter_registry()
-    out = registry.convert_request(req, "claude:chat", "gemini:cli", target_variant="antigravity")
+    out = registry.convert_request(req, "claude:chat", "gemini:chat", target_variant="antigravity")
 
     parts = out["contents"][1]["parts"]
     assert parts[0]["thought"] is True

tests/services/test_provider_transport_antigravity.py

@@ -19,7 +19,7 @@ class _DummyEndpoint:
 def test_antigravity_uses_v1internal_path_and_sets_contextvar() -> None:
     endpoint = _DummyEndpoint(
         base_url="https://ignored.example.com",
-        api_format="gemini:cli",
+        api_format="gemini:chat",
         provider=SimpleNamespace(provider_type="antigravity"),
     )
 

tests/services/test_request_builder_oauth.py

@@ -20,7 +20,7 @@ async def test_get_provider_auth_oauth_returns_decrypted_auth_config() -> None:
         "project_id": "project-1",
     }
 
-    endpoint = SimpleNamespace(api_format="gemini:cli")
+    endpoint = SimpleNamespace(api_format="gemini:chat")
     key = SimpleNamespace(
         id="k1",
         auth_type="oauth",