plugin-ai-api 1.0.28 → 1.1.1

package/src/locale/zh-CN.json

@@ -1,138 +1,139 @@
-{
-  "AI API Gateway": "AI API 网关",
-  "Configuration": "配置",
-  "Default AI Employee": "默认 AI 员工",
-  "Enabled LLM Services": "已启用 LLM 服务",
-  "Rate Limit": "速率限制",
-  "Save Configuration": "保存配置",
-  "Configuration saved": "配置已保存",
-  "Failed to save configuration": "保存配置失败",
-  "API mode": "API 模式",
-  "Direct LLM": "直接 LLM",
-  "AI Employee agent": "AI 员工代理",
-  "Default LLM service": "默认 LLM 服务",
-  "Enable user quotas": "启用用户配额",
-  "Default reserved output tokens": "默认预留输出令牌数",
-  "Refresh": "刷新",
-  "Model pricing": "模型定价",
-  "User quotas": "用户配额",
-  "Usage": "用量",
-  "LLM service": "LLM 服务",
-  "Model": "模型",
-  "Input price / 1M": "输入价格 / 百万令牌",
-  "Output price / 1M": "输出价格 / 百万令牌",
-  "Fixed request cost": "每次请求固定费用",
-  "Currency": "货币",
-  "Status": "状态",
-  "Enabled": "已启用",
-  "Disabled": "已禁用",
-  "Actions": "操作",
-  "Edit": "编辑",
-  "Delete": "删除",
-  "Delete this price?": "删除此价格？",
-  "Add price": "添加价格",
-  "Edit price": "编辑价格",
-  "Effective from": "生效时间",
-  "Effective to": "失效时间",
-  "Notes": "备注",
-  "Saved successfully": "保存成功",
-  "Deleted successfully": "删除成功",
-  "User": "用户",
-  "Period": "周期",
-  "Request limit": "请求限制",
-  "Token limit": "令牌限制",
-  "Cost limit": "费用限制",
-  "Timezone": "时区",
-  "Unlimited": "无限制",
-  "Add quota": "添加配额",
-  "Edit quota": "编辑配额",
-  "Delete this quota?": "删除此配额？",
-  "Daily": "每日",
-  "Monthly": "每月",
-  "Reject unpriced models": "拒绝未定价模型",
-  "Missing usage behavior": "缺少用量时的行为",
-  "Use reserved estimate": "使用预留估算",
-  "Allow without token charge": "允许且不计令牌",
-  "Context overflow behavior": "上下文超限处理",
-  "Reject request": "拒绝请求",
-  "Truncate oldest conversation turns": "截断最早的对话轮次",
-  "Started at": "开始时间",
-  "Requested model": "请求模型",
-  "Resolved service": "解析后的服务",
-  "Resolved model": "解析后的模型",
-  "Input tokens": "输入令牌",
-  "Output tokens": "输出令牌",
-  "Total tokens": "总令牌",
-  "Prompt cache tokens": "提示缓存令牌",
-  "Cost": "费用",
-  "Cost status": "费用状态",
-  "Request ID": "请求 ID",
-  "Failed to load models": "加载模型失败",
-  "Select a model": "选择模型",
-  "Select an AI Employee": "选择 AI 员工",
-  "Usage guide": "使用指南",
-  "OpenAI-compatible endpoint": "OpenAI 兼容端点",
-  "Base URL": "基础 URL",
-  "Use a NocoBase API key as the Bearer token.": "使用 NocoBase API 密钥作为 Bearer Token。",
-  "List available models": "列出可用模型",
-  "Send a chat completion": "发送聊天补全请求",
-  "Usage filters": "用量筛选",
-  "Time range": "时间范围",
-  "User ID": "用户 ID",
-  "Succeeded": "成功",
-  "Failed": "失败",
-  "Started": "已开始",
-  "Apply filters": "应用筛选",
-  "Reset": "重置",
-  "Requests": "请求数",
-  "Total cost": "总费用",
-  "Usage records": "用量记录",
-  "Model metadata": "模型元数据",
-  "Add override": "添加覆盖",
-  "Edit override": "编辑覆盖",
-  "Delete this override?": "确认删除此覆盖？",
-  "Context window": "上下文窗口",
-  "Max completion tokens": "最大输出 token 数",
-  "Owned by": "所属方",
-  "Display name": "显示名称",
-  "Description": "描述",
-  "Leave empty to not override": "留空则不覆盖",
-  "Total input + output token capacity reported to clients.": "返回给客户端的输入+输出 token 总容量。",
-  "Maximum output tokens reported to clients.": "返回给客户端的最大输出 token 数。",
-  "Initial system prompt": "初始系统提示词",
-  "Prepended as the first system message, before any system prompt sent by the client. If the client sends no system prompt, this becomes the system prompt sent to the provider.": "作为第一条 system 消息插入到客户端发送的任何 system 提示词之前。如果客户端未发送 system 提示词，此提示词将作为发送给提供商的 system 提示词。",
-  "AI API": "AI API",
-  "Allow this role to use the AI API": "允许此角色使用 AI API",
-  "Allow all AI Employees": "允许所有 AI 员工",
-  "Select which AI Employees this role may use:": "选择此角色可使用的 AI 员工：",
-  "Select allowed AI Employees": "选择允许的 AI 员工",
-  "Max request body size (MB)": "请求体大小上限（MB）",
-  "Raise this to accept inline base64 images. Base64 adds about 33% to the original file size.": "调高此值以接收内联 base64 图片。base64 编码会使体积增加约 33%。",
-  "Usage groups": "用量组",
-  "Add group": "添加分组",
-  "Edit group": "编辑分组",
-  "Delete this group?": "删除此分组？",
-  "Mode": "模式",
-  "Share": "共享",
-  "Per user": "按用户",
-  "Rate limit per minute": "每分钟速率限制",
-  "Members": "成员",
-  "Add member": "添加成员",
-  "Member added": "成员已添加",
-  "Member removed": "成员已移除",
-  "Remove member?": "移除成员？",
-  "Remove": "移除",
-  "Search group by user": "按用户搜索分组",
-  "User belongs to": "用户属于",
-  "User not found": "未找到用户",
-  "Default": "默认",
-  "Allowed LLM services": "允许的 LLM 服务",
-  "Allow all models": "允许所有模型",
-  "Allowed models": "允许的模型",
-  "Model access": "模型访问",
-  "All models": "所有模型",
-  "All services": "所有服务",
-  "No models": "未允许任何模型",
-  "Leave empty to allow every service enabled in the general configuration.": "留空则允许通用配置中已启用的所有服务。",
-  "Users who do not belong to any other group automatically use this default group — no need to add members.": "不属于其他分组的用户会自动使用此默认分组，无需手动添加成员。"
-}
+{
+  "AI API Gateway": "AI API 网关",
+  "Configuration": "配置",
+  "Default AI Employee": "默认 AI 员工",
+  "Enabled LLM Services": "已启用 LLM 服务",
+  "Rate Limit": "速率限制",
+  "Save Configuration": "保存配置",
+  "Configuration saved": "配置已保存",
+  "Failed to save configuration": "保存配置失败",
+  "API mode": "API 模式",
+  "Direct LLM": "直接 LLM",
+  "AI Employee agent": "AI 员工代理",
+  "Default LLM service": "默认 LLM 服务",
+  "Enable user quotas": "启用用户配额",
+  "Default reserved output tokens": "默认预留输出令牌数",
+  "Refresh": "刷新",
+  "Model pricing": "模型定价",
+  "User quotas": "用户配额",
+  "Usage": "用量",
+  "LLM service": "LLM 服务",
+  "Model": "模型",
+  "Input price / 1M": "输入价格 / 百万令牌",
+  "Cache input price / 1M": "缓存输入价格 / 百万令牌",
+  "Output price / 1M": "输出价格 / 百万令牌",
+  "Fixed request cost": "每次请求固定费用",
+  "Currency": "货币",
+  "Status": "状态",
+  "Enabled": "已启用",
+  "Disabled": "已禁用",
+  "Actions": "操作",
+  "Edit": "编辑",
+  "Delete": "删除",
+  "Delete this price?": "删除此价格？",
+  "Add price": "添加价格",
+  "Edit price": "编辑价格",
+  "Effective from": "生效时间",
+  "Effective to": "失效时间",
+  "Notes": "备注",
+  "Saved successfully": "保存成功",
+  "Deleted successfully": "删除成功",
+  "User": "用户",
+  "Period": "周期",
+  "Request limit": "请求限制",
+  "Token limit": "令牌限制",
+  "Cost limit": "费用限制",
+  "Timezone": "时区",
+  "Unlimited": "无限制",
+  "Add quota": "添加配额",
+  "Edit quota": "编辑配额",
+  "Delete this quota?": "删除此配额？",
+  "Daily": "每日",
+  "Monthly": "每月",
+  "Reject unpriced models": "拒绝未定价模型",
+  "Missing usage behavior": "缺少用量时的行为",
+  "Use reserved estimate": "使用预留估算",
+  "Allow without token charge": "允许且不计令牌",
+  "Context overflow behavior": "上下文超限处理",
+  "Reject request": "拒绝请求",
+  "Truncate oldest conversation turns": "截断最早的对话轮次",
+  "Started at": "开始时间",
+  "Requested model": "请求模型",
+  "Resolved service": "解析后的服务",
+  "Resolved model": "解析后的模型",
+  "Input tokens": "输入令牌",
+  "Output tokens": "输出令牌",
+  "Total tokens": "总令牌",
+  "Prompt cache tokens": "提示缓存令牌",
+  "Cost": "费用",
+  "Cost status": "费用状态",
+  "Request ID": "请求 ID",
+  "Failed to load models": "加载模型失败",
+  "Select a model": "选择模型",
+  "Select an AI Employee": "选择 AI 员工",
+  "Usage guide": "使用指南",
+  "OpenAI-compatible endpoint": "OpenAI 兼容端点",
+  "Base URL": "基础 URL",
+  "Use a NocoBase API key as the Bearer token.": "使用 NocoBase API 密钥作为 Bearer Token。",
+  "List available models": "列出可用模型",
+  "Send a chat completion": "发送聊天补全请求",
+  "Usage filters": "用量筛选",
+  "Time range": "时间范围",
+  "User ID": "用户 ID",
+  "Succeeded": "成功",
+  "Failed": "失败",
+  "Started": "已开始",
+  "Apply filters": "应用筛选",
+  "Reset": "重置",
+  "Requests": "请求数",
+  "Total cost": "总费用",
+  "Usage records": "用量记录",
+  "Model metadata": "模型元数据",
+  "Add override": "添加覆盖",
+  "Edit override": "编辑覆盖",
+  "Delete this override?": "确认删除此覆盖？",
+  "Context window": "上下文窗口",
+  "Max completion tokens": "最大输出 token 数",
+  "Owned by": "所属方",
+  "Display name": "显示名称",
+  "Description": "描述",
+  "Leave empty to not override": "留空则不覆盖",
+  "Total input + output token capacity reported to clients.": "返回给客户端的输入+输出 token 总容量。",
+  "Maximum output tokens reported to clients.": "返回给客户端的最大输出 token 数。",
+  "Initial system prompt": "初始系统提示词",
+  "Prepended as the first system message, before any system prompt sent by the client. If the client sends no system prompt, this becomes the system prompt sent to the provider.": "作为第一条 system 消息插入到客户端发送的任何 system 提示词之前。如果客户端未发送 system 提示词，此提示词将作为发送给提供商的 system 提示词。",
+  "AI API": "AI API",
+  "Allow this role to use the AI API": "允许此角色使用 AI API",
+  "Allow all AI Employees": "允许所有 AI 员工",
+  "Select which AI Employees this role may use:": "选择此角色可使用的 AI 员工：",
+  "Select allowed AI Employees": "选择允许的 AI 员工",
+  "Max request body size (MB)": "请求体大小上限（MB）",
+  "Raise this to accept inline base64 images. Base64 adds about 33% to the original file size.": "调高此值以接收内联 base64 图片。base64 编码会使体积增加约 33%。",
+  "Usage groups": "用量组",
+  "Add group": "添加分组",
+  "Edit group": "编辑分组",
+  "Delete this group?": "删除此分组？",
+  "Mode": "模式",
+  "Share": "共享",
+  "Per user": "按用户",
+  "Rate limit per minute": "每分钟速率限制",
+  "Members": "成员",
+  "Add member": "添加成员",
+  "Member added": "成员已添加",
+  "Member removed": "成员已移除",
+  "Remove member?": "移除成员？",
+  "Remove": "移除",
+  "Search group by user": "按用户搜索分组",
+  "User belongs to": "用户属于",
+  "User not found": "未找到用户",
+  "Default": "默认",
+  "Allowed LLM services": "允许的 LLM 服务",
+  "Allow all models": "允许所有模型",
+  "Allowed models": "允许的模型",
+  "Model access": "模型访问",
+  "All models": "所有模型",
+  "All services": "所有服务",
+  "No models": "未允许任何模型",
+  "Leave empty to allow every service enabled in the general configuration.": "留空则允许通用配置中已启用的所有服务。",
+  "Users who do not belong to any other group automatically use this default group — no need to add members.": "不属于其他分组的用户会自动使用此默认分组，无需手动添加成员。"
+}

package/src/server/__tests__/billing-quota.test.ts

@@ -23,6 +23,7 @@ describe('AI API group quota reservation', () => {
         { name: 'enabled', type: 'boolean' },
         { name: 'currency', type: 'string' },
         { name: 'inputPricePerMillionTokens', type: 'decimal', precision: 20, scale: 10 },
+        { name: 'cacheInputPricePerMillionTokens', type: 'decimal', precision: 20, scale: 10 },
         { name: 'outputPricePerMillionTokens', type: 'decimal', precision: 20, scale: 10 },
         { name: 'fixedCostPerRequest', type: 'decimal', precision: 20, scale: 10 },
         { name: 'effectiveFrom', type: 'datetimeTz' },
@@ -83,6 +84,7 @@ describe('AI API group quota reservation', () => {
         enabled: true,
         currency: 'USD',
         inputPricePerMillionTokens: '5',
+        cacheInputPricePerMillionTokens: '1',
         outputPricePerMillionTokens: '15',
         fixedCostPerRequest: '0',
         effectiveFrom: new Date('2020-01-01T00:00:00Z'),
@@ -140,10 +142,14 @@ describe('AI API group quota reservation', () => {
     markLlmProviderAttempted(first);
     const finalized = await finalizeLlmBilling(
       first,
-      { prompt_tokens: 10, completion_tokens: 5, total_tokens: 15 },
+      { prompt_tokens: 10, completion_tokens: 5, total_tokens: 15, prompt_cache_tokens: 6 },
       true,
     );
-    expect(finalized).toMatchObject({ estimatedCost: '0.00012500', costStatus: 'calculated' });
+    expect(finalized).toMatchObject({
+      estimatedCost: '0.00010100',
+      costStatus: 'calculated',
+      cacheInputPricePerMillionTokens: '1.0000000000',
+    });
 
     const bucket = await db.getRepository('aiApiGroupQuotaBuckets').findOne();
     expect(String(bucket?.get('requestCount'))).toBe('1');

package/src/server/__tests__/billing.test.ts

@@ -6,6 +6,7 @@ function price(values: Partial<PriceSnapshot> = {}): PriceSnapshot {
     id: 1,
     currency: 'USD',
     inputPricePerMillionTokens: '5.0000000000',
+    cacheInputPricePerMillionTokens: '0.0000000000',
     outputPricePerMillionTokens: '15.0000000000',
     fixedCostPerRequest: '0.0000000000',
     ...values,
@@ -17,6 +18,18 @@ describe('AI API LLM cost calculation', () => {
     expect(calculateLlmCost(10_000, 2_000, price())).toBe('0.08000000');
   });
 
+  it('prices cached prompt tokens separately from uncached input tokens', () => {
+    expect(calculateLlmCost(10_000, 2_000, price({ cacheInputPricePerMillionTokens: '1.0000000000' }), 6_000)).toBe(
+      '0.05600000',
+    );
+  });
+
+  it('does not charge cached tokens beyond the reported input token count', () => {
+    expect(calculateLlmCost(100, 0, price({ cacheInputPricePerMillionTokens: '1.0000000000' }), 200)).toBe(
+      '0.00010000',
+    );
+  });
+
   it('includes a fixed per-request cost', () => {
     expect(calculateLlmCost(0, 0, price({ fixedCostPerRequest: '0.1250000000' }))).toBe('0.12500000');
   });

package/src/server/__tests__/usage.test.ts

@@ -175,6 +175,7 @@ describe('AI API usage persistence', () => {
         inputTokens: 20,
         outputTokens: 4,
         totalTokens: 24,
+        cacheInputPricePerMillionTokens: null,
         providerRequestId: 'provider-id',
         responseMetadata: { usageSource: 'provider', gatewayResponseId: 'gateway-result-id' },
       }),

package/src/server/billing.ts

@@ -24,6 +24,7 @@ export interface PriceSnapshot {
   id: string | number | bigint;
   currency: string;
   inputPricePerMillionTokens: string;
+  cacheInputPricePerMillionTokens: string;
   outputPricePerMillionTokens: string;
   fixedCostPerRequest: string;
 }
@@ -60,6 +61,7 @@ export interface BillingFinalization {
   groupId?: string | number | bigint;
   quotaMode?: QuotaMode;
   inputPricePerMillionTokens?: string;
+  cacheInputPricePerMillionTokens?: string;
   outputPricePerMillionTokens?: string;
   fixedCostPerRequest?: string;
 }
@@ -112,9 +114,20 @@ function divideRounded(value: bigint, divisor: bigint): bigint {
   return (value + divisor / 2n) / divisor;
 }
 
-function calculateCostUnits(inputTokens: number, outputTokens: number, price: PriceSnapshot): bigint {
+function calculateCostUnits(
+  inputTokens: number,
+  outputTokens: number,
+  promptCacheTokens: number,
+  price: PriceSnapshot,
+): bigint {
+  const cachedInputTokens = Math.min(Math.max(promptCacheTokens, 0), inputTokens);
+  const uncachedInputTokens = inputTokens - cachedInputTokens;
   const input = divideRounded(
-    BigInt(inputTokens) * decimalUnits(price.inputPricePerMillionTokens, PRICE_SCALE),
+    BigInt(uncachedInputTokens) * decimalUnits(price.inputPricePerMillionTokens, PRICE_SCALE),
+    PRICE_TO_COST_DIVISOR,
+  );
+  const cacheInput = divideRounded(
+    BigInt(cachedInputTokens) * decimalUnits(price.cacheInputPricePerMillionTokens, PRICE_SCALE),
     PRICE_TO_COST_DIVISOR,
   );
   const output = divideRounded(
@@ -122,11 +135,16 @@ function calculateCostUnits(inputTokens: number, outputTokens: number, price: Pr
     PRICE_TO_COST_DIVISOR,
   );
   const fixed = divideRounded(decimalUnits(price.fixedCostPerRequest, PRICE_SCALE), 100n);
-  return input + output + fixed;
+  return input + cacheInput + output + fixed;
 }
 
-export function calculateLlmCost(inputTokens: number, outputTokens: number, price: PriceSnapshot): string {
-  return formatUnits(calculateCostUnits(inputTokens, outputTokens, price), COST_SCALE);
+export function calculateLlmCost(
+  inputTokens: number,
+  outputTokens: number,
+  price: PriceSnapshot,
+  promptCacheTokens = 0,
+): string {
+  return formatUnits(calculateCostUnits(inputTokens, outputTokens, promptCacheTokens, price), COST_SCALE);
 }
 
 function normalizePositiveInteger(value: unknown, fallback: number): number {
@@ -184,6 +202,7 @@ async function findPrice(
     id: valueOf(price, 'id'),
     currency: valueOf<string>(price, 'currency'),
     inputPricePerMillionTokens: decimalString(valueOf(price, 'inputPricePerMillionTokens'), PRICE_SCALE),
+    cacheInputPricePerMillionTokens: decimalString(valueOf(price, 'cacheInputPricePerMillionTokens'), PRICE_SCALE),
     outputPricePerMillionTokens: decimalString(valueOf(price, 'outputPricePerMillionTokens'), PRICE_SCALE),
     fixedCostPerRequest: decimalString(valueOf(price, 'fixedCostPerRequest'), PRICE_SCALE),
   };
@@ -325,7 +344,10 @@ export async function finalizeLlmBilling(
     costStatus = billing.price ? 'usage_unavailable' : 'unpriced';
   }
 
-  const cost = numbers && billing.price ? calculateLlmCost(numbers.input, numbers.output, billing.price) : undefined;
+  const cost =
+    numbers && billing.price
+      ? calculateLlmCost(numbers.input, numbers.output, billing.price, providerUsage?.prompt_cache_tokens ?? 0)
+      : undefined;
   const reservation = billing.reservation;
   if (reservation) {
     const Bucket = ctx.db.getModel('aiApiGroupQuotaBuckets');
@@ -378,6 +400,7 @@ export async function finalizeLlmBilling(
     groupId: reservation?.groupId,
     quotaMode: reservation?.quotaMode,
     inputPricePerMillionTokens: billing.price?.inputPricePerMillionTokens,
+    cacheInputPricePerMillionTokens: billing.price?.cacheInputPricePerMillionTokens,
     outputPricePerMillionTokens: billing.price?.outputPricePerMillionTokens,
     fixedCostPerRequest: billing.price?.fixedCostPerRequest,
   };

package/src/server/collections/ai-api-model-prices.ts

@@ -9,6 +9,14 @@ export default defineCollection({
     { name: 'model', type: 'string', allowNull: false, index: true },
     { name: 'currency', type: 'string', allowNull: false, defaultValue: 'USD' },
     { name: 'inputPricePerMillionTokens', type: 'decimal', precision: 20, scale: 10, allowNull: false },
+    {
+      name: 'cacheInputPricePerMillionTokens',
+      type: 'decimal',
+      precision: 20,
+      scale: 10,
+      allowNull: false,
+      defaultValue: 0,
+    },
     { name: 'outputPricePerMillionTokens', type: 'decimal', precision: 20, scale: 10, allowNull: false },
     { name: 'fixedCostPerRequest', type: 'decimal', precision: 20, scale: 10, allowNull: false, defaultValue: 0 },
     { name: 'effectiveFrom', type: 'datetimeTz', allowNull: false, index: true },

package/src/server/collections/ai-api-usage-records.ts

@@ -40,6 +40,7 @@ export default defineCollection({
     { name: 'quotaPolicyId', type: 'bigInt', allowNull: true, index: true },
     { name: 'groupId', type: 'bigInt', allowNull: true, index: true },
     { name: 'inputPricePerMillionTokens', type: 'decimal', allowNull: true, precision: 20, scale: 10 },
+    { name: 'cacheInputPricePerMillionTokens', type: 'decimal', allowNull: true, precision: 20, scale: 10 },
     { name: 'outputPricePerMillionTokens', type: 'decimal', allowNull: true, precision: 20, scale: 10 },
     { name: 'fixedCostPerRequest', type: 'decimal', allowNull: true, precision: 20, scale: 10 },
     { name: 'providerRequestId', type: 'string', allowNull: true },

package/src/server/migrations/20260827000000-add-cache-input-price.ts

@@ -0,0 +1,49 @@
+import { Migration } from '@nocobase/server';
+
+export default class AddCacheInputPriceToModelPrices extends Migration {
+  on = 'beforeLoad' as const;
+
+  async up() {
+    await this.addDecimalColumn('aiApiModelPrices', 'cacheInputPricePerMillionTokens', false, 0);
+    await this.addDecimalColumn('aiApiUsageRecords', 'cacheInputPricePerMillionTokens', true);
+  }
+
+  async down() {
+    await this.removeColumn('aiApiUsageRecords', 'cacheInputPricePerMillionTokens');
+    await this.removeColumn('aiApiModelPrices', 'cacheInputPricePerMillionTokens');
+  }
+
+  private async addDecimalColumn(
+    collectionName: string,
+    columnName: string,
+    allowNull: boolean,
+    defaultValue?: number,
+  ): Promise<void> {
+    const collection = this.db.getCollection(collectionName);
+    if (!collection || !(await collection.existsInDb())) return;
+
+    const tableName = collection.getTableNameWithSchema();
+    if (await this.columnExists(tableName, columnName)) return;
+
+    await this.queryInterface.addColumn(tableName, columnName, {
+      type: 'DECIMAL(20,10)',
+      allowNull,
+      ...(defaultValue === undefined ? {} : { defaultValue }),
+    });
+  }
+
+  private async removeColumn(collectionName: string, columnName: string): Promise<void> {
+    const collection = this.db.getCollection(collectionName);
+    if (!collection || !(await collection.existsInDb())) return;
+
+    const tableName = collection.getTableNameWithSchema();
+    if (await this.columnExists(tableName, columnName)) {
+      await this.queryInterface.removeColumn(tableName, columnName);
+    }
+  }
+
+  private async columnExists(tableName: string, columnName: string): Promise<boolean> {
+    const columns = await this.queryInterface.describeTable(tableName);
+    return Object.prototype.hasOwnProperty.call(columns, columnName);
+  }
+}

package/src/server/routes/auth.ts

@@ -75,7 +75,7 @@ export async function authenticateBearer(ctx: Context): Promise<boolean> {
       return false;
     }
 
-    let decoded: { userId?: string | number; roleName?: string; temp?: boolean };
+    let decoded: { userId?: string | number; roleName?: string; temp?: boolean; jti?: string };
     try {
       decoded = await jwt.decode(token);
     } catch (e) {
@@ -90,6 +90,26 @@ export async function authenticateBearer(ctx: Context): Promise<boolean> {
       return false;
     }
 
+    // Enforce the token blacklist here as well. Revoked API keys are blocked by
+    // the core NocoBase auth path (Auth.checkToken), but this API-key fallback
+    // decodes the JWT directly and would otherwise keep accepting them.
+    if (jwt.blacklist) {
+      let blocked = false;
+      try {
+        blocked = await jwt.blacklist.has(decoded.jti ?? token);
+      } catch (err) {
+        ctx.log.error('AI API token blacklist check failed:', err);
+        ctx.status = 401;
+        ctx.body = toOpenAIError(401, 'Unable to verify API key status', 'server_error');
+        return false;
+      }
+      if (blocked) {
+        ctx.status = 401;
+        ctx.body = toOpenAIError(401, 'API key has been revoked', 'invalid_request_error', 'invalid_api_key');
+        return false;
+      }
+    }
+
     // This fallback is for API keys only. A normal login/OIDC token should have
     // been resolved by NocoBase auth middleware and must not be treated as the
     // member role merely because it has no roleName claim.

package/src/server/routes/embeddings.ts

@@ -12,8 +12,9 @@ import { toOpenAIError, toOpenAIEmbeddingsResponse } from '../utils/openai-forma
 import { resolveModelString } from '../utils/resolve-service';
 import { enforceModelAccess } from '../utils/user-permissions';
 import { getAiApiConfig } from '../utils/request-cache';
-import { setAiApiUsageUnavailable } from '../usage';
+import { setAiApiUsageResult } from '../usage';
 import type PluginAiApiServer from '../plugin';
+import { markLlmProviderAttempted, prepareLlmBilling, finalizeLlmBilling } from '../billing';
 
 /**
  * POST /api/ai-llm/v1/embeddings
@@ -103,6 +104,9 @@ export async function handleEmbeddings(ctx: Context, plugin: PluginAiApiServer)
 
   const { service, modelId } = resolved;
 
+  // ─── Prepare billing/quota ────────────────────────────────────────────────
+  await prepareLlmBilling(ctx, resolved);
+
   if (service.enabled === false) {
     ctx.status = 404;
     ctx.body = toOpenAIError(
@@ -168,11 +172,21 @@ export async function handleEmbeddings(ctx: Context, plugin: PluginAiApiServer)
     // createEmbedding() returns a LangChain EmbeddingsInterface.
     // embedDocuments() accepts string[] and returns number[][] (one vector per input).
     const embeddingModel = embeddingProvider.createEmbedding();
+    markLlmProviderAttempted(ctx);
     const vectors: number[][] = await embeddingModel.embedDocuments(inputs);
 
+    // Estimate input tokens: average 4 chars per token (conservative).
+    const estimatedInputTokens = Math.ceil(inputs.reduce((sum, s) => sum + s.length, 0) / 4);
+    const usage = {
+      prompt_tokens: estimatedInputTokens,
+      completion_tokens: 0,
+      total_tokens: estimatedInputTokens,
+    };
+    setAiApiUsageResult(ctx, usage);
+    await finalizeLlmBilling(ctx, usage, true);
+
     ctx.status = 200;
     ctx.set('Content-Type', 'application/json');
-    setAiApiUsageUnavailable(ctx);
     ctx.body = toOpenAIEmbeddingsResponse({
       model: body.model,
       embeddings: vectors,
@@ -186,4 +200,4 @@ export async function handleEmbeddings(ctx: Context, plugin: PluginAiApiServer)
       ctx.body = toOpenAIError(500, err.message || 'Failed to generate embeddings', 'server_error');
     }
   }
-}
+}

package/src/server/usage.ts

@@ -233,6 +233,7 @@ export async function finishUsageRecord(ctx: Context, id: unknown, startedAt: nu
     quotaPolicyId: billing.quotaPolicyId ?? null,
     groupId: billing.groupId ?? null,
     inputPricePerMillionTokens: billing.inputPricePerMillionTokens ?? null,
+    cacheInputPricePerMillionTokens: billing.cacheInputPricePerMillionTokens ?? null,
     outputPricePerMillionTokens: billing.outputPricePerMillionTokens ?? null,
     fixedCostPerRequest: billing.fixedCostPerRequest ?? null,
     providerRequestId: usageResult.providerRequestId ?? null,