plugin-ai-api 1.1.0 → 1.1.1

package/dist/server/billing.js

@@ -86,9 +86,15 @@ function formatUnits(value, scale) {
 function divideRounded(value, divisor) {
   return (value + divisor / 2n) / divisor;
 }
-function calculateCostUnits(inputTokens, outputTokens, price) {
+function calculateCostUnits(inputTokens, outputTokens, promptCacheTokens, price) {
+  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(
@@ -96,10 +102,10 @@ function calculateCostUnits(inputTokens, outputTokens, price) {
     PRICE_TO_COST_DIVISOR
   );
   const fixed = divideRounded(decimalUnits(price.fixedCostPerRequest, PRICE_SCALE), 100n);
-  return input + output + fixed;
+  return input + cacheInput + output + fixed;
 }
-function calculateLlmCost(inputTokens, outputTokens, price) {
-  return formatUnits(calculateCostUnits(inputTokens, outputTokens, price), COST_SCALE);
+function calculateLlmCost(inputTokens, outputTokens, price, promptCacheTokens = 0) {
+  return formatUnits(calculateCostUnits(inputTokens, outputTokens, promptCacheTokens, price), COST_SCALE);
 }
 function normalizePositiveInteger(value, fallback) {
   const parsed = Number(value);
@@ -145,6 +151,7 @@ async function findPrice(ctx, service, modelId) {
     id: valueOf(price, "id"),
     currency: valueOf(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)
   };
@@ -252,7 +259,7 @@ function usageNumbers(usage) {
   };
 }
 async function finalizeLlmBilling(ctx, providerUsage, succeeded) {
-  var _a, _b, _c, _d, _e, _f;
+  var _a, _b, _c, _d, _e, _f, _g;
   const billing = stateOf(ctx).aiApiLlmBilling;
   if (!billing) return {};
   let numbers = usageNumbers(providerUsage);
@@ -269,7 +276,7 @@ async function finalizeLlmBilling(ctx, providerUsage, succeeded) {
   } else {
     costStatus = billing.price ? "usage_unavailable" : "unpriced";
   }
-  const cost = numbers && billing.price ? calculateLlmCost(numbers.input, numbers.output, billing.price) : void 0;
+  const cost = numbers && billing.price ? calculateLlmCost(numbers.input, numbers.output, billing.price, (providerUsage == null ? void 0 : providerUsage.prompt_cache_tokens) ?? 0) : void 0;
   const reservation = billing.reservation;
   if (reservation) {
     const Bucket = ctx.db.getModel("aiApiGroupQuotaBuckets");
@@ -315,8 +322,9 @@ async function finalizeLlmBilling(ctx, providerUsage, succeeded) {
     groupId: reservation == null ? void 0 : reservation.groupId,
     quotaMode: reservation == null ? void 0 : reservation.quotaMode,
     inputPricePerMillionTokens: (_d = billing.price) == null ? void 0 : _d.inputPricePerMillionTokens,
-    outputPricePerMillionTokens: (_e = billing.price) == null ? void 0 : _e.outputPricePerMillionTokens,
-    fixedCostPerRequest: (_f = billing.price) == null ? void 0 : _f.fixedCostPerRequest
+    cacheInputPricePerMillionTokens: (_e = billing.price) == null ? void 0 : _e.cacheInputPricePerMillionTokens,
+    outputPricePerMillionTokens: (_f = billing.price) == null ? void 0 : _f.outputPricePerMillionTokens,
+    fixedCostPerRequest: (_g = billing.price) == null ? void 0 : _g.fixedCostPerRequest
   };
 }
 // Annotate the CommonJS export names for ESM import in node:

package/dist/server/collections/ai-api-model-prices.js

@@ -39,6 +39,14 @@ var ai_api_model_prices_default = (0, import_database.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/dist/server/collections/ai-api-usage-records.js

@@ -70,6 +70,7 @@ var ai_api_usage_records_default = (0, import_database.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/dist/server/migrations/20260827000000-add-cache-input-price.js

@@ -0,0 +1,66 @@
+/**
+ * This file is part of the NocoBase (R) project.
+ * Copyright (c) 2020-2024 NocoBase Co., Ltd.
+ * Authors: NocoBase Team.
+ *
+ * This project is dual-licensed under AGPL-3.0 and NocoBase Commercial License.
+ * For more information, please refer to: https://www.nocobase.com/agreement.
+ */
+
+var __defProp = Object.defineProperty;
+var __getOwnPropDesc = Object.getOwnPropertyDescriptor;
+var __getOwnPropNames = Object.getOwnPropertyNames;
+var __hasOwnProp = Object.prototype.hasOwnProperty;
+var __export = (target, all) => {
+  for (var name in all)
+    __defProp(target, name, { get: all[name], enumerable: true });
+};
+var __copyProps = (to, from, except, desc) => {
+  if (from && typeof from === "object" || typeof from === "function") {
+    for (let key of __getOwnPropNames(from))
+      if (!__hasOwnProp.call(to, key) && key !== except)
+        __defProp(to, key, { get: () => from[key], enumerable: !(desc = __getOwnPropDesc(from, key)) || desc.enumerable });
+  }
+  return to;
+};
+var __toCommonJS = (mod) => __copyProps(__defProp({}, "__esModule", { value: true }), mod);
+var add_cache_input_price_exports = {};
+__export(add_cache_input_price_exports, {
+  default: () => AddCacheInputPriceToModelPrices
+});
+module.exports = __toCommonJS(add_cache_input_price_exports);
+var import_server = require("@nocobase/server");
+class AddCacheInputPriceToModelPrices extends import_server.Migration {
+  on = "beforeLoad";
+  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");
+  }
+  async addDecimalColumn(collectionName, columnName, allowNull, defaultValue) {
+    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 === void 0 ? {} : { defaultValue }
+    });
+  }
+  async removeColumn(collectionName, columnName) {
+    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);
+    }
+  }
+  async columnExists(tableName, columnName) {
+    const columns = await this.queryInterface.describeTable(tableName);
+    return Object.prototype.hasOwnProperty.call(columns, columnName);
+  }
+}

package/dist/server/routes/auth.js

@@ -91,6 +91,22 @@ async function authenticateBearer(ctx) {
       ctx.body = (0, import_openai_format.toOpenAIError)(401, "Invalid or expired API key", "invalid_request_error", "invalid_api_key");
       return false;
     }
+    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 = (0, import_openai_format.toOpenAIError)(401, "Unable to verify API key status", "server_error");
+        return false;
+      }
+      if (blocked) {
+        ctx.status = 401;
+        ctx.body = (0, import_openai_format.toOpenAIError)(401, "API key has been revoked", "invalid_request_error", "invalid_api_key");
+        return false;
+      }
+    }
     if (!decoded.roleName) {
       ctx.status = 401;
       ctx.body = (0, import_openai_format.toOpenAIError)(

package/dist/server/routes/embeddings.js

@@ -34,6 +34,7 @@ var import_resolve_service = require("../utils/resolve-service");
 var import_user_permissions = require("../utils/user-permissions");
 var import_request_cache = require("../utils/request-cache");
 var import_usage = require("../usage");
+var import_billing = require("../billing");
 async function handleEmbeddings(ctx, plugin) {
   const body = ctx.request.body;
   if (!(body == null ? void 0 : body.model)) {
@@ -93,6 +94,7 @@ async function handleEmbeddings(ctx, plugin) {
     return;
   }
   const { service, modelId } = resolved;
+  await (0, import_billing.prepareLlmBilling)(ctx, resolved);
   if (service.enabled === false) {
     ctx.status = 404;
     ctx.body = (0, import_openai_format.toOpenAIError)(
@@ -144,10 +146,18 @@ async function handleEmbeddings(ctx, plugin) {
       // The specific embedding model
     });
     const embeddingModel = embeddingProvider.createEmbedding();
+    (0, import_billing.markLlmProviderAttempted)(ctx);
     const vectors = await embeddingModel.embedDocuments(inputs);
+    const estimatedInputTokens = Math.ceil(inputs.reduce((sum, s) => sum + s.length, 0) / 4);
+    const usage = {
+      prompt_tokens: estimatedInputTokens,
+      completion_tokens: 0,
+      total_tokens: estimatedInputTokens
+    };
+    (0, import_usage.setAiApiUsageResult)(ctx, usage);
+    await (0, import_billing.finalizeLlmBilling)(ctx, usage, true);
     ctx.status = 200;
     ctx.set("Content-Type", "application/json");
-    (0, import_usage.setAiApiUsageUnavailable)(ctx);
     ctx.body = (0, import_openai_format.toOpenAIEmbeddingsResponse)({
       model: body.model,
       embeddings: vectors,

package/dist/server/usage.js

@@ -176,6 +176,7 @@ async function finishUsageRecord(ctx, id, startedAt, status) {
     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,

package/dist/server/utils/direct-llm-context.js

@@ -26,6 +26,8 @@ var __copyProps = (to, from, except, desc) => {
 var __toCommonJS = (mod) => __copyProps(__defProp({}, "__esModule", { value: true }), mod);
 var direct_llm_context_exports = {};
 __export(direct_llm_context_exports, {
+  DEFAULT_CONTEXT_WINDOW: () => DEFAULT_CONTEXT_WINDOW,
+  DEFAULT_MAX_COMPLETION_TOKENS: () => DEFAULT_MAX_COMPLETION_TOKENS,
   DirectLlmContextError: () => DirectLlmContextError,
   parseImageDimensions: () => parseImageDimensions,
   prepareDirectLlmContext: () => prepareDirectLlmContext
@@ -226,18 +228,15 @@ function messagesWithTurns(messages, turns) {
   const retainedMessages = new Set(turns.flat());
   return messages.filter((message) => isInstruction(message) || retainedMessages.has(message));
 }
+const DEFAULT_CONTEXT_WINDOW = 128e3;
+const DEFAULT_MAX_COMPLETION_TOKENS = 8192;
 async function loadModelMetadata(ctx, serviceName, modelId) {
   const row = await ctx.db.getRepository("aiApiModelMetadata").findOne({
     filter: { llmService: serviceName, model: modelId, enabled: true }
   });
-  const contextWindow = positiveInteger(getValue(row, "contextWindow"));
-  const maxCompletionTokens = positiveInteger(getValue(row, "maxCompletionTokens"));
-  if (!contextWindow || !maxCompletionTokens) {
-    throw new DirectLlmContextError(
-      "model_context_metadata_not_configured",
-      `Context metadata is not configured for '${serviceName}/${modelId}'. Configure context window and max completion tokens.`
-    );
-  }
+  const config = await (0, import_request_cache.getAiApiConfig)(ctx);
+  const contextWindow = positiveInteger(getValue(row, "contextWindow")) ?? positiveInteger(getValue(config, "defaultContextWindow")) ?? DEFAULT_CONTEXT_WINDOW;
+  const maxCompletionTokens = positiveInteger(getValue(row, "maxCompletionTokens")) ?? positiveInteger(getValue(config, "defaultMaxCompletionTokens")) ?? DEFAULT_MAX_COMPLETION_TOKENS;
   const systemPromptValue = getValue(row, "systemPrompt");
   const systemPrompt = typeof systemPromptValue === "string" ? systemPromptValue.trim() : "";
   return { contextWindow, maxCompletionTokens, ...systemPrompt ? { systemPrompt } : {} };
@@ -315,6 +314,8 @@ async function prepareDirectLlmContext(ctx, options) {
 }
 // Annotate the CommonJS export names for ESM import in node:
 0 && (module.exports = {
+  DEFAULT_CONTEXT_WINDOW,
+  DEFAULT_MAX_COMPLETION_TOKENS,
   DirectLlmContextError,
   parseImageDimensions,
   prepareDirectLlmContext

package/dist/server/validation.js

@@ -55,6 +55,7 @@ function requireNonNegativeIntegerOrNull(value, field) {
 }
 async function validateModelPrice(db, model) {
   requireNonNegativeDecimal(model.get("inputPricePerMillionTokens"), "inputPricePerMillionTokens");
+  requireNonNegativeDecimal(model.get("cacheInputPricePerMillionTokens") ?? 0, "cacheInputPricePerMillionTokens");
   requireNonNegativeDecimal(model.get("outputPricePerMillionTokens"), "outputPricePerMillionTokens");
   requireNonNegativeDecimal(model.get("fixedCostPerRequest") ?? 0, "fixedCostPerRequest");
   const effectiveFrom = new Date(String(model.get("effectiveFrom")));

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "plugin-ai-api",
-  "version": "1.1.0",
+  "version": "1.1.1",
   "main": "dist/server/index.js",
   "dependencies": {},
   "peerDependencies": {

package/src/client/index.tsx

@@ -1,10 +1,10 @@
-/**
- * This file is part of the NocoBase (R) project.
- * Copyright (c) 2020-2024 NocoBase Co., Ltd.
- * Authors: NocoBase Team.
- *
- * This project is dual-licensed under AGPL-3.0 and NocoBase Commercial License.
- * For more information, please refer to: https://www.nocobase.com/agreement.
- */
-
-export { default } from './plugin';
+/**
+ * This file is part of the NocoBase (R) project.
+ * Copyright (c) 2020-2024 NocoBase Co., Ltd.
+ * Authors: NocoBase Team.
+ *
+ * This project is dual-licensed under AGPL-3.0 and NocoBase Commercial License.
+ * For more information, please refer to: https://www.nocobase.com/agreement.
+ */
+
+export { default } from './plugin';

package/src/client/models/index.ts

@@ -1,12 +1,12 @@
-/**
- * This file is part of the NocoBase (R) project.
- * Copyright (c) 2020-2024 NocoBase Co., Ltd.
- * Authors: NocoBase Team.
- *
- * This project is dual-licensed under AGPL-3.0 and NocoBase Commercial License.
- * For more information, please refer to: https://www.nocobase.com/agreement.
- */
-
-import { ModelConstructor } from '@nocobase/flow-engine';
-
-export default {} as Record<string, ModelConstructor>;
+/**
+ * This file is part of the NocoBase (R) project.
+ * Copyright (c) 2020-2024 NocoBase Co., Ltd.
+ * Authors: NocoBase Team.
+ *
+ * This project is dual-licensed under AGPL-3.0 and NocoBase Commercial License.
+ * For more information, please refer to: https://www.nocobase.com/agreement.
+ */
+
+import { ModelConstructor } from '@nocobase/flow-engine';
+
+export default {} as Record<string, ModelConstructor>;

package/src/client-v2/pages/ModelPricingPage.tsx

@@ -28,6 +28,7 @@ interface ModelPrice {
   model: string;
   currency: string;
   inputPricePerMillionTokens: string;
+  cacheInputPricePerMillionTokens: string;
   outputPricePerMillionTokens: string;
   fixedCostPerRequest: string;
   effectiveFrom: string;
@@ -95,6 +96,7 @@ export default function ModelPricingPage() {
     form.setFieldsValue({
       currency: 'USD',
       inputPricePerMillionTokens: '0',
+      cacheInputPricePerMillionTokens: '0',
       outputPricePerMillionTokens: '0',
       fixedCostPerRequest: '0',
       effectiveFrom: dayjs(),
@@ -182,6 +184,12 @@ export default function ModelPricingPage() {
     { title: t('LLM service'), dataIndex: 'llmService', key: 'llmService', width: 180 },
     { title: t('Model'), dataIndex: 'model', key: 'model', width: 180 },
     { title: t('Input price / 1M'), dataIndex: 'inputPricePerMillionTokens', key: 'inputPrice', width: 150 },
+    {
+      title: t('Cache input price / 1M'),
+      dataIndex: 'cacheInputPricePerMillionTokens',
+      key: 'cacheInputPrice',
+      width: 170,
+    },
     { title: t('Output price / 1M'), dataIndex: 'outputPricePerMillionTokens', key: 'outputPrice', width: 150 },
     { title: t('Fixed request cost'), dataIndex: 'fixedCostPerRequest', key: 'fixedCost', width: 150 },
     { title: t('Currency'), dataIndex: 'currency', key: 'currency', width: 90 },
@@ -260,6 +268,13 @@ export default function ModelPricingPage() {
           <Form.Item name="inputPricePerMillionTokens" label={t('Input price / 1M')} rules={[{ required: true }]}>
             <InputNumber min={0} stringMode style={{ width: '100%' }} />
           </Form.Item>
+          <Form.Item
+            name="cacheInputPricePerMillionTokens"
+            label={t('Cache input price / 1M')}
+            rules={[{ required: true }]}
+          >
+            <InputNumber min={0} stringMode style={{ width: '100%' }} />
+          </Form.Item>
           <Form.Item name="outputPricePerMillionTokens" label={t('Output price / 1M')} rules={[{ required: true }]}>
             <InputNumber min={0} stringMode style={{ width: '100%' }} />
           </Form.Item>

package/src/index.ts

@@ -1,11 +1,11 @@
-/**
- * This file is part of the NocoBase (R) project.
- * Copyright (c) 2020-2024 NocoBase Co., Ltd.
- * Authors: NocoBase Team.
- *
- * This project is dual-licensed under AGPL-3.0 and NocoBase Commercial License.
- * For more information, please refer to: https://www.nocobase.com/agreement.
- */
-
-export * from './server';
-export { default } from './server';
+/**
+ * This file is part of the NocoBase (R) project.
+ * Copyright (c) 2020-2024 NocoBase Co., Ltd.
+ * Authors: NocoBase Team.
+ *
+ * This project is dual-licensed under AGPL-3.0 and NocoBase Commercial License.
+ * For more information, please refer to: https://www.nocobase.com/agreement.
+ */
+
+export * from './server';
+export { default } from './server';