plugin-ai-api 1.0.25 → 1.0.28

package/dist/server/routes/completions.js

@@ -34,6 +34,7 @@ var import_resolve_service = require("../utils/resolve-service");
 var import_user_permissions = require("../utils/user-permissions");
 var import_streaming = require("../utils/streaming");
 var import_chat_completions = require("./chat-completions");
+var import_request_cache = require("../utils/request-cache");
 var import_usage = require("../usage");
 var import_billing = require("../billing");
 var import_direct_llm_context = require("../utils/direct-llm-context");
@@ -91,7 +92,7 @@ async function handleCompletions(ctx, plugin) {
       );
       return;
     }
-    const config = await ctx.db.getRepository("aiApiConfig").findOne();
+    const config = await (0, import_request_cache.getAiApiConfig)(ctx);
     if (!await (0, import_user_permissions.enforceModelAccess)(ctx, config == null ? void 0 : config.enabledLlmServices, service, modelId)) {
       return;
     }
@@ -110,19 +111,7 @@ async function handleCompletions(ctx, plugin) {
     if (body.max_tokens !== void 0) modelOptions.maxTokens = body.max_tokens;
     if (body.stop !== void 0) modelOptions.stop = body.stop;
     const prompt = typeof body.prompt === "string" ? body.prompt : Array.isArray(body.prompt) ? body.prompt.join("\n") : String(body.prompt);
-    const messages = [];
-    if (config == null ? void 0 : config.defaultAiEmployee) {
-      const employee = await ctx.db.getRepository("aiEmployees").findOne({
-        filter: { username: config.defaultAiEmployee }
-      });
-      if (employee) {
-        const systemPrompt = employee.about || employee.defaultPrompt || "";
-        if (systemPrompt) {
-          messages.push({ role: "system", content: systemPrompt });
-        }
-      }
-    }
-    messages.push({ role: "user", content: prompt });
+    const messages = [{ role: "user", content: prompt }];
     const preparedContext = await (0, import_direct_llm_context.prepareDirectLlmContext)(ctx, {
       serviceName: service.name,
       modelId,
@@ -211,7 +200,7 @@ async function handleNonStreamingTextCompletion(ctx, chatModel, messages, comple
         text,
         index: 0,
         logprobs: null,
-        finish_reason: "stop"
+        finish_reason: (0, import_chat_completions.extractFinishReason)(result) ?? "stop"
       }
     ],
     usage: usage ? {
@@ -232,7 +221,9 @@ async function handleStreamingTextCompletion(ctx, chatModel, messages, completio
   ctx.status = 200;
   const requestAbort = (0, import_streaming.createRequestAbortController)(ctx);
   let usage;
+  let usageResponseMetadata;
   let providerRequestId;
+  let providerFinishReason;
   try {
     const stream = await chatModel.stream(messages, {
       ...providerRequestParameters,
@@ -271,8 +262,14 @@ async function handleStreamingTextCompletion(ctx, chatModel, messages, completio
         );
       }
       if (chunk.usage_metadata) {
-        usage = (0, import_usage.normalizeUsage)(chunk.usage_metadata) ?? usage;
+        const normalized = (0, import_usage.normalizeUsage)(chunk.usage_metadata);
+        if (normalized) {
+          usage = normalized;
+          usageResponseMetadata = chunk.response_metadata;
+        }
       }
+      const chunkFinishReason = (0, import_chat_completions.extractFinishReason)(chunk);
+      if (chunkFinishReason) providerFinishReason = chunkFinishReason;
       providerRequestId = providerRequestId ?? (0, import_usage.extractProviderRequestId)(chunk);
     }
     await (0, import_streaming.writeResponse)(
@@ -288,7 +285,7 @@ async function handleStreamingTextCompletion(ctx, chatModel, messages, completio
             text: "",
             index: 0,
             logprobs: null,
-            finish_reason: "stop"
+            finish_reason: providerFinishReason ?? "stop"
           }
         ],
         usage: null
@@ -308,7 +305,7 @@ async function handleStreamingTextCompletion(ctx, chatModel, messages, completio
       );
     }
     await (0, import_streaming.writeResponse)(ctx, (0, import_openai_format.formatSSEDone)());
-    (0, import_usage.setAiApiUsageResult)(ctx, usage, { gatewayResponseId: completionId, providerRequestId });
+    (0, import_usage.setAiApiUsageResult)(ctx, usage, { gatewayResponseId: completionId, providerRequestId }, usageResponseMetadata);
     ctx.state.aiApiStreamResult = { succeeded: true, id: completionId };
   } catch (err) {
     const cancelled = (0, import_streaming.isClientDisconnected)(ctx, err);
@@ -324,7 +321,7 @@ async function handleStreamingTextCompletion(ctx, chatModel, messages, completio
         })
       );
     }
-    (0, import_usage.setAiApiUsageResult)(ctx, usage, { gatewayResponseId: completionId, providerRequestId });
+    (0, import_usage.setAiApiUsageResult)(ctx, usage, { gatewayResponseId: completionId, providerRequestId }, usageResponseMetadata);
     ctx.state.aiApiStreamResult = {
       succeeded: false,
       id: completionId,

package/dist/server/routes/embeddings.js

@@ -32,6 +32,7 @@ module.exports = __toCommonJS(embeddings_exports);
 var import_openai_format = require("../utils/openai-format");
 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");
 async function handleEmbeddings(ctx, plugin) {
   const body = ctx.request.body;
@@ -104,7 +105,7 @@ async function handleEmbeddings(ctx, plugin) {
   }
   let globalEnabledServices = [];
   try {
-    const config = await ctx.db.getRepository("aiApiConfig").findOne();
+    const config = await (0, import_request_cache.getAiApiConfig)(ctx);
     globalEnabledServices = (config == null ? void 0 : config.enabledLlmServices) ?? [];
   } catch {
   }

package/dist/server/routes/models.js

@@ -33,6 +33,7 @@ __export(models_exports, {
 module.exports = __toCommonJS(models_exports);
 var import_openai_format = require("../utils/openai-format");
 var import_user_permissions = require("../utils/user-permissions");
+var import_request_cache = require("../utils/request-cache");
 async function handleListModels(ctx, plugin) {
   var _a;
   try {
@@ -198,7 +199,7 @@ function toPositiveInt(value) {
   return Number.isSafeInteger(n) && n > 0 ? n : null;
 }
 async function getPluginConfig(ctx) {
-  return ctx.db.getRepository("aiApiConfig").findOne();
+  return (0, import_request_cache.getAiApiConfig)(ctx);
 }
 function resolveEnabledModels(service) {
   const raw = service.enabledModels;

package/dist/server/routes/router.js

@@ -55,6 +55,7 @@ var import_rate_limit = require("../middleware/rate-limit");
 var import_role_permission = require("../middleware/role-permission");
 var import_usage = require("../usage");
 var import_streaming = require("../utils/streaming");
+var import_request_cache = require("../utils/request-cache");
 var import_billing = require("../billing");
 var import_app_observability = require("../utils/app-observability");
 const API_PREFIX = "/api/ai-llm/v1";
@@ -273,7 +274,7 @@ async function resolveMaxBodyBytes(ctx) {
   var _a, _b;
   let configuredMb;
   try {
-    const config = await ctx.db.getRepository("aiApiConfig").findOne();
+    const config = await (0, import_request_cache.getAiApiConfig)(ctx);
     configuredMb = config == null ? void 0 : config.get("maxRequestBodyMb");
   } catch (err) {
     (_b = (_a = ctx.log) == null ? void 0 : _a.warn) == null ? void 0 : _b.call(_a, "AI API: could not read maxRequestBodyMb, using default:", err);
@@ -322,7 +323,7 @@ async function resolveMode(ctx) {
     return headerMode;
   }
   try {
-    const config = await ctx.db.getRepository("aiApiConfig").findOne();
+    const config = await (0, import_request_cache.getAiApiConfig)(ctx);
     if (config) {
       const dbMode = config.get("mode") || config.mode;
       if (dbMode === "agent" || dbMode === "llm") {

package/dist/server/services/file-processor.js

@@ -7,9 +7,11 @@
  * For more information, please refer to: https://www.nocobase.com/agreement.
  */
 
+var __create = Object.create;
 var __defProp = Object.defineProperty;
 var __getOwnPropDesc = Object.getOwnPropertyDescriptor;
 var __getOwnPropNames = Object.getOwnPropertyNames;
+var __getProtoOf = Object.getPrototypeOf;
 var __hasOwnProp = Object.prototype.hasOwnProperty;
 var __export = (target, all) => {
   for (var name in all)
@@ -23,6 +25,14 @@ var __copyProps = (to, from, except, desc) => {
   }
   return to;
 };
+var __toESM = (mod, isNodeMode, target) => (target = mod != null ? __create(__getProtoOf(mod)) : {}, __copyProps(
+  // If the importer is in node compatibility mode or this is not an ESM
+  // file that has been converted to a CommonJS file using a Babel-
+  // compatible transform (i.e. "__esModule" has not been set), then set
+  // "default" to the CommonJS "module.exports" for node compatibility.
+  isNodeMode || !mod || !mod.__esModule ? __defProp(target, "default", { value: mod, enumerable: true }) : target,
+  mod
+));
 var __toCommonJS = (mod) => __copyProps(__defProp({}, "__esModule", { value: true }), mod);
 var file_processor_exports = {};
 __export(file_processor_exports, {
@@ -31,10 +41,14 @@ __export(file_processor_exports, {
   base64FileForwarder: () => base64FileForwarder,
   fetchFileAsBase64: () => fetchFileAsBase64,
   httpFileUrlFetcher: () => httpFileUrlFetcher,
+  isBlockedAddress: () => isBlockedAddress,
   pdfFileProcessor: () => pdfFileProcessor
 });
 module.exports = __toCommonJS(file_processor_exports);
+var import_dns = __toESM(require("dns"));
+var import_net = require("net");
 var import_path = require("path");
+var import_request_cache = require("../utils/request-cache");
 class FileProcessorError extends Error {
   constructor(code, message) {
     super(message);
@@ -44,7 +58,9 @@ class FileProcessorError extends Error {
 }
 const DEFAULT_MAX_FILE_SIZE = 50 * 1024 * 1024;
 const DEFAULT_TIMEOUT_MS = 3e4;
+const DEFAULT_MAX_REDIRECTS = 5;
 const ALLOWED_PROTOCOLS = /* @__PURE__ */ new Set(["http:", "https:"]);
+const REDIRECT_STATUSES = /* @__PURE__ */ new Set([301, 302, 303, 307, 308]);
 class FileProcessorService {
   processors = [];
   pdfRenderer = null;
@@ -82,11 +98,6 @@ class FileProcessorService {
 function isRecord(value) {
   return typeof value === "object" && value !== null && !Array.isArray(value);
 }
-function getUrlString(value) {
-  if (typeof value === "string") return value;
-  if (isRecord(value) && typeof value.url === "string") return value.url;
-  return void 0;
-}
 function extractFilename(url, contentDisposition) {
   if (contentDisposition) {
     const match = contentDisposition.match(/filename="?([^"]+)"?/);
@@ -99,35 +110,187 @@ function extractFilename(url, contentDisposition) {
   }
   return void 0;
 }
+function parseIpv4Octets(ip) {
+  const parts = ip.split(".");
+  if (parts.length !== 4) return void 0;
+  const octets = [];
+  for (const part of parts) {
+    if (!/^\d{1,3}$/.test(part)) return void 0;
+    const value = Number(part);
+    if (value > 255) return void 0;
+    octets.push(value);
+  }
+  return octets;
+}
+function isBlockedIpv4(ip) {
+  const octets = parseIpv4Octets(ip);
+  if (!octets) return true;
+  const [a, b] = octets;
+  if (a === 0) return true;
+  if (a === 10) return true;
+  if (a === 100 && b >= 64 && b <= 127) return true;
+  if (a === 127) return true;
+  if (a === 169 && b === 254) return true;
+  if (a === 172 && b >= 16 && b <= 31) return true;
+  if (a === 192 && b === 0) return true;
+  if (a === 192 && b === 168) return true;
+  if (a === 198 && (b === 18 || b === 19)) return true;
+  if (a >= 224) return true;
+  return false;
+}
+function parseHexGroup(group) {
+  if (group.length < 1 || group.length > 4 || !/^[0-9a-fA-F]+$/.test(group)) return void 0;
+  return parseInt(group, 16);
+}
+function expandIpv6(input) {
+  let address = input;
+  const zoneIndex = address.indexOf("%");
+  if (zoneIndex !== -1) address = address.slice(0, zoneIndex);
+  if (!address) return void 0;
+  const lastColon = address.lastIndexOf(":");
+  if (lastColon !== -1 && address.includes(".", lastColon)) {
+    const octets = parseIpv4Octets(address.slice(lastColon + 1));
+    if (!octets) return void 0;
+    const high = (octets[0] << 8 | octets[1]).toString(16);
+    const low = (octets[2] << 8 | octets[3]).toString(16);
+    address = `${address.slice(0, lastColon + 1)}${high}:${low}`;
+  }
+  const groups = [];
+  const doubleColonIndex = address.indexOf("::");
+  if (doubleColonIndex !== -1) {
+    if (address.indexOf("::", doubleColonIndex + 1) !== -1) return void 0;
+    const head = address.slice(0, doubleColonIndex);
+    const tail = address.slice(doubleColonIndex + 2);
+    const headGroups = head === "" ? [] : head.split(":");
+    const tailGroups = tail === "" ? [] : tail.split(":");
+    const fillCount = 8 - headGroups.length - tailGroups.length;
+    if (fillCount < 1) return void 0;
+    const allGroups = [...headGroups];
+    for (let i = 0; i < fillCount; i += 1) allGroups.push("0");
+    allGroups.push(...tailGroups);
+    for (const group of allGroups) {
+      const parsed = parseHexGroup(group);
+      if (parsed === void 0) return void 0;
+      groups.push(parsed);
+    }
+  } else {
+    for (const group of address.split(":")) {
+      const parsed = parseHexGroup(group);
+      if (parsed === void 0) return void 0;
+      groups.push(parsed);
+    }
+  }
+  return groups.length === 8 ? groups : void 0;
+}
+function isBlockedIpv6(ip) {
+  const groups = expandIpv6(ip);
+  if (!groups) return true;
+  const [g0, g1, g2, g3, g4, g5, g6, g7] = groups;
+  if (groups.every((group) => group === 0)) return true;
+  if (g0 === 0 && g1 === 0 && g2 === 0 && g3 === 0 && g4 === 0 && g5 === 0 && g6 === 0 && g7 === 1) {
+    return true;
+  }
+  const isIpv4Mapped = g0 === 0 && g1 === 0 && g2 === 0 && g3 === 0 && g4 === 0 && g5 === 65535;
+  const isIpv4Compatible = g0 === 0 && g1 === 0 && g2 === 0 && g3 === 0 && g4 === 0 && g5 === 0;
+  const isNat64 = g0 === 100 && g1 === 65435 && g2 === 0 && g3 === 0 && g4 === 0 && g5 === 0;
+  if (isIpv4Mapped || isIpv4Compatible || isNat64) {
+    const embedded = `${g6 >> 8 & 255}.${g6 & 255}.${g7 >> 8 & 255}.${g7 & 255}`;
+    return isBlockedIpv4(embedded);
+  }
+  if ((g0 & 65024) === 64512) return true;
+  if ((g0 & 65472) === 65152) return true;
+  if ((g0 & 65280) === 65280) return true;
+  return false;
+}
+function isBlockedAddress(ip) {
+  const version = (0, import_net.isIP)(ip);
+  if (version === 4) return isBlockedIpv4(ip);
+  if (version === 6) return isBlockedIpv6(ip);
+  return true;
+}
+async function assertHostAllowed(hostname) {
+  if ((0, import_net.isIP)(hostname) !== 0) {
+    if (isBlockedAddress(hostname)) {
+      throw new FileProcessorError("blocked_host", `Host '${hostname}' is a blocked address.`);
+    }
+    return;
+  }
+  let addresses;
+  try {
+    addresses = await import_dns.default.promises.lookup(hostname, { all: true });
+  } catch (error) {
+    const message = error instanceof Error && error.message ? error.message : String(error);
+    throw new FileProcessorError("fetch_failed", `Could not resolve host '${hostname}': ${message}`);
+  }
+  if (!addresses.length) {
+    throw new FileProcessorError("fetch_failed", `Could not resolve host '${hostname}'.`);
+  }
+  for (const { address } of addresses) {
+    if (isBlockedAddress(address)) {
+      throw new FileProcessorError("blocked_host", `Host '${hostname}' resolves to blocked address '${address}'.`);
+    }
+  }
+}
+async function fetchWithSsrfGuard(initialUrl, options) {
+  let currentUrl = initialUrl;
+  for (let hop = 0; ; hop += 1) {
+    let parsed;
+    try {
+      parsed = new URL(currentUrl);
+    } catch {
+      throw new FileProcessorError("invalid_url", `File URL '${currentUrl}' is not a valid URL.`);
+    }
+    if (!options.allowedProtocols.has(parsed.protocol)) {
+      throw new FileProcessorError("unsupported_protocol", `File URL protocol '${parsed.protocol}' is not allowed.`);
+    }
+    const hostname = parsed.hostname.startsWith("[") && parsed.hostname.endsWith("]") ? parsed.hostname.slice(1, -1) : parsed.hostname;
+    await assertHostAllowed(hostname);
+    const response = await fetch(currentUrl, { signal: options.signal, redirect: "manual" });
+    if (!REDIRECT_STATUSES.has(response.status)) {
+      return { response, finalUrl: currentUrl };
+    }
+    if (hop >= options.maxRedirects) {
+      throw new FileProcessorError(
+        "too_many_redirects",
+        `File URL '${initialUrl}' exceeded the limit of ${options.maxRedirects} redirects.`
+      );
+    }
+    const location = response.headers.get("location");
+    if (!location) {
+      throw new FileProcessorError("fetch_failed", `Redirect from '${currentUrl}' is missing a Location header.`);
+    }
+    try {
+      currentUrl = new URL(location, currentUrl).toString();
+    } catch {
+      throw new FileProcessorError(
+        "invalid_url",
+        `Redirect Location '${location}' from '${currentUrl}' is not a valid URL.`
+      );
+    }
+  }
+}
 async function fetchFileAsBase64(url, options = {}) {
   const maxSize = options.maxSizeBytes ?? DEFAULT_MAX_FILE_SIZE;
   const timeoutMs = options.timeoutMs ?? DEFAULT_TIMEOUT_MS;
+  const maxRedirects = options.maxRedirects ?? DEFAULT_MAX_REDIRECTS;
   const allowedProtocols = options.allowedProtocols ? new Set(options.allowedProtocols) : ALLOWED_PROTOCOLS;
-  let protocol;
-  try {
-    protocol = new URL(url).protocol;
-  } catch {
-    throw new FileProcessorError("invalid_url", `File URL '${url}' is not a valid URL.`);
-  }
-  if (!allowedProtocols.has(protocol)) {
-    throw new FileProcessorError("unsupported_protocol", `File URL protocol '${protocol}' is not allowed.`);
-  }
   const controller = new AbortController();
   const timeout = setTimeout(() => controller.abort(), timeoutMs);
   try {
-    const response = await fetch(url, {
-      signal: controller.signal,
-      redirect: "follow"
+    const { response, finalUrl } = await fetchWithSsrfGuard(url, {
+      allowedProtocols,
+      maxRedirects,
+      signal: controller.signal
     });
     if (!response.ok) {
       throw new FileProcessorError(
         "fetch_failed",
-        `Failed to fetch file from '${url}': ${response.status} ${response.statusText}`
+        `Failed to fetch file from '${finalUrl}': ${response.status} ${response.statusText}`
       );
     }
     const contentLength = response.headers.get("content-length");
     if (contentLength && Number(contentLength) > maxSize) {
-      throw new FileProcessorError("file_too_large", `File at '${url}' exceeds maximum allowed size.`);
+      throw new FileProcessorError("file_too_large", `File at '${finalUrl}' exceeds maximum allowed size.`);
     }
     const contentType = response.headers.get("content-type") || void 0;
     if (options.allowedContentTypes && contentType && !options.allowedContentTypes.some((type) => contentType.includes(type))) {
@@ -135,11 +298,11 @@ async function fetchFileAsBase64(url, options = {}) {
     }
     const buffer = Buffer.from(await response.arrayBuffer());
     if (buffer.length > maxSize) {
-      throw new FileProcessorError("file_too_large", `File at '${url}' exceeds maximum allowed size.`);
+      throw new FileProcessorError("file_too_large", `File at '${finalUrl}' exceeds maximum allowed size.`);
     }
     const mimeType = (contentType == null ? void 0 : contentType.split(";")[0].trim()) ?? "application/octet-stream";
     const contentDisposition = response.headers.get("content-disposition");
-    const filename = extractFilename(url, contentDisposition) ?? "file";
+    const filename = extractFilename(finalUrl, contentDisposition) ?? "file";
     return {
       fileData: `data:${mimeType};base64,${buffer.toString("base64")}`,
       mimeType,
@@ -225,7 +388,7 @@ const pdfFileProcessor = {
   },
   async process(block, context) {
     var _a, _b, _c, _d;
-    const config = await context.ctx.db.getRepository("aiApiConfig").findOne();
+    const config = await (0, import_request_cache.getAiApiConfig)(context.ctx);
     if (!(config == null ? void 0 : config.pdfRenderPagesAsImages)) {
       return block;
     }
@@ -258,5 +421,6 @@ const pdfFileProcessor = {
   base64FileForwarder,
   fetchFileAsBase64,
   httpFileUrlFetcher,
+  isBlockedAddress,
   pdfFileProcessor
 });

package/dist/server/usage.js

@@ -74,7 +74,10 @@ function normalizeUsage(value) {
     prompt_tokens: prompt,
     completion_tokens: completion,
     total_tokens: total,
-    prompt_cache_tokens: extractPromptCacheTokens(source)
+    // Keep normalizeUsage idempotent: streaming routes normalize the chunk
+    // usage once and hand the result to setAiApiUsageResult, which normalizes
+    // again — an already-extracted prompt_cache_tokens must survive that pass.
+    prompt_cache_tokens: extractPromptCacheTokens(source) ?? normalizeTokenCount(source.prompt_cache_tokens)
   };
 }
 function setAiApiUsageResult(ctx, value, metadata = {}, responseMetadata) {
@@ -171,6 +174,7 @@ async function finishUsageRecord(ctx, id, startedAt, status) {
     costStatus: billing.costStatus ?? null,
     modelPriceId: billing.modelPriceId ?? null,
     quotaPolicyId: billing.quotaPolicyId ?? null,
+    groupId: billing.groupId ?? null,
     inputPricePerMillionTokens: billing.inputPricePerMillionTokens ?? null,
     outputPricePerMillionTokens: billing.outputPricePerMillionTokens ?? null,
     fixedCostPerRequest: billing.fixedCostPerRequest ?? null,

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

@@ -31,6 +31,7 @@ __export(direct_llm_context_exports, {
   prepareDirectLlmContext: () => prepareDirectLlmContext
 });
 module.exports = __toCommonJS(direct_llm_context_exports);
+var import_request_cache = require("./request-cache");
 class DirectLlmContextError extends Error {
   constructor(code, message) {
     super(message);
@@ -237,16 +238,16 @@ async function loadModelMetadata(ctx, serviceName, modelId) {
       `Context metadata is not configured for '${serviceName}/${modelId}'. Configure context window and max completion tokens.`
     );
   }
-  return { contextWindow, maxCompletionTokens };
+  const systemPromptValue = getValue(row, "systemPrompt");
+  const systemPrompt = typeof systemPromptValue === "string" ? systemPromptValue.trim() : "";
+  return { contextWindow, maxCompletionTokens, ...systemPrompt ? { systemPrompt } : {} };
 }
 async function resolveOverflowBehavior(ctx) {
   var _a;
   const userId = (_a = ctx.state.currentUser) == null ? void 0 : _a.id;
   if (userId === null || userId === void 0) return "reject";
-  const policy = await ctx.db.getRepository("aiApiUserQuotaPolicies").findOne({
-    filter: { userId, enabled: true }
-  });
-  return getValue(policy, "contextOverflowBehavior") === "truncate" ? "truncate" : "reject";
+  const group = await (0, import_request_cache.resolveRequestUserGroup)(ctx, userId);
+  return group.contextOverflowBehavior === "truncate" ? "truncate" : "reject";
 }
 function resolveReservedOutputTokens(options, metadata) {
   const requested = positiveInteger(options.maxCompletionTokens ?? options.maxTokens);
@@ -272,10 +273,11 @@ async function prepareDirectLlmContext(ctx, options) {
     );
   }
   const fixedOverheadTokens = estimateValueTokens(options.tools) + (options.tools === void 0 ? 0 : 4);
-  const originalEstimate = estimateMessagesTokens(options.messages) + fixedOverheadTokens;
+  const baseMessages = metadata.systemPrompt ? [{ role: "system", content: metadata.systemPrompt }, ...options.messages] : options.messages;
+  const originalEstimate = estimateMessagesTokens(baseMessages) + fixedOverheadTokens;
   if (originalEstimate <= inputTokenBudget) {
     return {
-      messages: options.messages,
+      messages: baseMessages,
       estimatedInputTokens: originalEstimate,
       inputTokenBudget,
       reservedOutputTokens,
@@ -288,13 +290,13 @@ async function prepareDirectLlmContext(ctx, options) {
       `Estimated input tokens (${originalEstimate}) exceed the allowed input budget (${inputTokenBudget}).`
     );
   }
-  const { turns } = splitTurns(options.messages);
+  const { turns } = splitTurns(baseMessages);
   let remainingTurns = turns;
-  let messages = messagesWithTurns(options.messages, remainingTurns);
+  let messages = messagesWithTurns(baseMessages, remainingTurns);
   let estimatedInputTokens = estimateMessagesTokens(messages) + fixedOverheadTokens;
   while (remainingTurns.length > 1 && estimatedInputTokens > inputTokenBudget) {
     remainingTurns = remainingTurns.slice(1);
-    messages = messagesWithTurns(options.messages, remainingTurns);
+    messages = messagesWithTurns(baseMessages, remainingTurns);
     estimatedInputTokens = estimateMessagesTokens(messages) + fixedOverheadTokens;
   }
   if (estimatedInputTokens > inputTokenBudget) {
@@ -308,7 +310,7 @@ async function prepareDirectLlmContext(ctx, options) {
     estimatedInputTokens,
     inputTokenBudget,
     reservedOutputTokens,
-    truncated: messages.length !== options.messages.length
+    truncated: messages.length !== baseMessages.length
   };
 }
 // Annotate the CommonJS export names for ESM import in node:

package/dist/server/utils/rate-limiter.js

@@ -39,7 +39,7 @@ class RateLimiter {
    * Check and record a request for a user.
    *
    * @param userId  The user ID (string or numeric)
-   * @param limit   Max allowed requests per window (from aiApiConfig.rateLimitPerMinute)
+   * @param limit   Max allowed requests per window (from the user's usage group)
    * @returns { allowed: true } or { allowed: false, retryAfterMs: number }
    */
   check(userId, limit) {

package/dist/server/utils/request-cache.js

@@ -0,0 +1,61 @@
+/**
+ * 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 request_cache_exports = {};
+__export(request_cache_exports, {
+  getAiApiConfig: () => getAiApiConfig,
+  resolveRequestUserGroup: () => resolveRequestUserGroup
+});
+module.exports = __toCommonJS(request_cache_exports);
+var import_quota_groups = require("../quota-groups");
+function getCache(ctx) {
+  if (!ctx.state.aiApiRequestCache) {
+    ctx.state.aiApiRequestCache = { configLoaded: false, config: null };
+  }
+  return ctx.state.aiApiRequestCache;
+}
+async function getAiApiConfig(ctx) {
+  const cache = getCache(ctx);
+  if (!cache.configLoaded) {
+    cache.config = await ctx.db.getRepository("aiApiConfig").findOne() ?? null;
+    cache.configLoaded = true;
+  }
+  return cache.config;
+}
+async function resolveRequestUserGroup(ctx, userId) {
+  const cache = getCache(ctx);
+  const groupKey = userId === void 0 || userId === null ? "" : String(userId);
+  if (!cache.group || cache.groupKey !== groupKey) {
+    cache.group = await (0, import_quota_groups.resolveUserGroup)(ctx, userId);
+    cache.groupKey = groupKey;
+  }
+  return cache.group;
+}
+// Annotate the CommonJS export names for ESM import in node:
+0 && (module.exports = {
+  getAiApiConfig,
+  resolveRequestUserGroup
+});

package/dist/server/utils/resolve-service.js

@@ -30,6 +30,7 @@ __export(resolve_service_exports, {
   resolveModelString: () => resolveModelString
 });
 module.exports = __toCommonJS(resolve_service_exports);
+var import_request_cache = require("./request-cache");
 async function resolveLlmService(ctx, serviceKey) {
   const repo = ctx.db.getRepository("llmServices");
   let service = await repo.findOne({ filter: { name: serviceKey } });
@@ -60,7 +61,7 @@ async function resolveModelString(ctx, modelString) {
       }
     }
   }
-  const config = await ctx.db.getRepository("aiApiConfig").findOne();
+  const config = await (0, import_request_cache.getAiApiConfig)(ctx);
   if (config == null ? void 0 : config.defaultLlmService) {
     const service = await repo.findOne({ filter: { name: config.defaultLlmService } });
     if (service) {