feat: 添加 API 格式访问限制

扩展访问限制功能，支持 API Key 级别的 API 格式限制（OPENAI、CLAUDE、GEMINI）。

- AccessRestrictions 新增 allowed_api_formats 字段
- 新增 is_api_format_allowed() 方法检查格式权限
- models.py 添加 _filter_formats_by_restrictions() 函数过滤 API 格式
- 在所有模型列表和查询端点应用格式限制检查
- 添加 _build_empty_list_response() 统一空响应构建逻辑

LICENSE

@@ -5,12 +5,17 @@ Aether 非商业开源许可证
 特此授予任何获得本软件及其相关文档文件（以下简称"软件"）副本的人免费使用、
 复制、修改、合并、发布和分发本软件的权限，但须遵守以下条件：
 
-1. 仅限非商业用途
-   本软件不得用于商业目的。商业目的包括但不限于：
+1. 仅限非盈利用途
+   本软件不得用于盈利目的。盈利目的包括但不限于：
    - 出售本软件或任何衍生作品
    - 使用本软件提供付费服务
-   - 将本软件用于商业产品或服务
-   - 将本软件用于任何旨在获取商业利益或金钱报酬的活动
+   - 将本软件用于以盈利为目的的商业产品或服务
+
+   以下用途被明确允许：
+   - 个人学习和研究
+   - 教育机构的教学和研究
+   - 非盈利组织的内部使用
+   - 企业内部非盈利性质的使用（如内部工具、测试环境等）
 
 2. 署名要求
    上述版权声明和本许可声明应包含在本软件的所有副本或主要部分中。
@@ -22,7 +27,7 @@ Aether 非商业开源许可证
    您不得以不同的条款将本软件再许可给他人。
 
 5. 商业许可
-   如需商业使用，请联系版权持有人以获取单独的商业许可。
+   如需将本软件用于盈利目的，请联系版权持有人以获取单独的商业许可。
 
 本软件按"原样"提供，不提供任何明示或暗示的保证，包括但不限于对适销性、
 特定用途适用性和非侵权性的保证。在任何情况下，作者或版权持有人均不对任何

README.md

@@ -143,7 +143,7 @@ cd frontend && npm install && npm run dev
 - **模型级别**: 在模型管理中针对指定模型开启 1H缓存策略
 - **密钥级别**: 在密钥管理中针对指定密钥使用 1H缓存策略
 
-> **注意**: 若对密钥设置强制 1H缓存, 则该密钥只能调用支持 1H缓存的模型
+> **注意**: 若对密钥设置强制 1H缓存, 则该密钥只能使用支持 1H缓存的模型, 匹配提供商Key, 将会导致这个Key无法同时用于Claude Code、Codex、GeminiCLI, 因为更推荐使用模型开启1H缓存.
 
 ### Q: 如何配置负载均衡？
 
@@ -162,4 +162,16 @@ cd frontend && npm install && npm run dev
 
 ## 许可证
 
-本项目采用 [Aether 非商业开源许可证](LICENSE)。
+本项目采用 [Aether 非商业开源许可证](LICENSE)。允许个人学习、教育研究、非盈利组织及企业内部非盈利性质的使用；禁止用于盈利目的。商业使用请联系获取商业许可。
+
+## 联系作者
+
+<p align="center">
+  <img src="docs/author/qq_qrcode.jpg" width="200" alt="QQ二维码">
+</p>
+
+## Star History
+
+[![Star History Chart](https://api.star-history.com/svg?repos=fawney19/Aether&type=Date)](https://star-history.com/#fawney19/Aether&Date)
+
+

frontend/src/api/dashboard.ts

@@ -87,6 +87,8 @@ export interface DashboardStatsResponse {
   cache_stats?: CacheStats
   users?: UserStats
   token_breakdown?: TokenBreakdown
+  // 普通用户专用字段
+  monthly_cost?: number
 }
 
 export interface RecentRequestsResponse {

frontend/src/api/endpoints/global-models.ts

@@ -4,7 +4,8 @@ import type {
   GlobalModelUpdate,
   GlobalModelResponse,
   GlobalModelWithStats,
-  GlobalModelListResponse
+  GlobalModelListResponse,
+  ModelCatalogProviderDetail,
 } from './types'
 
 /**
@@ -83,3 +84,16 @@ export async function batchAssignToProviders(
   )
   return response.data
 }
+
+/**
+ * 获取 GlobalModel 的所有关联提供商（包括非活跃的）
+ */
+export async function getGlobalModelProviders(globalModelId: string): Promise<{
+  providers: ModelCatalogProviderDetail[]
+  total: number
+}> {
+  const response = await client.get(
+    `/api/admin/models/global/${globalModelId}/providers`
+  )
+  return response.data
+}

frontend/src/api/endpoints/keys.ts

@@ -110,6 +110,14 @@ export async function updateEndpointKey(
   return response.data
 }
 
+/**
+ * 获取完整的 API Key（用于查看和复制）
+ */
+export async function revealEndpointKey(keyId: string): Promise<{ api_key: string }> {
+  const response = await client.get(`/api/admin/endpoints/keys/${keyId}/reveal`)
+  return response.data
+}
+
 /**
  * 删除 Endpoint Key
  */

frontend/src/api/global-models.ts

@@ -20,4 +20,5 @@ export {
   updateGlobalModel,
   deleteGlobalModel,
   batchAssignToProviders,
+  getGlobalModelProviders,
 } from './endpoints/global-models'

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

@@ -116,6 +116,19 @@
                 {{ model.global_model_name }}
               </div>
             </div>
+
+            <!-- 测试按钮 -->
+            <Button
+              variant="ghost"
+              size="icon"
+              class="h-7 w-7 shrink-0"
+              title="测试模型连接"
+              :disabled="testingModelName === model.global_model_name"
+              @click.stop="testModelConnection(model)"
+            >
+              <Loader2 v-if="testingModelName === model.global_model_name" class="w-3.5 h-3.5 animate-spin" />
+              <Play v-else class="w-3.5 h-3.5" />
+            </Button>
           </div>
         </div>
       </div>
@@ -148,16 +161,17 @@
 
 <script setup lang="ts">
 import { ref, computed, watch } from 'vue'
-import { Box, Loader2, Settings2 } from 'lucide-vue-next'
+import { Box, Loader2, Settings2, Play } from 'lucide-vue-next'
 import { Dialog } from '@/components/ui'
 import Button from '@/components/ui/button.vue'
 import Badge from '@/components/ui/badge.vue'
 import Checkbox from '@/components/ui/checkbox.vue'
 import { useToast } from '@/composables/useToast'
-import { parseApiError } from '@/utils/errorParser'
+import { parseApiError, parseTestModelError } from '@/utils/errorParser'
 import {
   updateEndpointKey,
   getProviderAvailableSourceModels,
+  testModel,
   type EndpointAPIKey,
   type ProviderAvailableSourceModel
 } from '@/api/endpoints'
@@ -181,6 +195,7 @@ const loadingModels = ref(false)
 const availableModels = ref<ProviderAvailableSourceModel[]>([])
 const selectedModels = ref<string[]>([])
 const initialModels = ref<string[]>([])
+const testingModelName = ref<string | null>(null)
 
 // 监听对话框打开
 watch(() => props.open, (open) => {
@@ -268,6 +283,32 @@ function clearModels() {
   selectedModels.value = []
 }
 
+// 测试模型连接
+async function testModelConnection(model: ProviderAvailableSourceModel) {
+  if (!props.providerId || !props.apiKey || testingModelName.value) return
+
+  testingModelName.value = model.global_model_name
+  try {
+    const result = await testModel({
+      provider_id: props.providerId,
+      model_name: model.provider_model_name,
+      api_key_id: props.apiKey.id,
+      message: "hello"
+    })
+
+    if (result.success) {
+      success(`模型 "${model.display_name}" 测试成功`)
+    } else {
+      showError(`模型测试失败: ${parseTestModelError(result)}`)
+    }
+  } catch (err: any) {
+    const errorMsg = err.response?.data?.detail || err.message || '测试请求失败'
+    showError(`模型测试失败: ${errorMsg}`)
+  } finally {
+    testingModelName.value = null
+  }
+}
+
 function areArraysEqual(a: string[], b: string[]): boolean {
   if (a.length !== b.length) return false
   const sortedA = [...a].sort()

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

@@ -337,8 +337,40 @@
                                         {{ key.is_active ? '活跃' : '禁用' }}
                                       </Badge>
                                     </div>
-                                    <div class="text-[10px] font-mono text-muted-foreground truncate">
-                                      {{ key.api_key_masked }}
+                                    <div class="flex items-center gap-1">
+                                      <span class="text-[10px] font-mono text-muted-foreground truncate max-w-[180px]">
+                                        {{ revealedKeys.has(key.id) ? revealedKeys.get(key.id) : key.api_key_masked }}
+                                      </span>
+                                      <Button
+                                        variant="ghost"
+                                        size="icon"
+                                        class="h-5 w-5 shrink-0"
+                                        :title="revealedKeys.has(key.id) ? '隐藏密钥' : '显示密钥'"
+                                        :disabled="revealingKeyId === key.id"
+                                        @click.stop="toggleKeyReveal(key)"
+                                      >
+                                        <Loader2
+                                          v-if="revealingKeyId === key.id"
+                                          class="w-3 h-3 animate-spin"
+                                        />
+                                        <EyeOff
+                                          v-else-if="revealedKeys.has(key.id)"
+                                          class="w-3 h-3"
+                                        />
+                                        <Eye
+                                          v-else
+                                          class="w-3 h-3"
+                                        />
+                                      </Button>
+                                      <Button
+                                        variant="ghost"
+                                        size="icon"
+                                        class="h-5 w-5 shrink-0"
+                                        title="复制密钥"
+                                        @click.stop="copyFullKey(key)"
+                                      >
+                                        <Copy class="w-3 h-3" />
+                                      </Button>
                                     </div>
                                   </div>
                                   <div class="flex items-center gap-1.5 ml-auto shrink-0">
@@ -654,7 +686,9 @@ import {
   Power,
   Layers,
   GripVertical,
-  Copy
+  Copy,
+  Eye,
+  EyeOff
 } from 'lucide-vue-next'
 import { useEscapeKey } from '@/composables/useEscapeKey'
 import Button from '@/components/ui/button.vue'
@@ -681,6 +715,7 @@ import {
   updateEndpoint,
   updateEndpointKey,
   batchUpdateKeyPriority,
+  revealEndpointKey,
   type ProviderEndpoint,
   type EndpointAPIKey,
   type Model
@@ -731,6 +766,10 @@ const recoveringEndpointId = ref<string | null>(null)
 const togglingEndpointId = ref<string | null>(null)
 const togglingKeyId = ref<string | null>(null)
 
+// 密钥显示状态：key_id -> 完整密钥
+const revealedKeys = ref<Map<string, string>>(new Map())
+const revealingKeyId = ref<string | null>(null)
+
 // 模型相关状态
 const modelFormDialogOpen = ref(false)
 const editingModel = ref<Model | null>(null)
@@ -800,6 +839,9 @@ watch(() => props.open, (newOpen) => {
     currentEndpoint.value = null
     editingKey.value = null
     keyToDelete.value = null
+
+    // 清除已显示的密钥（安全考虑）
+    revealedKeys.value.clear()
   }
 })
 
@@ -888,6 +930,43 @@ function handleConfigKeyModels(key: EndpointAPIKey) {
   keyAllowedModelsDialogOpen.value = true
 }
 
+// 切换密钥显示/隐藏
+async function toggleKeyReveal(key: EndpointAPIKey) {
+  if (revealedKeys.value.has(key.id)) {
+    // 已显示，隐藏它
+    revealedKeys.value.delete(key.id)
+    return
+  }
+
+  // 未显示，调用 API 获取完整密钥
+  revealingKeyId.value = key.id
+  try {
+    const result = await revealEndpointKey(key.id)
+    revealedKeys.value.set(key.id, result.api_key)
+  } catch (err: any) {
+    showError(err.response?.data?.detail || '获取密钥失败', '错误')
+  } finally {
+    revealingKeyId.value = null
+  }
+}
+
+// 复制完整密钥
+async function copyFullKey(key: EndpointAPIKey) {
+  // 如果已经显示了，直接复制
+  if (revealedKeys.value.has(key.id)) {
+    copyToClipboard(revealedKeys.value.get(key.id)!)
+    return
+  }
+
+  // 否则先获取再复制
+  try {
+    const result = await revealEndpointKey(key.id)
+    copyToClipboard(result.api_key)
+  } catch (err: any) {
+    showError(err.response?.data?.detail || '获取密钥失败', '错误')
+  }
+}
+
 function handleDeleteKey(key: EndpointAPIKey) {
   keyToDelete.value = key
   deleteKeyConfirmOpen.value = true

frontend/src/features/providers/components/provider-tabs/ModelsTab.vue

@@ -156,17 +156,6 @@
             </td>
             <td class="align-top px-4 py-3">
               <div class="flex justify-center gap-1.5">
-                <Button
-                  variant="ghost"
-                  size="icon"
-                  class="h-8 w-8"
-                  title="测试模型"
-                  :disabled="testingModelId === model.id"
-                  @click="testModelConnection(model)"
-                >
-                  <Loader2 v-if="testingModelId === model.id" class="w-3.5 h-3.5 animate-spin" />
-                  <Play v-else class="w-3.5 h-3.5" />
-                </Button>
                 <Button
                   variant="ghost"
                   size="icon"
@@ -220,14 +209,13 @@
 
 <script setup lang="ts">
 import { ref, computed, onMounted } from 'vue'
-import { Box, Edit, Trash2, Layers, Eye, Wrench, Zap, Brain, Power, Copy, Image, Loader2, Play } from 'lucide-vue-next'
+import { Box, Edit, Trash2, Layers, Eye, Wrench, Zap, Brain, Power, Copy, Image } from 'lucide-vue-next'
 import Card from '@/components/ui/card.vue'
 import Button from '@/components/ui/button.vue'
 import { useToast } from '@/composables/useToast'
 import { useClipboard } from '@/composables/useClipboard'
-import { getProviderModels, testModel, type Model } from '@/api/endpoints'
+import { getProviderModels, type Model } from '@/api/endpoints'
 import { updateModel } from '@/api/endpoints/models'
-import { parseTestModelError } from '@/utils/errorParser'
 
 const props = defineProps<{
   provider: any
@@ -246,7 +234,6 @@ const { copyToClipboard } = useClipboard()
 const loading = ref(false)
 const models = ref<Model[]>([])
 const togglingModelId = ref<string | null>(null)
-const testingModelId = ref<string | null>(null)
 
 // 按名称排序的模型列表
 const sortedModels = computed(() => {
@@ -390,39 +377,6 @@ async function toggleModelActive(model: Model) {
   }
 }
 
-// 测试模型连接性
-async function testModelConnection(model: Model) {
-  if (testingModelId.value) return
-
-  testingModelId.value = model.id
-  try {
-    const result = await testModel({
-      provider_id: props.provider.id,
-      model_name: model.provider_model_name,
-      message: "hello"
-    })
-
-    if (result.success) {
-      showSuccess(`模型 "${model.provider_model_name}" 测试成功`)
-
-      // 如果有响应内容，可以显示更多信息
-      if (result.data?.response?.choices?.[0]?.message?.content) {
-        const content = result.data.response.choices[0].message.content
-        showSuccess(`测试成功，响应: ${content.substring(0, 100)}${content.length > 100 ? '...' : ''}`)
-      } else if (result.data?.content_preview) {
-        showSuccess(`流式测试成功，预览: ${result.data.content_preview}`)
-      }
-    } else {
-      showError(`模型测试失败: ${parseTestModelError(result)}`)
-    }
-  } catch (err: any) {
-    const errorMsg = err.response?.data?.detail || err.message || '测试请求失败'
-    showError(`模型测试失败: ${errorMsg}`)
-  } finally {
-    testingModelId.value = null
-  }
-}
-
 onMounted(() => {
   loadModels()
 })

frontend/src/mocks/data.ts

@@ -5,7 +5,7 @@
 
 import type { User, LoginResponse } from '@/api/auth'
 import type { DashboardStatsResponse, RecentRequest, ProviderStatus, DailyStatsResponse } from '@/api/dashboard'
-import type { User as AdminUser, ApiKey } from '@/api/users'
+import type { User as AdminUser } from '@/api/users'
 import type { AdminApiKeysResponse } from '@/api/admin'
 import type { Profile, UsageResponse } from '@/api/me'
 import type { ProviderWithEndpointsSummary, GlobalModelResponse } from '@/api/endpoints/types'
@@ -185,18 +185,20 @@ export const MOCK_DASHBOARD_STATS: DashboardStatsResponse = {
     output: 700000,
     cache_creation: 50000,
     cache_read: 200000
-  }
+  },
+  // 普通用户专用字段
+  monthly_cost: 45.67
 }
 
 export const MOCK_RECENT_REQUESTS: RecentRequest[] = [
-  { id: 'req-001', user: 'alice', model: 'claude-sonnet-4-20250514', tokens: 15234, time: '2 分钟前' },
-  { id: 'req-002', user: 'bob', model: 'gpt-4o', tokens: 8765, time: '5 分钟前' },
-  { id: 'req-003', user: 'charlie', model: 'claude-opus-4-20250514', tokens: 32100, time: '8 分钟前' },
-  { id: 'req-004', user: 'diana', model: 'gemini-2.0-flash', tokens: 4521, time: '12 分钟前' },
-  { id: 'req-005', user: 'eve', model: 'claude-sonnet-4-20250514', tokens: 9876, time: '15 分钟前' },
-  { id: 'req-006', user: 'frank', model: 'gpt-4o-mini', tokens: 2345, time: '18 分钟前' },
-  { id: 'req-007', user: 'grace', model: 'claude-haiku-3-5-20241022', tokens: 6789, time: '22 分钟前' },
-  { id: 'req-008', user: 'henry', model: 'gemini-2.5-pro', tokens: 12345, time: '25 分钟前' }
+  { id: 'req-001', user: 'alice', model: 'claude-sonnet-4-5-20250929', tokens: 15234, time: '2 分钟前' },
+  { id: 'req-002', user: 'bob', model: 'gpt-5.1', tokens: 8765, time: '5 分钟前' },
+  { id: 'req-003', user: 'charlie', model: 'claude-opus-4-5-20251101', tokens: 32100, time: '8 分钟前' },
+  { id: 'req-004', user: 'diana', model: 'gemini-3-pro-preview', tokens: 4521, time: '12 分钟前' },
+  { id: 'req-005', user: 'eve', model: 'claude-sonnet-4-5-20250929', tokens: 9876, time: '15 分钟前' },
+  { id: 'req-006', user: 'frank', model: 'gpt-5.1-codex-mini', tokens: 2345, time: '18 分钟前' },
+  { id: 'req-007', user: 'grace', model: 'claude-haiku-4-5-20251001', tokens: 6789, time: '22 分钟前' },
+  { id: 'req-008', user: 'henry', model: 'gemini-3-pro-preview', tokens: 12345, time: '25 分钟前' }
 ]
 
 export const MOCK_PROVIDER_STATUS: ProviderStatus[] = [
@@ -231,11 +233,11 @@ function generateDailyStats(): DailyStatsResponse {
       unique_models: 8 + Math.floor(Math.random() * 5),
       unique_providers: 4 + Math.floor(Math.random() * 3),
       model_breakdown: [
-        { model: 'claude-sonnet-4-20250514', requests: Math.floor(baseRequests * 0.35), tokens: Math.floor(baseTokens * 0.35), cost: Number((baseCost * 0.35).toFixed(2)) },
-        { model: 'gpt-4o', requests: Math.floor(baseRequests * 0.25), tokens: Math.floor(baseTokens * 0.25), cost: Number((baseCost * 0.25).toFixed(2)) },
-        { model: 'claude-opus-4-20250514', requests: Math.floor(baseRequests * 0.15), tokens: Math.floor(baseTokens * 0.15), cost: Number((baseCost * 0.20).toFixed(2)) },
-        { model: 'gemini-2.0-flash', requests: Math.floor(baseRequests * 0.15), tokens: Math.floor(baseTokens * 0.15), cost: Number((baseCost * 0.10).toFixed(2)) },
-        { model: 'claude-haiku-3-5-20241022', requests: Math.floor(baseRequests * 0.10), tokens: Math.floor(baseTokens * 0.10), cost: Number((baseCost * 0.10).toFixed(2)) }
+        { model: 'claude-sonnet-4-5-20250929', requests: Math.floor(baseRequests * 0.35), tokens: Math.floor(baseTokens * 0.35), cost: Number((baseCost * 0.35).toFixed(2)) },
+        { model: 'gpt-5.1', requests: Math.floor(baseRequests * 0.25), tokens: Math.floor(baseTokens * 0.25), cost: Number((baseCost * 0.25).toFixed(2)) },
+        { model: 'claude-opus-4-5-20251101', requests: Math.floor(baseRequests * 0.15), tokens: Math.floor(baseTokens * 0.15), cost: Number((baseCost * 0.20).toFixed(2)) },
+        { model: 'gemini-3-pro-preview', requests: Math.floor(baseRequests * 0.15), tokens: Math.floor(baseTokens * 0.15), cost: Number((baseCost * 0.10).toFixed(2)) },
+        { model: 'claude-haiku-4-5-20251001', requests: Math.floor(baseRequests * 0.10), tokens: Math.floor(baseTokens * 0.10), cost: Number((baseCost * 0.10).toFixed(2)) }
       ]
     })
   }
@@ -243,11 +245,11 @@ function generateDailyStats(): DailyStatsResponse {
   return {
     daily_stats: dailyStats,
     model_summary: [
-      { model: 'claude-sonnet-4-20250514', requests: 2456, tokens: 8500000, cost: 125.45, avg_response_time: 1.2, cost_per_request: 0.051, tokens_per_request: 3461 },
-      { model: 'gpt-4o', requests: 1823, tokens: 6200000, cost: 98.32, avg_response_time: 0.9, cost_per_request: 0.054, tokens_per_request: 3401 },
-      { model: 'claude-opus-4-20250514', requests: 987, tokens: 4100000, cost: 156.78, avg_response_time: 2.1, cost_per_request: 0.159, tokens_per_request: 4154 },
-      { model: 'gemini-2.0-flash', requests: 1234, tokens: 3800000, cost: 28.56, avg_response_time: 0.6, cost_per_request: 0.023, tokens_per_request: 3079 },
-      { model: 'claude-haiku-3-5-20241022', requests: 2100, tokens: 5200000, cost: 32.10, avg_response_time: 0.5, cost_per_request: 0.015, tokens_per_request: 2476 }
+      { model: 'claude-sonnet-4-5-20250929', requests: 2456, tokens: 8500000, cost: 125.45, avg_response_time: 1.2, cost_per_request: 0.051, tokens_per_request: 3461 },
+      { model: 'gpt-5.1', requests: 1823, tokens: 6200000, cost: 98.32, avg_response_time: 0.9, cost_per_request: 0.054, tokens_per_request: 3401 },
+      { model: 'claude-opus-4-5-20251101', requests: 987, tokens: 4100000, cost: 156.78, avg_response_time: 2.1, cost_per_request: 0.159, tokens_per_request: 4154 },
+      { model: 'gemini-3-pro-preview', requests: 1234, tokens: 3800000, cost: 28.56, avg_response_time: 0.6, cost_per_request: 0.023, tokens_per_request: 3079 },
+      { model: 'claude-haiku-4-5-20251001', requests: 2100, tokens: 5200000, cost: 32.10, avg_response_time: 0.5, cost_per_request: 0.015, tokens_per_request: 2476 }
     ],
     period: {
       start_date: dailyStats[0].date,
@@ -336,7 +338,7 @@ export const MOCK_ALL_USERS: AdminUser[] = [
 
 // ========== API Key 数据 ==========
 
-export const MOCK_USER_API_KEYS: ApiKey[] = [
+export const MOCK_USER_API_KEYS = [
   {
     id: 'key-uuid-001',
     key_display: 'sk-ae...x7f9',
@@ -346,7 +348,8 @@ export const MOCK_USER_API_KEYS: ApiKey[] = [
     is_active: true,
     is_standalone: false,
     total_requests: 1234,
-    total_cost_usd: 45.67
+    total_cost_usd: 45.67,
+    force_capabilities: null
   },
   {
     id: 'key-uuid-002',
@@ -357,7 +360,8 @@ export const MOCK_USER_API_KEYS: ApiKey[] = [
     is_active: true,
     is_standalone: false,
     total_requests: 5678,
-    total_cost_usd: 123.45
+    total_cost_usd: 123.45,
+    force_capabilities: { cache_1h: true }
   },
   {
     id: 'key-uuid-003',
@@ -367,7 +371,8 @@ export const MOCK_USER_API_KEYS: ApiKey[] = [
     is_active: false,
     is_standalone: false,
     total_requests: 100,
-    total_cost_usd: 2.34
+    total_cost_usd: 2.34,
+    force_capabilities: null
   }
 ]
 
@@ -813,16 +818,16 @@ export const MOCK_USAGE_RESPONSE: UsageResponse = {
   quota_usd: 100,
   used_usd: 45.32,
   summary_by_model: [
-    { model: 'claude-sonnet-4-20250514', requests: 456, input_tokens: 650000, output_tokens: 250000, total_tokens: 900000, total_cost_usd: 18.50, actual_total_cost_usd: 13.50 },
-    { model: 'gpt-4o', requests: 312, input_tokens: 480000, output_tokens: 180000, total_tokens: 660000, total_cost_usd: 12.30, actual_total_cost_usd: 9.20 },
-    { model: 'claude-haiku-3-5-20241022', requests: 289, input_tokens: 420000, output_tokens: 170000, total_tokens: 590000, total_cost_usd: 8.50, actual_total_cost_usd: 6.30 },
-    { model: 'gemini-2.0-flash', requests: 177, input_tokens: 250000, output_tokens: 100000, total_tokens: 350000, total_cost_usd: 6.37, actual_total_cost_usd: 4.33 }
+    { model: 'claude-sonnet-4-5-20250929', requests: 456, input_tokens: 650000, output_tokens: 250000, total_tokens: 900000, total_cost_usd: 18.50, actual_total_cost_usd: 13.50 },
+    { model: 'gpt-5.1', requests: 312, input_tokens: 480000, output_tokens: 180000, total_tokens: 660000, total_cost_usd: 12.30, actual_total_cost_usd: 9.20 },
+    { model: 'claude-haiku-4-5-20251001', requests: 289, input_tokens: 420000, output_tokens: 170000, total_tokens: 590000, total_cost_usd: 8.50, actual_total_cost_usd: 6.30 },
+    { model: 'gemini-3-pro-preview', requests: 177, input_tokens: 250000, output_tokens: 100000, total_tokens: 350000, total_cost_usd: 6.37, actual_total_cost_usd: 4.33 }
   ],
   records: [
     {
       id: 'usage-001',
       provider: 'anthropic',
-      model: 'claude-sonnet-4-20250514',
+      model: 'claude-sonnet-4-5-20250929',
       input_tokens: 1500,
       output_tokens: 800,
       total_tokens: 2300,
@@ -837,7 +842,7 @@ export const MOCK_USAGE_RESPONSE: UsageResponse = {
     {
       id: 'usage-002',
       provider: 'openai',
-      model: 'gpt-4o',
+      model: 'gpt-5.1',
       input_tokens: 2000,
       output_tokens: 500,
       total_tokens: 2500,

frontend/src/mocks/handler.ts

@@ -405,10 +405,10 @@ function getUsageRecords() {
 
 // Mock 映射数据
 const MOCK_ALIASES = [
-  { id: 'alias-001', source_model: 'claude-4-sonnet', target_global_model_id: 'gm-001', target_global_model_name: 'claude-sonnet-4-20250514', target_global_model_display_name: 'Claude Sonnet 4', provider_id: null, provider_name: null, scope: 'global', mapping_type: 'alias', is_active: true, created_at: '2024-01-01T00:00:00Z', updated_at: '2024-01-01T00:00:00Z' },
-  { id: 'alias-002', source_model: 'claude-4-opus', target_global_model_id: 'gm-002', target_global_model_name: 'claude-opus-4-20250514', target_global_model_display_name: 'Claude Opus 4', provider_id: null, provider_name: null, scope: 'global', mapping_type: 'alias', is_active: true, created_at: '2024-01-01T00:00:00Z', updated_at: '2024-01-01T00:00:00Z' },
-  { id: 'alias-003', source_model: 'gpt4o', target_global_model_id: 'gm-004', target_global_model_name: 'gpt-4o', target_global_model_display_name: 'GPT-4o', provider_id: null, provider_name: null, scope: 'global', mapping_type: 'alias', is_active: true, created_at: '2024-01-01T00:00:00Z', updated_at: '2024-01-01T00:00:00Z' },
-  { id: 'alias-004', source_model: 'gemini-flash', target_global_model_id: 'gm-005', target_global_model_name: 'gemini-2.0-flash', target_global_model_display_name: 'Gemini 2.0 Flash', provider_id: null, provider_name: null, scope: 'global', mapping_type: 'alias', is_active: true, created_at: '2024-01-01T00:00:00Z', updated_at: '2024-01-01T00:00:00Z' }
+  { id: 'alias-001', source_model: 'claude-4-sonnet', target_global_model_id: 'gm-003', target_global_model_name: 'claude-sonnet-4-5-20250929', target_global_model_display_name: 'Claude Sonnet 4.5', provider_id: null, provider_name: null, scope: 'global', mapping_type: 'alias', is_active: true, created_at: '2024-01-01T00:00:00Z', updated_at: '2024-01-01T00:00:00Z' },
+  { id: 'alias-002', source_model: 'claude-4-opus', target_global_model_id: 'gm-002', target_global_model_name: 'claude-opus-4-5-20251101', target_global_model_display_name: 'Claude Opus 4.5', provider_id: null, provider_name: null, scope: 'global', mapping_type: 'alias', is_active: true, created_at: '2024-01-01T00:00:00Z', updated_at: '2024-01-01T00:00:00Z' },
+  { id: 'alias-003', source_model: 'gpt5', target_global_model_id: 'gm-006', target_global_model_name: 'gpt-5.1', target_global_model_display_name: 'GPT-5.1', provider_id: null, provider_name: null, scope: 'global', mapping_type: 'alias', is_active: true, created_at: '2024-01-01T00:00:00Z', updated_at: '2024-01-01T00:00:00Z' },
+  { id: 'alias-004', source_model: 'gemini-pro', target_global_model_id: 'gm-005', target_global_model_name: 'gemini-3-pro-preview', target_global_model_display_name: 'Gemini 3 Pro Preview', provider_id: null, provider_name: null, scope: 'global', mapping_type: 'alias', is_active: true, created_at: '2024-01-01T00:00:00Z', updated_at: '2024-01-01T00:00:00Z' }
 ]
 
 // Mock Endpoint Keys
@@ -2172,10 +2172,10 @@ function generateIntervalTimelineData(
 
   // 模型列表（用于按模型区分颜色）
   const models = [
-    'claude-sonnet-4-20250514',
-    'claude-3-5-sonnet-20241022',
-    'claude-3-5-haiku-20241022',
-    'claude-opus-4-20250514'
+    'claude-sonnet-4-5-20250929',
+    'claude-haiku-4-5-20251001',
+    'claude-opus-4-5-20251101',
+    'gpt-5.1'
   ]
 
   // 生成模拟的请求间隔数据

frontend/src/views/admin/ModelManagement.vue

@@ -737,6 +737,7 @@ import {
   updateGlobalModel,
   deleteGlobalModel,
   batchAssignToProviders,
+  getGlobalModelProviders,
   type GlobalModelResponse,
 } from '@/api/global-models'
 import { log } from '@/utils/logger'
@@ -1080,18 +1081,11 @@ async function selectModel(model: GlobalModelResponse) {
 async function loadModelProviders(_globalModelId: string) {
   loadingModelProviders.value = true
   try {
-    // 使用 ModelCatalog API 获取详细的关联提供商信息
-    const { getModelCatalog } = await import('@/api/endpoints')
-    const catalogResponse = await getModelCatalog()
+    // 使用新的 API 获取所有关联提供商（包括非活跃的）
+    const response = await getGlobalModelProviders(_globalModelId)
 
-    // 查找当前 GlobalModel 对应的 catalog item
-    const catalogItem = catalogResponse.models.find(
-      m => m.global_model_name === selectedModel.value?.name
-    )
-
-    if (catalogItem) {
-      // 转换为展示格式，包含完整的模型实现信息
-      selectedModelProviders.value = catalogItem.providers.map(p => ({
+    // 转换为展示格式
+    selectedModelProviders.value = response.providers.map(p => ({
       id: p.provider_id,
       model_id: p.model_id,
       display_name: p.provider_display_name || p.provider_name,
@@ -1113,9 +1107,6 @@ async function loadModelProviders(_globalModelId: string) {
       supports_function_calling: p.supports_function_calling,
       supports_streaming: p.supports_streaming
     }))
-    } else {
-      selectedModelProviders.value = []
-    }
   } catch (err: any) {
     log.error('加载关联提供商失败:', err)
     showError(parseApiError(err, '加载关联提供商失败'), '错误')

frontend/src/views/shared/Dashboard.vue

@@ -145,10 +145,10 @@
               <DollarSign class="absolute top-3 right-3 h-3.5 w-3.5 sm:h-4 sm:w-4 text-muted-foreground" />
               <div class="pr-6">
                 <p class="text-[9px] sm:text-[10px] font-semibold uppercase tracking-[0.2em] sm:tracking-[0.3em] text-muted-foreground">
-                  实际成本
+                  本月费用
                 </p>
                 <p class="mt-1.5 sm:mt-2 text-lg sm:text-xl font-semibold text-foreground">
-                  {{ formatCurrency(costStats.total_actual_cost) }}
+                  {{ formatCurrency(costStats.total_cost) }}
                 </p>
                 <Badge
                   v-if="costStats.cost_savings > 0"
@@ -162,14 +162,14 @@
           </div>
         </div>
 
-        <!-- 普通用户：缓存统计 -->
+        <!-- 普通用户：月度统计 -->
         <div
-          v-else-if="!isAdmin && cacheStats && cacheStats.total_cache_tokens > 0"
+          v-else-if="!isAdmin && (hasCacheData || (userMonthlyCost !== null && userMonthlyCost > 0))"
           class="mt-6"
         >
           <div class="mb-3 flex items-center justify-between">
             <h3 class="text-sm font-medium text-foreground">
-              本月缓存使用
+              本月统计
             </h3>
             <Badge
               variant="outline"
@@ -178,8 +178,16 @@
               Monthly
             </Badge>
           </div>
-          <div class="grid grid-cols-2 gap-2 sm:gap-3 xl:grid-cols-4">
-            <Card class="relative p-3 sm:p-4 border-book-cloth/30">
+          <div
+            :class="[
+              'grid gap-2 sm:gap-3',
+              hasCacheData ? 'grid-cols-2 xl:grid-cols-4' : 'grid-cols-1 max-w-xs'
+            ]"
+          >
+            <Card
+              v-if="cacheStats"
+              class="relative p-3 sm:p-4 border-book-cloth/30"
+            >
               <Database class="absolute top-3 right-3 h-3.5 w-3.5 sm:h-4 sm:w-4 text-muted-foreground" />
               <div class="pr-6">
                 <p class="text-[9px] sm:text-[10px] font-semibold uppercase tracking-[0.2em] sm:tracking-[0.3em] text-muted-foreground">
@@ -190,7 +198,10 @@
                 </p>
               </div>
             </Card>
-            <Card class="relative p-3 sm:p-4 border-kraft/30">
+            <Card
+              v-if="cacheStats"
+              class="relative p-3 sm:p-4 border-kraft/30"
+            >
               <Hash class="absolute top-3 right-3 h-3.5 w-3.5 sm:h-4 sm:w-4 text-muted-foreground" />
               <div class="pr-6">
                 <p class="text-[9px] sm:text-[10px] font-semibold uppercase tracking-[0.2em] sm:tracking-[0.3em] text-muted-foreground">
@@ -201,7 +212,10 @@
                 </p>
               </div>
             </Card>
-            <Card class="relative p-3 sm:p-4 border-book-cloth/25">
+            <Card
+              v-if="cacheStats"
+              class="relative p-3 sm:p-4 border-book-cloth/25"
+            >
               <Database class="absolute top-3 right-3 h-3.5 w-3.5 sm:h-4 sm:w-4 text-muted-foreground" />
               <div class="pr-6">
                 <p class="text-[9px] sm:text-[10px] font-semibold uppercase tracking-[0.2em] sm:tracking-[0.3em] text-muted-foreground">
@@ -213,19 +227,16 @@
               </div>
             </Card>
             <Card
-              v-if="tokenBreakdown"
+              v-if="userMonthlyCost !== null"
               class="relative p-3 sm:p-4 border-manilla/40"
             >
-              <Hash class="absolute top-3 right-3 h-3.5 w-3.5 sm:h-4 sm:w-4 text-muted-foreground" />
+              <DollarSign class="absolute top-3 right-3 h-3.5 w-3.5 sm:h-4 sm:w-4 text-muted-foreground" />
               <div class="pr-6">
                 <p class="text-[9px] sm:text-[10px] font-semibold uppercase tracking-[0.2em] sm:tracking-[0.3em] text-muted-foreground">
-                  总Token
+                  本月费用
                 </p>
                 <p class="mt-1.5 sm:mt-2 text-lg sm:text-xl font-semibold text-foreground">
-                  {{ formatTokens((tokenBreakdown.input || 0) + (tokenBreakdown.output || 0)) }}
-                </p>
-                <p class="mt-0.5 sm:mt-1 text-[9px] sm:text-[10px] text-muted-foreground">
-                  输入 {{ formatTokens(tokenBreakdown.input || 0) }} / 输出 {{ formatTokens(tokenBreakdown.output || 0) }}
+                  {{ formatCurrency(userMonthlyCost) }}
                 </p>
               </div>
             </Card>
@@ -831,6 +842,12 @@ const cacheStats = ref<{
   total_cache_tokens: number
 } | null>(null)
 
+const userMonthlyCost = ref<number | null>(null)
+
+const hasCacheData = computed(() =>
+  cacheStats.value && cacheStats.value.total_cache_tokens > 0
+)
+
 const tokenBreakdown = ref<{
   input: number
   output: number
@@ -1086,6 +1103,7 @@ async function loadDashboardData() {
     } else {
       if (statsData.cache_stats) cacheStats.value = statsData.cache_stats
       if (statsData.token_breakdown) tokenBreakdown.value = statsData.token_breakdown
+      if (statsData.monthly_cost !== undefined) userMonthlyCost.value = statsData.monthly_cost
     }
   } finally {
     loading.value = false

pyproject.toml

@@ -13,7 +13,7 @@ authors = [
 classifiers = [
     "Development Status :: 4 - Beta",
     "Intended Audience :: Developers",
-    "License :: OSI Approved :: MIT License",
+    "License :: Other/Proprietary License",
     "Programming Language :: Python :: 3",
     "Programming Language :: Python :: 3.8",
     "Programming Language :: Python :: 3.9",

src/api/admin/endpoints/keys.py

@@ -80,6 +80,17 @@ async def get_keys_grouped_by_format(
     return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
 
 
+@router.get("/keys/{key_id}/reveal")
+async def reveal_endpoint_key(
+    key_id: str,
+    request: Request,
+    db: Session = Depends(get_db),
+) -> dict:
+    """获取完整的 API Key（用于查看和复制）"""
+    adapter = AdminRevealEndpointKeyAdapter(key_id=key_id)
+    return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
+
+
 @router.delete("/keys/{key_id}")
 async def delete_endpoint_key(
     key_id: str,
@@ -293,6 +304,30 @@ class AdminUpdateEndpointKeyAdapter(AdminApiAdapter):
         return EndpointAPIKeyResponse(**response_dict)
 
 
+@dataclass
+class AdminRevealEndpointKeyAdapter(AdminApiAdapter):
+    """获取完整的 API Key（用于查看和复制）"""
+
+    key_id: str
+
+    async def handle(self, context):  # type: ignore[override]
+        db = context.db
+        key = db.query(ProviderAPIKey).filter(ProviderAPIKey.id == self.key_id).first()
+        if not key:
+            raise NotFoundException(f"Key {self.key_id} 不存在")
+
+        try:
+            decrypted_key = crypto_service.decrypt(key.api_key)
+        except Exception as e:
+            logger.error(f"解密 Key 失败: ID={self.key_id}, Error={e}")
+            raise InvalidRequestException(
+                "无法解密 API Key，可能是加密密钥已更改。请重新添加该密钥。"
+            )
+
+        logger.info(f"[REVEAL] 查看完整 Key: ID={self.key_id}, Name={key.name}")
+        return {"api_key": decrypted_key}
+
+
 @dataclass
 class AdminDeleteEndpointKeyAdapter(AdminApiAdapter):
     key_id: str

src/api/admin/models/global_models.py

@@ -5,7 +5,7 @@ GlobalModel Admin API
 """
 
 from dataclasses import dataclass
-from typing import List, Optional
+from typing import Optional
 
 from fastapi import APIRouter, Depends, Query, Request
 from sqlalchemy.orm import Session
@@ -19,9 +19,11 @@ from src.models.pydantic_models import (
     BatchAssignToProvidersResponse,
     GlobalModelCreate,
     GlobalModelListResponse,
+    GlobalModelProvidersResponse,
     GlobalModelResponse,
     GlobalModelUpdate,
     GlobalModelWithStats,
+    ModelCatalogProviderDetail,
 )
 from src.services.model.global_model import GlobalModelService
 
@@ -108,6 +110,17 @@ async def batch_assign_to_providers(
     return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
 
 
+@router.get("/{global_model_id}/providers", response_model=GlobalModelProvidersResponse)
+async def get_global_model_providers(
+    request: Request,
+    global_model_id: str,
+    db: Session = Depends(get_db),
+) -> GlobalModelProvidersResponse:
+    """获取 GlobalModel 的所有关联提供商（包括非活跃的）"""
+    adapter = AdminGetGlobalModelProvidersAdapter(global_model_id=global_model_id)
+    return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
+
+
 # ========== Adapters ==========
 
 
@@ -275,3 +288,61 @@ class AdminBatchAssignToProvidersAdapter(AdminApiAdapter):
         logger.info(f"批量为 Provider 添加 GlobalModel: global_model_id={self.global_model_id} success={len(result['success'])} errors={len(result['errors'])}")
 
         return BatchAssignToProvidersResponse(**result)
+
+
+@dataclass
+class AdminGetGlobalModelProvidersAdapter(AdminApiAdapter):
+    """获取 GlobalModel 的所有关联提供商（包括非活跃的）"""
+
+    global_model_id: str
+
+    async def handle(self, context):  # type: ignore[override]
+        from sqlalchemy.orm import joinedload
+
+        from src.models.database import Model
+
+        global_model = GlobalModelService.get_global_model(context.db, self.global_model_id)
+
+        # 获取所有关联的 Model（包括非活跃的）
+        models = (
+            context.db.query(Model)
+            .options(joinedload(Model.provider), joinedload(Model.global_model))
+            .filter(Model.global_model_id == global_model.id)
+            .all()
+        )
+
+        provider_entries = []
+        for model in models:
+            provider = model.provider
+            if not provider:
+                continue
+
+            effective_tiered = model.get_effective_tiered_pricing()
+            tier_count = len(effective_tiered.get("tiers", [])) if effective_tiered else 1
+
+            provider_entries.append(
+                ModelCatalogProviderDetail(
+                    provider_id=provider.id,
+                    provider_name=provider.name,
+                    provider_display_name=provider.display_name,
+                    model_id=model.id,
+                    target_model=model.provider_model_name,
+                    input_price_per_1m=model.get_effective_input_price(),
+                    output_price_per_1m=model.get_effective_output_price(),
+                    cache_creation_price_per_1m=model.get_effective_cache_creation_price(),
+                    cache_read_price_per_1m=model.get_effective_cache_read_price(),
+                    cache_1h_creation_price_per_1m=model.get_effective_1h_cache_creation_price(),
+                    price_per_request=model.get_effective_price_per_request(),
+                    effective_tiered_pricing=effective_tiered,
+                    tier_count=tier_count,
+                    supports_vision=model.get_effective_supports_vision(),
+                    supports_function_calling=model.get_effective_supports_function_calling(),
+                    supports_streaming=model.get_effective_supports_streaming(),
+                    is_active=bool(model.is_active),
+                )
+            )
+
+        return GlobalModelProvidersResponse(
+            providers=provider_entries,
+            total=len(provider_entries),
+        )

src/api/base/models_service.py

@@ -18,7 +18,15 @@ from sqlalchemy.orm import Session, joinedload
 from src.config.constants import CacheTTL
 from src.core.cache_service import CacheService
 from src.core.logger import logger
-from src.models.database import GlobalModel, Model, Provider, ProviderAPIKey, ProviderEndpoint
+from src.models.database import (
+    ApiKey,
+    GlobalModel,
+    Model,
+    Provider,
+    ProviderAPIKey,
+    ProviderEndpoint,
+    User,
+)
 
 # 缓存 key 前缀
 _CACHE_KEY_PREFIX = "models:list"
@@ -82,6 +90,7 @@ class ModelInfo:
     created_at: Optional[str]  # ISO 格式
     created_timestamp: int  # Unix 时间戳
     provider_name: str
+    provider_id: str = ""  # Provider ID，用于权限过滤
     # 能力配置
     streaming: bool = True
     vision: bool = False
@@ -99,6 +108,92 @@ class ModelInfo:
     output_modalities: Optional[list[str]] = None
 
 
+@dataclass
+class AccessRestrictions:
+    """API Key 或 User 的访问限制"""
+
+    allowed_providers: Optional[list[str]] = None  # 允许的 Provider ID 列表
+    allowed_models: Optional[list[str]] = None  # 允许的模型名称列表
+    allowed_api_formats: Optional[list[str]] = None  # 允许的 API 格式列表
+
+    @classmethod
+    def from_api_key_and_user(
+        cls, api_key: Optional[ApiKey], user: Optional[User]
+    ) -> "AccessRestrictions":
+        """
+        从 API Key 和 User 合并访问限制
+
+        限制逻辑:
+        - API Key 的限制优先于 User 的限制
+        - 如果 API Key 有限制，使用 API Key 的限制
+        - 如果 API Key 无限制但 User 有限制，使用 User 的限制
+        - 两者都无限制则返回空限制
+        """
+        allowed_providers: Optional[list[str]] = None
+        allowed_models: Optional[list[str]] = None
+        allowed_api_formats: Optional[list[str]] = None
+
+        # 优先使用 API Key 的限制
+        if api_key:
+            if api_key.allowed_providers is not None:
+                allowed_providers = api_key.allowed_providers
+            if api_key.allowed_models is not None:
+                allowed_models = api_key.allowed_models
+            if api_key.allowed_api_formats is not None:
+                allowed_api_formats = api_key.allowed_api_formats
+
+        # 如果 API Key 没有限制，检查 User 的限制
+        # 注意: User 没有 allowed_api_formats 字段
+        if user:
+            if allowed_providers is None and user.allowed_providers is not None:
+                allowed_providers = user.allowed_providers
+            if allowed_models is None and user.allowed_models is not None:
+                allowed_models = user.allowed_models
+
+        return cls(
+            allowed_providers=allowed_providers,
+            allowed_models=allowed_models,
+            allowed_api_formats=allowed_api_formats,
+        )
+
+    def is_api_format_allowed(self, api_format: str) -> bool:
+        """
+        检查 API 格式是否被允许
+
+        Args:
+            api_format: API 格式 (如 "OPENAI", "CLAUDE", "GEMINI")
+
+        Returns:
+            True 如果格式被允许，False 否则
+        """
+        if self.allowed_api_formats is None:
+            return True
+        return api_format in self.allowed_api_formats
+
+    def is_model_allowed(self, model_id: str, provider_id: str) -> bool:
+        """
+        检查模型是否被允许访问
+
+        Args:
+            model_id: 模型 ID
+            provider_id: Provider ID
+
+        Returns:
+            True 如果模型被允许，False 否则
+        """
+        # 检查 Provider 限制
+        if self.allowed_providers is not None:
+            if provider_id not in self.allowed_providers:
+                return False
+
+        # 检查模型限制
+        if self.allowed_models is not None:
+            if model_id not in self.allowed_models:
+                return False
+
+        return True
+
+
 def get_available_provider_ids(db: Session, api_formats: list[str]) -> set[str]:
     """
     返回有可用端点的 Provider IDs
@@ -218,6 +313,7 @@ def _extract_model_info(model: Any) -> ModelInfo:
     )
     created_timestamp: int = int(model.created_at.timestamp()) if model.created_at else 0
     provider_name: str = model.provider.name if model.provider else "unknown"
+    provider_id: str = model.provider_id or ""
 
     # 从 GlobalModel.config 提取配置信息
     config: dict = {}
@@ -233,6 +329,7 @@ def _extract_model_info(model: Any) -> ModelInfo:
         created_at=created_at,
         created_timestamp=created_timestamp,
         provider_name=provider_name,
+        provider_id=provider_id,
         # 能力配置
         streaming=config.get("streaming", True),
         vision=config.get("vision", False),
@@ -255,6 +352,7 @@ async def list_available_models(
     db: Session,
     available_provider_ids: set[str],
     api_formats: Optional[list[str]] = None,
+    restrictions: Optional[AccessRestrictions] = None,
 ) -> list[ModelInfo]:
     """
     获取可用模型列表（已去重，带缓存）
@@ -263,6 +361,7 @@ async def list_available_models(
         db: 数据库会话
         available_provider_ids: 有可用端点的 Provider ID 集合
         api_formats: API 格式列表，用于检查 Key 的 allowed_models
+        restrictions: API Key/User 的访问限制
 
     Returns:
         去重后的 ModelInfo 列表，按创建时间倒序
@@ -270,8 +369,16 @@ async def list_available_models(
     if not available_provider_ids:
         return []
 
+    # 缓存策略：只有完全无访问限制时才使用缓存
+    # - restrictions is None: 未传入限制对象
+    # - restrictions 的两个字段都为 None: 传入了限制对象但无实际限制
+    # 以上两种情况返回的结果相同，可以共享全局缓存
+    use_cache = restrictions is None or (
+        restrictions.allowed_providers is None and restrictions.allowed_models is None
+    )
+
     # 尝试从缓存获取
-    if api_formats:
+    if api_formats and use_cache:
         cached = await _get_cached_models(api_formats)
         if cached is not None:
             return cached
@@ -306,14 +413,19 @@ async def list_available_models(
         if available_model_ids is not None and info.id not in available_model_ids:
             continue
 
+        # 检查 API Key/User 访问限制
+        if restrictions is not None:
+            if not restrictions.is_model_allowed(info.id, info.provider_id):
+                continue
+
         if info.id in seen_model_ids:
             continue
         seen_model_ids.add(info.id)
 
         result.append(info)
 
-    # 写入缓存
-    if api_formats:
+    # 只有无限制的情况才写入缓存
+    if api_formats and use_cache:
         await _set_cached_models(api_formats, result)
 
     return result
@@ -324,6 +436,7 @@ def find_model_by_id(
     model_id: str,
     available_provider_ids: set[str],
     api_formats: Optional[list[str]] = None,
+    restrictions: Optional[AccessRestrictions] = None,
 ) -> Optional[ModelInfo]:
     """
     按 ID 查找模型
@@ -338,6 +451,7 @@ def find_model_by_id(
         model_id: 模型 ID
         available_provider_ids: 有可用端点的 Provider ID 集合
         api_formats: API 格式列表，用于检查 Key 的 allowed_models
+        restrictions: API Key/User 的访问限制
 
     Returns:
         ModelInfo 或 None
@@ -353,6 +467,11 @@ def find_model_by_id(
         if available_model_ids is not None and model_id not in available_model_ids:
             return None
 
+    # 快速检查：如果 restrictions 明确限制了模型列表且目标模型不在其中，直接返回 None
+    if restrictions is not None and restrictions.allowed_models is not None:
+        if model_id not in restrictions.allowed_models:
+            return None
+
     # 先按 GlobalModel.name 查找
     models_by_global = (
         db.query(Model)
@@ -368,8 +487,19 @@ def find_model_by_id(
         .all()
     )
 
+    def is_model_accessible(m: Model) -> bool:
+        """检查模型是否可访问"""
+        if m.provider_id not in available_provider_ids:
+            return False
+        # 检查 API Key/User 访问限制
+        if restrictions is not None:
+            provider_id = m.provider_id or ""
+            if not restrictions.is_model_allowed(model_id, provider_id):
+                return False
+        return True
+
     model = next(
-        (m for m in models_by_global if m.provider_id in available_provider_ids),
+        (m for m in models_by_global if is_model_accessible(m)),
         None,
     )
 
@@ -393,7 +523,7 @@ def find_model_by_id(
         )
 
         model = next(
-            (m for m in models_by_provider_name if m.provider_id in available_provider_ids),
+            (m for m in models_by_provider_name if is_model_accessible(m)),
             None,
         )
 

src/api/dashboard/routes.py

@@ -118,7 +118,9 @@ class AdminDashboardStatsAdapter(AdminApiAdapter):
         # 转换为 UTC 用于与 stats_daily.date 比较（存储的是业务日期对应的 UTC 开始时间）
         today = today_local.astimezone(timezone.utc)
         yesterday = (today_local - timedelta(days=1)).astimezone(timezone.utc)
-        last_month = (today_local - timedelta(days=30)).astimezone(timezone.utc)
+        # 本月第一天（自然月）
+        month_start_local = today_local.replace(day=1)
+        month_start = month_start_local.astimezone(timezone.utc)
 
         # ==================== 使用预聚合数据 ====================
         # 从 stats_summary + 今日实时数据获取全局统计
@@ -208,7 +210,7 @@ class AdminDashboardStatsAdapter(AdminApiAdapter):
                 func.sum(StatsDaily.cache_read_cost).label("cache_read_cost"),
                 func.sum(StatsDaily.fallback_count).label("fallback_count"),
             )
-            .filter(StatsDaily.date >= last_month, StatsDaily.date < today)
+            .filter(StatsDaily.date >= month_start, StatsDaily.date < today)
             .first()
         )
 
@@ -227,24 +229,24 @@ class AdminDashboardStatsAdapter(AdminApiAdapter):
         else:
             # 回退到实时查询（没有预聚合数据时）
             total_requests = (
-                db.query(func.count(Usage.id)).filter(Usage.created_at >= last_month).scalar() or 0
+                db.query(func.count(Usage.id)).filter(Usage.created_at >= month_start).scalar() or 0
             )
             total_cost = (
-                db.query(func.sum(Usage.total_cost_usd)).filter(Usage.created_at >= last_month).scalar() or 0
+                db.query(func.sum(Usage.total_cost_usd)).filter(Usage.created_at >= month_start).scalar() or 0
             )
             total_actual_cost = (
                 db.query(func.sum(Usage.actual_total_cost_usd))
-                .filter(Usage.created_at >= last_month).scalar() or 0
+                .filter(Usage.created_at >= month_start).scalar() or 0
             )
             error_requests = (
                 db.query(func.count(Usage.id))
                 .filter(
-                    Usage.created_at >= last_month,
+                    Usage.created_at >= month_start,
                     (Usage.status_code >= 400) | (Usage.error_message.isnot(None)),
                 ).scalar() or 0
             )
             total_tokens = (
-                db.query(func.sum(Usage.total_tokens)).filter(Usage.created_at >= last_month).scalar() or 0
+                db.query(func.sum(Usage.total_tokens)).filter(Usage.created_at >= month_start).scalar() or 0
             )
             cache_stats = (
                 db.query(
@@ -253,7 +255,7 @@ class AdminDashboardStatsAdapter(AdminApiAdapter):
                     func.sum(Usage.cache_creation_cost_usd).label("cache_creation_cost"),
                     func.sum(Usage.cache_read_cost_usd).label("cache_read_cost"),
                 )
-                .filter(Usage.created_at >= last_month)
+                .filter(Usage.created_at >= month_start)
                 .first()
             )
             cache_creation_tokens = int(cache_stats.cache_creation_tokens or 0) if cache_stats else 0
@@ -267,7 +269,7 @@ class AdminDashboardStatsAdapter(AdminApiAdapter):
                     RequestCandidate.request_id, func.count(RequestCandidate.id).label("executed_count")
                 )
                 .filter(
-                    RequestCandidate.created_at >= last_month,
+                    RequestCandidate.created_at >= month_start,
                     RequestCandidate.status.in_(["success", "failed"]),
                 )
                 .group_by(RequestCandidate.request_id)
@@ -447,7 +449,9 @@ class UserDashboardStatsAdapter(DashboardAdapter):
         # 转换为 UTC 用于数据库查询
         today = today_local.astimezone(timezone.utc)
         yesterday = (today_local - timedelta(days=1)).astimezone(timezone.utc)
-        last_month = (today_local - timedelta(days=30)).astimezone(timezone.utc)
+        # 本月第一天（自然月）
+        month_start_local = today_local.replace(day=1)
+        month_start = month_start_local.astimezone(timezone.utc)
 
         user_api_keys = db.query(func.count(ApiKey.id)).filter(ApiKey.user_id == user.id).scalar()
         active_keys = (
@@ -483,12 +487,12 @@ class UserDashboardStatsAdapter(DashboardAdapter):
         # 本月请求统计
         user_requests = (
             db.query(func.count(Usage.id))
-            .filter(and_(Usage.user_id == user.id, Usage.created_at >= last_month))
+            .filter(and_(Usage.user_id == user.id, Usage.created_at >= month_start))
             .scalar()
         )
         user_cost = (
             db.query(func.sum(Usage.total_cost_usd))
-            .filter(and_(Usage.user_id == user.id, Usage.created_at >= last_month))
+            .filter(and_(Usage.user_id == user.id, Usage.created_at >= month_start))
             .scalar()
             or 0
         )
@@ -532,18 +536,19 @@ class UserDashboardStatsAdapter(DashboardAdapter):
                 func.sum(Usage.cache_read_input_tokens).label("cache_read_tokens"),
                 func.sum(Usage.input_tokens).label("total_input_tokens"),
             )
-            .filter(and_(Usage.user_id == user.id, Usage.created_at >= last_month))
+            .filter(and_(Usage.user_id == user.id, Usage.created_at >= month_start))
             .first()
         )
         cache_creation_tokens = int(cache_stats.cache_creation_tokens or 0) if cache_stats else 0
         cache_read_tokens = int(cache_stats.cache_read_tokens or 0) if cache_stats else 0
+        monthly_input_tokens = int(cache_stats.total_input_tokens or 0) if cache_stats else 0
 
-        # 计算缓存命中率：cache_read / (input_tokens + cache_read)
+        # 计算本月缓存命中率：cache_read / (input_tokens + cache_read)
         # input_tokens 是实际发送给模型的输入（不含缓存读取），cache_read 是从缓存读取的
         # 总输入 = input_tokens + cache_read，缓存命中率 = cache_read / 总输入
-        total_input_with_cache = all_time_input_tokens + all_time_cache_read
+        total_input_with_cache = monthly_input_tokens + cache_read_tokens
         cache_hit_rate = (
-            round((all_time_cache_read / total_input_with_cache) * 100, 1)
+            round((cache_read_tokens / total_input_with_cache) * 100, 1)
             if total_input_with_cache > 0
             else 0
         )
@@ -569,15 +574,15 @@ class UserDashboardStatsAdapter(DashboardAdapter):
             quota_value = "无限制"
             quota_change = f"已用 ${user.used_usd:.2f}"
             quota_high = False
-        elif user.quota_usd and user.quota_usd > 0:
+        elif user.quota_usd > 0:
             percent = min(100, int((user.used_usd / user.quota_usd) * 100))
-            quota_value = "无限制"
+            quota_value = f"${user.quota_usd:.0f}"
             quota_change = f"已用 ${user.used_usd:.2f}"
             quota_high = percent > 80
         else:
-            quota_value = "0%"
+            quota_value = "$0"
             quota_change = f"已用 ${user.used_usd:.2f}"
-            quota_high = False
+            quota_high = True
 
         return {
             "stats": [
@@ -605,9 +610,15 @@ class UserDashboardStatsAdapter(DashboardAdapter):
                     "icon": "TrendingUp",
                 },
                 {
-                    "name": "本月费用",
-                    "value": f"${user_cost:.2f}",
-                    "icon": "DollarSign",
+                    "name": "总Token",
+                    "value": format_tokens(
+                        all_time_input_tokens
+                        + all_time_output_tokens
+                        + all_time_cache_creation
+                        + all_time_cache_read
+                    ),
+                    "subValue": f"输入 {format_tokens(all_time_input_tokens)} / 输出 {format_tokens(all_time_output_tokens)}",
+                    "icon": "Hash",
                 },
             ],
             "today": {
@@ -631,6 +642,8 @@ class UserDashboardStatsAdapter(DashboardAdapter):
                 "cache_hit_rate": cache_hit_rate,
                 "total_cache_tokens": cache_creation_tokens + cache_read_tokens,
             },
+            # 本月费用（用于下方缓存区域显示）
+            "monthly_cost": float(user_cost),
         }
 
 

src/api/public/models.py

@@ -14,6 +14,7 @@ from fastapi.responses import JSONResponse
 from sqlalchemy.orm import Session
 
 from src.api.base.models_service import (
+    AccessRestrictions,
     ModelInfo,
     find_model_by_id,
     get_available_provider_ids,
@@ -103,6 +104,35 @@ def _get_formats_for_api(api_format: str) -> list[str]:
         return _OPENAI_FORMATS
 
 
+def _build_empty_list_response(api_format: str) -> dict:
+    """根据 API 格式构建空列表响应"""
+    if api_format == "claude":
+        return {"data": [], "has_more": False, "first_id": None, "last_id": None}
+    elif api_format == "gemini":
+        return {"models": []}
+    else:
+        return {"object": "list", "data": []}
+
+
+def _filter_formats_by_restrictions(
+    formats: list[str], restrictions: AccessRestrictions, api_format: str
+) -> Tuple[list[str], Optional[dict]]:
+    """
+    根据访问限制过滤 API 格式
+
+    Returns:
+        (过滤后的格式列表, 空响应或None)
+        如果过滤后为空，返回对应格式的空响应
+    """
+    if restrictions.allowed_api_formats is None:
+        return formats, None
+    filtered = [f for f in formats if f in restrictions.allowed_api_formats]
+    if not filtered:
+        logger.info(f"[Models] API Key 不允许访问格式 {api_format}")
+        return [], _build_empty_list_response(api_format)
+    return filtered, None
+
+
 def _authenticate(db: Session, api_key: Optional[str]) -> Tuple[Optional[User], Optional[ApiKey]]:
     """
     认证 API Key
@@ -375,22 +405,24 @@ async def list_models(
     logger.info(f"[Models] GET /v1/models | format={api_format}")
 
     # 认证
-    user, _ = _authenticate(db, api_key)
+    user, key_record = _authenticate(db, api_key)
     if not user:
         return _build_auth_error_response(api_format)
 
+    # 构建访问限制
+    restrictions = AccessRestrictions.from_api_key_and_user(key_record, user)
+
+    # 检查 API 格式限制
     formats = _get_formats_for_api(api_format)
+    formats, empty_response = _filter_formats_by_restrictions(formats, restrictions, api_format)
+    if empty_response is not None:
+        return empty_response
 
     available_provider_ids = get_available_provider_ids(db, formats)
     if not available_provider_ids:
-        if api_format == "claude":
-            return {"data": [], "has_more": False, "first_id": None, "last_id": None}
-        elif api_format == "gemini":
-            return {"models": []}
-        else:
-            return {"object": "list", "data": []}
+        return _build_empty_list_response(api_format)
 
-    models = await list_available_models(db, available_provider_ids, formats)
+    models = await list_available_models(db, available_provider_ids, formats, restrictions)
     logger.debug(f"[Models] 返回 {len(models)} 个模型")
 
     if api_format == "claude":
@@ -419,14 +451,21 @@ async def retrieve_model(
     logger.info(f"[Models] GET /v1/models/{model_id} | format={api_format}")
 
     # 认证
-    user, _ = _authenticate(db, api_key)
+    user, key_record = _authenticate(db, api_key)
     if not user:
         return _build_auth_error_response(api_format)
 
+    # 构建访问限制
+    restrictions = AccessRestrictions.from_api_key_and_user(key_record, user)
+
+    # 检查 API 格式限制
     formats = _get_formats_for_api(api_format)
+    formats, _ = _filter_formats_by_restrictions(formats, restrictions, api_format)
+    if not formats:
+        return _build_404_response(model_id, api_format)
 
     available_provider_ids = get_available_provider_ids(db, formats)
-    model_info = find_model_by_id(db, model_id, available_provider_ids, formats)
+    model_info = find_model_by_id(db, model_id, available_provider_ids, formats, restrictions)
 
     if not model_info:
         return _build_404_response(model_id, api_format)
@@ -455,15 +494,25 @@ async def list_models_gemini(
     api_key = _extract_api_key_from_request(request, gemini_def)
 
     # 认证
-    user, _ = _authenticate(db, api_key)
+    user, key_record = _authenticate(db, api_key)
     if not user:
         return _build_auth_error_response("gemini")
 
-    available_provider_ids = get_available_provider_ids(db, _GEMINI_FORMATS)
+    # 构建访问限制
+    restrictions = AccessRestrictions.from_api_key_and_user(key_record, user)
+
+    # 检查 API 格式限制
+    formats, empty_response = _filter_formats_by_restrictions(
+        _GEMINI_FORMATS, restrictions, "gemini"
+    )
+    if empty_response is not None:
+        return empty_response
+
+    available_provider_ids = get_available_provider_ids(db, formats)
     if not available_provider_ids:
         return {"models": []}
 
-    models = await list_available_models(db, available_provider_ids, _GEMINI_FORMATS)
+    models = await list_available_models(db, available_provider_ids, formats, restrictions)
     logger.debug(f"[Models] 返回 {len(models)} 个模型")
     response = _build_gemini_list_response(models, page_size, page_token)
     logger.debug(f"[Models] Gemini 响应: {response}")
@@ -486,12 +535,22 @@ async def get_model_gemini(
     api_key = _extract_api_key_from_request(request, gemini_def)
 
     # 认证
-    user, _ = _authenticate(db, api_key)
+    user, key_record = _authenticate(db, api_key)
     if not user:
         return _build_auth_error_response("gemini")
 
-    available_provider_ids = get_available_provider_ids(db, _GEMINI_FORMATS)
-    model_info = find_model_by_id(db, model_id, available_provider_ids, _GEMINI_FORMATS)
+    # 构建访问限制
+    restrictions = AccessRestrictions.from_api_key_and_user(key_record, user)
+
+    # 检查 API 格式限制
+    formats, _ = _filter_formats_by_restrictions(_GEMINI_FORMATS, restrictions, "gemini")
+    if not formats:
+        return _build_404_response(model_id, "gemini")
+
+    available_provider_ids = get_available_provider_ids(db, formats)
+    model_info = find_model_by_id(
+        db, model_id, available_provider_ids, formats, restrictions
+    )
 
     if not model_info:
         return _build_404_response(model_id, "gemini")

src/clients/http_client.py

@@ -9,6 +9,7 @@ from urllib.parse import quote, urlparse
 
 import httpx
 
+from src.config import config
 from src.core.logger import logger
 
 
@@ -83,10 +84,10 @@ class HTTPClientPool:
                 http2=False,  # 暂时禁用HTTP/2以提高兼容性
                 verify=True,  # 启用SSL验证
                 timeout=httpx.Timeout(
-                    connect=10.0,  # 连接超时
-                    read=300.0,  # 读取超时(5分钟,适合流式响应)
-                    write=60.0,  # 写入超时(60秒,支持大请求体)
-                    pool=5.0,  # 连接池超时
+                    connect=config.http_connect_timeout,
+                    read=config.http_read_timeout,
+                    write=config.http_write_timeout,
+                    pool=config.http_pool_timeout,
                 ),
                 limits=httpx.Limits(
                     max_connections=100,  # 最大连接数
@@ -111,15 +112,20 @@ class HTTPClientPool:
         """
         if name not in cls._clients:
             # 合并默认配置和自定义配置
-            config = {
+            default_config = {
                 "http2": False,
                 "verify": True,
-                "timeout": httpx.Timeout(10.0, read=300.0),
+                "timeout": httpx.Timeout(
+                    connect=config.http_connect_timeout,
+                    read=config.http_read_timeout,
+                    write=config.http_write_timeout,
+                    pool=config.http_pool_timeout,
+                ),
                 "follow_redirects": True,
             }
-            config.update(kwargs)
+            default_config.update(kwargs)
 
-            cls._clients[name] = httpx.AsyncClient(**config)
+            cls._clients[name] = httpx.AsyncClient(**default_config)
             logger.debug(f"创建命名HTTP客户端: {name}")
 
         return cls._clients[name]
@@ -151,14 +157,19 @@ class HTTPClientPool:
             async with HTTPClientPool.get_temp_client() as client:
                 response = await client.get('https://example.com')
         """
-        config = {
+        default_config = {
             "http2": False,
             "verify": True,
-            "timeout": httpx.Timeout(10.0),
+            "timeout": httpx.Timeout(
+                connect=config.http_connect_timeout,
+                read=config.http_read_timeout,
+                write=config.http_write_timeout,
+                pool=config.http_pool_timeout,
+            ),
         }
-        config.update(kwargs)
+        default_config.update(kwargs)
 
-        client = httpx.AsyncClient(**config)
+        client = httpx.AsyncClient(**default_config)
         try:
             yield client
         finally:
@@ -182,25 +193,30 @@ class HTTPClientPool:
         Returns:
             配置好的 httpx.AsyncClient 实例
         """
-        config: Dict[str, Any] = {
+        client_config: Dict[str, Any] = {
             "http2": False,
             "verify": True,
             "follow_redirects": True,
         }
 
         if timeout:
-            config["timeout"] = timeout
+            client_config["timeout"] = timeout
         else:
-            config["timeout"] = httpx.Timeout(10.0, read=300.0)
+            client_config["timeout"] = httpx.Timeout(
+                connect=config.http_connect_timeout,
+                read=config.http_read_timeout,
+                write=config.http_write_timeout,
+                pool=config.http_pool_timeout,
+            )
 
         # 添加代理配置
         proxy_url = build_proxy_url(proxy_config) if proxy_config else None
         if proxy_url:
-            config["proxy"] = proxy_url
+            client_config["proxy"] = proxy_url
             logger.debug(f"创建带代理的HTTP客户端: {proxy_config.get('url', 'unknown')}")
 
-        config.update(kwargs)
-        return httpx.AsyncClient(**config)
+        client_config.update(kwargs)
+        return httpx.AsyncClient(**client_config)
 
 
 # 便捷访问函数

src/config/settings.py

@@ -148,6 +148,7 @@ class Config:
 
         # HTTP 请求超时配置（秒）
         self.http_connect_timeout = float(os.getenv("HTTP_CONNECT_TIMEOUT", "10.0"))
+        self.http_read_timeout = float(os.getenv("HTTP_READ_TIMEOUT", "300.0"))
         self.http_write_timeout = float(os.getenv("HTTP_WRITE_TIMEOUT", "60.0"))
         self.http_pool_timeout = float(os.getenv("HTTP_POOL_TIMEOUT", "10.0"))
 

src/core/logger.py

@@ -96,13 +96,15 @@ if not DISABLE_FILE_LOG:
     log_dir.mkdir(exist_ok=True)
 
     # 文件日志通用配置
+    # 注意: enqueue=False 使用同步模式，避免 multiprocessing 信号量泄漏
+    # 在 macOS 上，进程异常退出时 POSIX 信号量不会自动释放，导致资源耗尽
     file_log_config = {
         "format": FILE_FORMAT,
         "filter": _log_filter,
         "rotation": "100 MB",
         "retention": "30 days",
         "compression": "gz",
-        "enqueue": True,
+        "enqueue": False,
         "encoding": "utf-8",
         "catch": True,
     }

src/database/database.py

@@ -360,6 +360,9 @@ def init_db():
 
     注意：数据库表结构由 Alembic 管理，部署时请运行 ./migrate.sh
     """
+    import sys
+    from sqlalchemy.exc import OperationalError
+
     logger.info("初始化数据库...")
 
     # 确保引擎已创建
@@ -382,6 +385,38 @@ def init_db():
         db.commit()
         logger.info("数据库初始化完成")
 
+    except OperationalError as e:
+        db.rollback()
+        # 提取数据库连接信息用于提示
+        db_url = config.database_url
+        # 隐藏密码，只显示 host:port/database
+        if "@" in db_url:
+            db_info = db_url.split("@")[-1]
+        else:
+            db_info = db_url
+
+        import os
+
+        # 直接打印到 stderr，确保消息显示
+        print("", file=sys.stderr)
+        print("=" * 60, file=sys.stderr)
+        print("数据库连接失败", file=sys.stderr)
+        print("=" * 60, file=sys.stderr)
+        print("", file=sys.stderr)
+        print(f"无法连接到数据库: {db_info}", file=sys.stderr)
+        print("", file=sys.stderr)
+        print("请检查以下事项:", file=sys.stderr)
+        print("  1. PostgreSQL 服务是否正在运行", file=sys.stderr)
+        print("  2. 数据库连接配置是否正确 (DATABASE_URL)", file=sys.stderr)
+        print("  3. 数据库用户名和密码是否正确", file=sys.stderr)
+        print("", file=sys.stderr)
+        print("如果使用 Docker，请先运行:", file=sys.stderr)
+        print("  docker-compose up -d postgres redis", file=sys.stderr)
+        print("", file=sys.stderr)
+        print("=" * 60, file=sys.stderr)
+        # 使用 os._exit 直接退出，避免 uvicorn 捕获并打印堆栈
+        os._exit(1)
+
     except Exception as e:
         logger.error(f"数据库初始化失败: {e}")
         db.rollback()

src/models/pydantic_models.py

@@ -274,6 +274,13 @@ class GlobalModelListResponse(BaseModel):
     total: int
 
 
+class GlobalModelProvidersResponse(BaseModel):
+    """GlobalModel 关联提供商列表响应"""
+
+    providers: List[ModelCatalogProviderDetail]
+    total: int
+
+
 class BatchAssignToProvidersRequest(BaseModel):
     """批量为 Provider 添加 GlobalModel 实现"""
 

src/utils/task_coordinator.py

@@ -1,8 +1,16 @@
-"""分布式任务协调器，确保仅有一个 worker 执行特定任务"""
+"""分布式任务协调器，确保仅有一个 worker 执行特定任务
+
+锁清理策略：
+- 单实例模式（默认）：启动时使用原子操作清理旧锁并获取新锁
+- 多实例模式：使用 NX 选项竞争锁，依赖 TTL 处理异常退出
+
+使用方式：
+- 默认行为：启动时清理旧锁（适用于单机部署）
+- 多实例部署：设置 SINGLE_INSTANCE_MODE=false 禁用启动清理
+"""
 
 from __future__ import annotations
 
-import asyncio
 import os
 import pathlib
 import uuid
@@ -19,6 +27,10 @@ except ImportError:  # pragma: no cover - Windows 环境
 class StartupTaskCoordinator:
     """利用 Redis 或文件锁，保证任务只在单个进程/实例中运行"""
 
+    # 类级别标记：在当前进程中是否已尝试过启动清理
+    # 注意：这在 fork 模式下每个 worker 都是独立的
+    _startup_cleanup_attempted = False
+
     def __init__(self, redis_client=None, lock_dir: Optional[str] = None):
         self.redis = redis_client
         self._tokens: Dict[str, str] = {}
@@ -26,6 +38,8 @@ class StartupTaskCoordinator:
         self._lock_dir = pathlib.Path(lock_dir or os.getenv("TASK_LOCK_DIR", "./.locks"))
         if not self._lock_dir.exists():
             self._lock_dir.mkdir(parents=True, exist_ok=True)
+        # 单实例模式：启动时清理旧锁（适用于单机部署，避免残留锁问题）
+        self._single_instance_mode = os.getenv("SINGLE_INSTANCE_MODE", "true").lower() == "true"
 
     def _redis_key(self, name: str) -> str:
         return f"task_lock:{name}"
@@ -36,7 +50,46 @@ class StartupTaskCoordinator:
         if self.redis:
             token = str(uuid.uuid4())
             try:
-                acquired = await self.redis.set(self._redis_key(name), token, nx=True, ex=ttl)
+                if self._single_instance_mode:
+                    # 单实例模式：使用 Lua 脚本原子性地"清理旧锁 + 竞争获取"
+                    # 只有当锁不存在或成功获取时才返回 1
+                    # 这样第一个执行的 worker 会清理旧锁并获取，后续 worker 会正常竞争
+                    script = """
+                    local key = KEYS[1]
+                    local token = ARGV[1]
+                    local ttl = tonumber(ARGV[2])
+                    local startup_key = KEYS[1] .. ':startup'
+
+                    -- 检查是否已有 worker 执行过启动清理
+                    local cleaned = redis.call('GET', startup_key)
+                    if not cleaned then
+                        -- 第一个 worker：删除旧锁，标记已清理
+                        redis.call('DEL', key)
+                        redis.call('SET', startup_key, '1', 'EX', 60)
+                    end
+
+                    -- 尝试获取锁（NX 模式）
+                    local result = redis.call('SET', key, token, 'NX', 'EX', ttl)
+                    if result then
+                        return 1
+                    end
+                    return 0
+                    """
+                    result = await self.redis.eval(
+                        script, 2,
+                        self._redis_key(name), self._redis_key(name),
+                        token, ttl
+                    )
+                    if result == 1:
+                        self._tokens[name] = token
+                        logger.info(f"任务 {name} 通过 Redis 锁独占执行")
+                        return True
+                    return False
+                else:
+                    # 多实例模式：直接使用 NX 选项竞争锁
+                    acquired = await self.redis.set(
+                        self._redis_key(name), token, nx=True, ex=ttl
+                    )
                     if acquired:
                         self._tokens[name] = token
                         logger.info(f"任务 {name} 通过 Redis 锁独占执行")