plugin-ai-api 1.0.24 → 1.0.28

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

@@ -27,9 +27,11 @@ var __toCommonJS = (mod) => __copyProps(__defProp({}, "__esModule", { value: tru
 var direct_llm_context_exports = {};
 __export(direct_llm_context_exports, {
   DirectLlmContextError: () => DirectLlmContextError,
+  parseImageDimensions: () => parseImageDimensions,
   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);
@@ -45,20 +47,159 @@ function positiveInteger(value) {
   const parsed = Number(value);
   return Number.isSafeInteger(parsed) && parsed > 0 ? parsed : void 0;
 }
-function hasImageContent(value) {
-  return Array.isArray(value) && value.some(
-    (block) => typeof block === "object" && block !== null && block.type === "image_url"
-  );
-}
 function estimateValueTokens(value) {
   if (value === void 0) return 0;
   return Math.ceil(Buffer.byteLength(JSON.stringify(value), "utf8") / 3);
 }
-function estimateMessagesTokens(messages) {
-  return messages.reduce((total, message) => total + estimateValueTokens(message) + 4, 0);
+function isRecord(value) {
+  return typeof value === "object" && value !== null && !Array.isArray(value);
+}
+function readBigEndian(buf, offset) {
+  return buf.readUInt32BE(offset);
+}
+function readUInt16BE(buf, offset) {
+  return buf.readUInt16BE(offset);
+}
+function readLittleEndian(buf, offset) {
+  return buf.readUInt16LE(offset);
+}
+function parsePngDimensions(buffer) {
+  if (buffer.length < 24) return void 0;
+  return { width: readBigEndian(buffer, 16), height: readBigEndian(buffer, 20) };
+}
+function parseJpegDimensions(buffer) {
+  let offset = 2;
+  while (offset < buffer.length) {
+    if (buffer[offset] !== 255) {
+      offset++;
+      continue;
+    }
+    const marker = buffer[offset + 1];
+    if (marker >= 192 && marker <= 195 || marker >= 197 && marker <= 199 || marker >= 201 && marker <= 203 || marker >= 205 && marker <= 207) {
+      if (offset + 9 <= buffer.length) {
+        return { height: readUInt16BE(buffer, offset + 5), width: readUInt16BE(buffer, offset + 7) };
+      }
+      return void 0;
+    }
+    if (marker === 217 || offset + 4 >= buffer.length) break;
+    const segmentLength = buffer.readUInt16BE(offset + 2);
+    offset += 2 + segmentLength;
+  }
+  return void 0;
 }
-function containsUnsupportedContent(messages) {
-  return messages.some((message) => hasImageContent(message.content));
+function parseGifDimensions(buffer) {
+  if (buffer.length < 10) return void 0;
+  return { width: readLittleEndian(buffer, 6), height: readLittleEndian(buffer, 8) };
+}
+function parseWebpDimensions(buffer) {
+  if (buffer.length < 30) return void 0;
+  const riff = buffer.toString("ascii", 0, 4);
+  const webp = buffer.toString("ascii", 8, 12);
+  if (riff !== "RIFF" || webp !== "WEBP") return void 0;
+  const chunkType = buffer.toString("ascii", 12, 16);
+  if (chunkType === "VP8 " && buffer.length >= 26) {
+    return { width: readLittleEndian(buffer, 26), height: readLittleEndian(buffer, 28) };
+  }
+  if (chunkType === "VP8L" && buffer.length >= 24) {
+    const bits = buffer.readUInt32LE(21);
+    return {
+      width: (bits & 16383) + 1,
+      height: (bits >> 14 & 16383) + 1
+    };
+  }
+  if (chunkType === "VP8X" && buffer.length >= 30) {
+    return {
+      width: ((buffer[24] | buffer[25] << 8 | buffer[26] << 16) & 16777215) + 1,
+      height: ((buffer[27] | buffer[28] << 8 | buffer[29] << 16) & 16777215) + 1
+    };
+  }
+  return void 0;
+}
+function parseImageDimensions(buffer) {
+  if (buffer.length < 12) return void 0;
+  const header = buffer.toString("binary", 0, 4);
+  if (header === "\x89PNG") return parsePngDimensions(buffer);
+  if (header === "GIF8") return parseGifDimensions(buffer);
+  if (header === "RIFF") return parseWebpDimensions(buffer);
+  if (buffer[0] === 255 && buffer[1] === 216) return parseJpegDimensions(buffer);
+  return void 0;
+}
+const VISION_TILE_SIZE = 512;
+const VISION_LOW_DETAIL_TOKENS = 85;
+const VISION_TILE_TOKENS = 170;
+const VISION_HTTP_URL_ESTIMATE = 1024;
+const FILE_BASE64_FALLBACK_TOKENS = 1024;
+function estimateVisionTokensForDimensions(width, height) {
+  if (!Number.isFinite(width) || !Number.isFinite(height) || width <= 0 || height <= 0) {
+    return VISION_LOW_DETAIL_TOKENS;
+  }
+  const tilesX = Math.ceil(width / VISION_TILE_SIZE);
+  const tilesY = Math.ceil(height / VISION_TILE_SIZE);
+  return VISION_LOW_DETAIL_TOKENS + tilesX * tilesY * VISION_TILE_TOKENS;
+}
+function decodeBase64DataUrl(url) {
+  const match = /^data:([^;]+);base64,([A-Za-z0-9+/]+=?=?)$/.exec(url);
+  if (!match) return void 0;
+  try {
+    const buffer = Buffer.from(match[2], "base64");
+    return { mimeType: match[1].toLowerCase(), buffer };
+  } catch {
+    return void 0;
+  }
+}
+function estimateImageUrlTokens(imageUrl) {
+  const url = typeof imageUrl === "string" ? imageUrl : isRecord(imageUrl) ? String(imageUrl.url ?? "") : "";
+  if (!url) return 0;
+  if (url.startsWith("data:")) {
+    const decoded = decodeBase64DataUrl(url);
+    if (!decoded) return FILE_BASE64_FALLBACK_TOKENS;
+    if (!decoded.mimeType.startsWith("image/")) return FILE_BASE64_FALLBACK_TOKENS;
+    const dimensions = parseImageDimensions(decoded.buffer);
+    return dimensions ? estimateVisionTokensForDimensions(dimensions.width, dimensions.height) : VISION_LOW_DETAIL_TOKENS;
+  }
+  if (url.startsWith("http://") || url.startsWith("https://")) {
+    return VISION_HTTP_URL_ESTIMATE;
+  }
+  return 0;
+}
+function estimateFileBlockTokens(block) {
+  const file = isRecord(block.file) ? block.file : void 0;
+  if (!file) return 0;
+  const fileData = String(file.file_data ?? "");
+  if (fileData.startsWith("data:")) {
+    const decoded = decodeBase64DataUrl(fileData);
+    if (decoded) {
+      return Math.max(1, Math.ceil(decoded.buffer.length / 3));
+    }
+    return FILE_BASE64_FALLBACK_TOKENS;
+  }
+  return FILE_BASE64_FALLBACK_TOKENS;
+}
+function estimateContentBlockTokens(block) {
+  if (!isRecord(block)) return estimateValueTokens(block);
+  const type = typeof block.type === "string" ? block.type : void 0;
+  if (type === "text") {
+    return typeof block.text === "string" ? estimateValueTokens(block.text) + 4 : 4;
+  }
+  if (type === "image_url") {
+    return estimateImageUrlTokens(block.image_url) + 4;
+  }
+  if (type === "file") {
+    return estimateFileBlockTokens(block) + 4;
+  }
+  if (type === "file_url") {
+    return VISION_HTTP_URL_ESTIMATE + 4;
+  }
+  return estimateValueTokens(block) + 4;
+}
+function estimateMessageTokens(message) {
+  if (Array.isArray(message.content)) {
+    return message.content.reduce((total, block) => total + estimateContentBlockTokens(block), 4);
+  }
+  return estimateValueTokens(message) + 4;
+}
+function estimateMessagesTokens(messages) {
+  return messages.reduce((total, message) => total + estimateMessageTokens(message), 0);
 }
 function isInstruction(message) {
   return message.role === "system" || message.role === "developer";
@@ -97,16 +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);
@@ -119,12 +260,6 @@ function resolveReservedOutputTokens(options, metadata) {
   return requested ?? metadata.maxCompletionTokens;
 }
 async function prepareDirectLlmContext(ctx, options) {
-  if (containsUnsupportedContent(options.messages)) {
-    throw new DirectLlmContextError(
-      "context_estimation_unsupported",
-      "Context enforcement does not support image_url content without a model-specific vision token estimator."
-    );
-  }
   const [metadata, behavior] = await Promise.all([
     loadModelMetadata(ctx, options.serviceName, options.modelId),
     resolveOverflowBehavior(ctx)
@@ -138,10 +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,
@@ -154,13 +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) {
@@ -174,11 +310,12 @@ 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:
 0 && (module.exports = {
   DirectLlmContextError,
+  parseImageDimensions,
   prepareDirectLlmContext
 });

package/dist/server/utils/openai-format.js

@@ -99,7 +99,19 @@ function toOpenAIResponse(options) {
         finish_reason: finishReason
       }
     ],
-    usage: usage || { prompt_tokens: null, completion_tokens: null, total_tokens: null }
+    usage: usage ? {
+      prompt_tokens: usage.prompt_tokens ?? null,
+      completion_tokens: usage.completion_tokens ?? null,
+      total_tokens: usage.total_tokens ?? null,
+      prompt_tokens_details: {
+        cached_tokens: usage.prompt_cache_tokens ?? null
+      }
+    } : {
+      prompt_tokens: null,
+      completion_tokens: null,
+      total_tokens: null,
+      prompt_tokens_details: { cached_tokens: null }
+    }
   };
 }
 function toOpenAIStreamChunk(options) {
@@ -129,7 +141,14 @@ function toOpenAIUsageChunk(options) {
     created: Math.floor(Date.now() / 1e3),
     model,
     choices: [],
-    usage
+    usage: {
+      prompt_tokens: usage.prompt_tokens,
+      completion_tokens: usage.completion_tokens,
+      total_tokens: usage.total_tokens,
+      prompt_tokens_details: {
+        cached_tokens: usage.prompt_cache_tokens ?? null
+      }
+    }
   };
 }
 function toOpenAIEmbeddingsResponse(options) {

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) {

package/dist/server/utils/user-permissions.js

@@ -28,73 +28,60 @@ var user_permissions_exports = {};
 __export(user_permissions_exports, {
   buildAccessScope: () => buildAccessScope,
   enforceModelAccess: () => enforceModelAccess,
-  invalidateUserPermissionCache: () => invalidateUserPermissionCache,
+  invalidateGroupAccessCache: () => invalidateGroupAccessCache,
   isModelAllowed: () => isModelAllowed,
   isServiceAllowed: () => isServiceAllowed,
   resolveUserAccessScope: () => resolveUserAccessScope
 });
 module.exports = __toCommonJS(user_permissions_exports);
 var import_openai_format = require("./openai-format");
+var import_request_cache = require("./request-cache");
 const SCOPE_TTL_MS = 15e3;
 const scopeCache = /* @__PURE__ */ new Map();
-const NO_RECORD_SCOPE = {
-  hasUserRecord: false,
-  denyAll: false,
-  allowedServices: null,
+const OPEN_SCOPE = {
+  allowedServices: [],
   allowAllModels: true,
   allowedModels: /* @__PURE__ */ new Set(),
   lookupFailed: false
 };
-const LOOKUP_FAILED_SCOPE = { ...NO_RECORD_SCOPE, denyAll: true, lookupFailed: true };
-function invalidateUserPermissionCache(userId) {
-  if (userId === void 0 || userId === null) {
+const LOOKUP_FAILED_SCOPE = { ...OPEN_SCOPE, lookupFailed: true };
+function invalidateGroupAccessCache(groupId) {
+  if (groupId === void 0 || groupId === null) {
     scopeCache.clear();
     return;
   }
-  const suffix = `:${userId}`;
+  const suffix = `:group:${groupId}`;
   for (const key of scopeCache.keys()) {
     if (key.endsWith(suffix)) scopeCache.delete(key);
   }
 }
-function valueOf(row, name) {
-  if (!row) return void 0;
-  const candidate = row;
-  if (typeof candidate.get === "function") return candidate.get(name);
-  return row[name];
-}
 function toStringArray(value) {
   if (!Array.isArray(value)) return [];
   return value.filter((item) => typeof item === "string" && item.length > 0);
 }
-function buildAccessScope(row) {
-  if (!row) return NO_RECORD_SCOPE;
-  if (valueOf(row, "enabled") === false) {
-    return { ...NO_RECORD_SCOPE, hasUserRecord: true, denyAll: true, allowedServices: [] };
-  }
+function buildAccessScope(group) {
   return {
-    hasUserRecord: true,
-    denyAll: false,
-    allowedServices: toStringArray(valueOf(row, "allowedLlmServices")),
-    allowAllModels: valueOf(row, "allowAllModels") !== false,
-    allowedModels: new Set(toStringArray(valueOf(row, "allowedModels"))),
+    groupId: group.id,
+    allowedServices: toStringArray(group.allowedLlmServices),
+    allowAllModels: group.allowAllModels !== false,
+    allowedModels: new Set(toStringArray(group.allowedModels)),
     lookupFailed: false
   };
 }
 async function resolveUserAccessScope(ctx) {
   var _a, _b, _c, _d;
   const userId = (_a = ctx.state.currentUser) == null ? void 0 : _a.id;
-  if (userId === void 0 || userId === null) return NO_RECORD_SCOPE;
-  const key = `${((_b = ctx.app) == null ? void 0 : _b.name) ?? "main"}:${userId}`;
-  const cached = scopeCache.get(key);
-  if (cached && cached.expiresAt > Date.now()) return cached.scope;
-  let scope;
+  let group;
   try {
-    const row = await ctx.db.getRepository("aiApiUserPermissions").findOne({ filter: { userId } });
-    scope = buildAccessScope(row);
+    group = await (0, import_request_cache.resolveRequestUserGroup)(ctx, userId);
   } catch (err) {
-    (_d = (_c = ctx.log) == null ? void 0 : _c.error) == null ? void 0 : _d.call(_c, "AI API user permissions lookup failed, denying access:", err);
+    (_c = (_b = ctx.log) == null ? void 0 : _b.error) == null ? void 0 : _c.call(_b, "AI API group access lookup failed, denying access:", err);
     return LOOKUP_FAILED_SCOPE;
   }
+  const key = `${((_d = ctx.app) == null ? void 0 : _d.name) ?? "main"}:group:${group.id}`;
+  const cached = scopeCache.get(key);
+  if (cached && cached.expiresAt > Date.now()) return cached.scope;
+  const scope = buildAccessScope(group);
   scopeCache.set(key, { scope, expiresAt: Date.now() + SCOPE_TTL_MS });
   return scope;
 }
@@ -102,15 +89,14 @@ function matchesService(list, serviceName, serviceTitle) {
   return list.some((entry) => entry === serviceName || entry === serviceTitle);
 }
 function isServiceAllowed(scope, globalEnabledServices, service) {
-  if (scope.denyAll) return false;
+  if (scope.lookupFailed) return false;
   const globalList = toStringArray(globalEnabledServices);
   if (globalList.length && !matchesService(globalList, service.name, service.title)) return false;
-  if (!scope.hasUserRecord) return true;
-  return matchesService(scope.allowedServices ?? [], service.name, service.title);
+  if (!scope.allowedServices.length) return true;
+  return matchesService(scope.allowedServices, service.name, service.title);
 }
 function isModelAllowed(scope, fullModelId) {
-  if (scope.denyAll) return false;
-  if (!scope.hasUserRecord) return true;
+  if (scope.lookupFailed) return false;
   if (scope.allowAllModels) return true;
   return scope.allowedModels.has(fullModelId);
 }
@@ -153,7 +139,7 @@ async function enforceModelAccess(ctx, globalEnabledServices, service, modelId)
 0 && (module.exports = {
   buildAccessScope,
   enforceModelAccess,
-  invalidateUserPermissionCache,
+  invalidateGroupAccessCache,
   isModelAllowed,
   isServiceAllowed,
   resolveUserAccessScope

package/dist/server/validation.js

@@ -88,6 +88,10 @@ function validateModelMetadata(model) {
   if (!String(model.get("model") ?? "").trim()) throw new Error("model is required.");
   requirePositiveIntegerOrNull(model.get("contextWindow"), "contextWindow");
   requirePositiveIntegerOrNull(model.get("maxCompletionTokens"), "maxCompletionTokens");
+  const systemPrompt = model.get("systemPrompt");
+  if (systemPrompt !== null && systemPrompt !== void 0 && typeof systemPrompt !== "string") {
+    throw new Error("systemPrompt must be a string.");
+  }
   const contextWindow = model.get("contextWindow");
   const maxCompletionTokens = model.get("maxCompletionTokens");
   if (contextWindow !== null && contextWindow !== void 0 && contextWindow !== "" && maxCompletionTokens !== null && maxCompletionTokens !== void 0 && maxCompletionTokens !== "" && Number(maxCompletionTokens) > Number(contextWindow)) {
@@ -98,6 +102,9 @@ function validateQuotaPolicy(model) {
   if (!["daily", "monthly"].includes(String(model.get("periodType")))) {
     throw new Error("periodType must be daily or monthly.");
   }
+  if (!["share", "per_user"].includes(String(model.get("quotaMode")))) {
+    throw new Error("quotaMode must be share or per_user.");
+  }
   if (!["allow", "use_reserved"].includes(String(model.get("missingUsageBehavior")))) {
     throw new Error("missingUsageBehavior must be allow or use_reserved.");
   }

package/dist/swagger.js

@@ -78,7 +78,7 @@ var swagger_default = {
       get: {
         tags: ["ai-llm"],
         summary: "List available models",
-        description: "Returns the LLM models available to the authenticated caller across registered services. Model IDs are formatted as `serviceName/modelId`.\n\nThe catalog is user-scoped: it starts from `enabledLlmServices` in the AI API configuration, then narrows to the caller's `aiApiUserPermissions` record when one exists. A user grant can only narrow the global whitelist, never widen it, so two users may receive different lists from the same request.",
+        description: "Returns the LLM models available to the authenticated caller across registered services. Model IDs are formatted as `serviceName/modelId`.\n\nThe catalog is user-scoped: it starts from `enabledLlmServices` in the AI API configuration, then narrows to the caller's usage group settings (`allowedLlmServices` / `allowedModels`). Group settings can only narrow the global whitelist, never widen it, so two users may receive different lists from the same request.",
         security: [{ BearerAuth: [] }],
         responses: {
           200: {
@@ -268,16 +268,15 @@ var swagger_default = {
             enum: ["llm", "agent"],
             description: "Default AI mode"
           },
-          defaultAiEmployee: { type: "string", description: "Default AI employee name" },
+          defaultAiEmployee: {
+            type: "string",
+            description: "Default AI employee name (agent mode only; direct LLM mode ignores it)"
+          },
           defaultLlmService: { type: "string", description: "Default LLM service name" },
           enabledLlmServices: {
             type: "array",
             items: { type: "string" },
-            description: "List of enabled LLM service names. This is the outer bound for every caller; per-user `aiApiUserPermissions` records can only narrow it further."
-          },
-          rateLimitPerMinute: {
-            type: "integer",
-            description: "Max requests per minute per user (0 = unlimited)"
+            description: "List of enabled LLM service names. This is the outer bound for every caller; usage group settings can only narrow it further."
           },
           maxRequestBodyMb: {
             type: "integer",
@@ -285,6 +284,11 @@ var swagger_default = {
             maximum: 100,
             default: 10,
             description: "Max request body size in megabytes. Requests above this return 413. The gateway buffers each body in memory, so values above 100 are rejected."
+          },
+          pdfRenderPagesAsImages: {
+            type: "boolean",
+            default: false,
+            description: "When true, PDF file/file_url blocks are rendered to per-page PNG images and sent as image_url blocks. Requires a registered PdfToImageRenderer. When false or no renderer is available, PDFs are forwarded as file blocks."
           }
         }
       },
@@ -299,9 +303,9 @@ var swagger_default = {
       },
       ContentBlock: {
         type: "object",
-        description: "A multimodal content block. Only text and image_url blocks are forwarded to the provider; any other type is rejected with 400 unsupported_content_block.",
+        description: "A multimodal content block. text, image_url, file and file_url blocks are forwarded; file and file_url blocks are first run through the configurable file processor service.",
         properties: {
-          type: { type: "string", enum: ["text", "image_url"] },
+          type: { type: "string", enum: ["text", "image_url", "file", "file_url"] },
           text: { type: "string" },
           image_url: {
             type: "object",
@@ -314,6 +318,34 @@ var swagger_default = {
               detail: { type: "string", enum: ["auto", "low", "high"] }
             },
             required: ["url"]
+          },
+          file: {
+            type: "object",
+            properties: {
+              file_data: {
+                type: "string",
+                description: "A base64 data URL, e.g. data:application/pdf;base64,JVBERi0...",
+                example: "data:application/pdf;base64,JVBERi0..."
+              },
+              filename: { type: "string" },
+              mime_type: {
+                type: "string",
+                description: "MIME type of the file, e.g. application/pdf",
+                example: "application/pdf"
+              }
+            },
+            required: ["file_data"]
+          },
+          file_url: {
+            type: "object",
+            properties: {
+              url: {
+                type: "string",
+                description: "An http(s) URL pointing to a file. The gateway downloads the file and converts it to a file block.",
+                example: "https://example.com/document.pdf"
+              }
+            },
+            required: ["url"]
           }
         },
         required: ["type"]
@@ -374,7 +406,13 @@ var swagger_default = {
             properties: {
               prompt_tokens: { type: "integer" },
               completion_tokens: { type: "integer" },
-              total_tokens: { type: "integer" }
+              total_tokens: { type: "integer" },
+              prompt_tokens_details: {
+                type: "object",
+                properties: {
+                  cached_tokens: { type: "integer" }
+                }
+              }
             }
           }
         }

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "plugin-ai-api",
-  "version": "1.0.24",
+  "version": "1.0.28",
   "main": "dist/server/index.js",
   "dependencies": {},
   "peerDependencies": {