plugin-ai-api 1.0.24 → 1.0.28

package/dist/server/routes/chat-completions.js

@@ -27,6 +27,7 @@ var __toCommonJS = (mod) => __copyProps(__defProp({}, "__esModule", { value: tru
 var chat_completions_exports = {};
 __export(chat_completions_exports, {
   applyProviderRequestParameters: () => applyProviderRequestParameters,
+  extractFinishReason: () => extractFinishReason,
   findContentBlockProblem: () => findContentBlockProblem,
   findMessageProblem: () => findMessageProblem,
   getProviderRequestParameters: () => getProviderRequestParameters,
@@ -37,12 +38,13 @@ module.exports = __toCommonJS(chat_completions_exports);
 var import_openai_format = require("../utils/openai-format");
 var import_resolve_service = require("../utils/resolve-service");
 var import_streaming = require("../utils/streaming");
-var import_role_permission = require("../middleware/role-permission");
 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");
 var import_direct_llm_context = require("../utils/direct-llm-context");
 var import_app_observability = require("../utils/app-observability");
+var import_file_processor = require("../services/file-processor");
 async function handleChatCompletions(ctx, plugin) {
   var _a;
   const body = ctx.request.body;
@@ -118,7 +120,7 @@ async function handleChatCompletions(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;
     }
@@ -147,30 +149,13 @@ async function handleChatCompletions(ctx, plugin) {
     if (body.frequency_penalty !== void 0) modelOptions.frequencyPenalty = body.frequency_penalty;
     if (body.presence_penalty !== void 0) modelOptions.presencePenalty = body.presence_penalty;
     if (body.stop !== void 0) modelOptions.stop = body.stop;
-    let systemPrompt = "";
-    if (config == null ? void 0 : config.defaultAiEmployee) {
-      if (!(0, import_role_permission.checkEmployeeAccess)(ctx, config.defaultAiEmployee)) {
-        ctx.status = 403;
-        ctx.body = (0, import_openai_format.toOpenAIError)(
-          403,
-          `Role is not permitted to use AI Employee '${config.defaultAiEmployee}'. An admin must grant access in Settings \u2192 Users & Permissions \u2192 [Role] \u2192 AI API.`,
-          "permission_denied",
-          "employee_not_permitted"
-        );
-        return;
-      }
-      const employee = await ctx.db.getRepository("aiEmployees").findOne({
-        filter: { username: config.defaultAiEmployee }
-      });
-      if (employee) {
-        systemPrompt = employee.about || employee.defaultPrompt || "";
-      }
-    }
     let messages = [...body.messages];
-    const hasSystemMessage = messages.some((m) => m.role === "system");
-    if (systemPrompt && !hasSystemMessage) {
-      messages.unshift({ role: "system", content: systemPrompt });
-    }
+    messages = await Promise.all(
+      messages.map(async (msg) => ({
+        ...msg,
+        content: await processMessageContentFileBlocks(msg.content, ctx, plugin)
+      }))
+    );
     const preparedContext = await (0, import_direct_llm_context.prepareDirectLlmContext)(ctx, {
       serviceName: service.name,
       modelId,
@@ -232,13 +217,14 @@ async function handleChatCompletions(ctx, plugin) {
     if (!((_a = ctx.res) == null ? void 0 : _a.headersSent)) {
       const isQuotaError = err instanceof import_billing.AiApiQuotaError;
       const isContextError = err instanceof import_direct_llm_context.DirectLlmContextError;
-      ctx.status = isQuotaError ? 429 : isContextError ? 400 : 500;
+      const isFileError = err instanceof import_file_processor.FileProcessorError;
+      ctx.status = isQuotaError ? 429 : isContextError || isFileError ? 400 : 500;
       if (isQuotaError) ctx.set("X-RateLimit-Reason", err.code);
       ctx.body = (0, import_openai_format.toOpenAIError)(
         ctx.status,
         getErrorMessage(err, "Internal server error"),
-        isQuotaError ? "quota_error" : isContextError ? "invalid_request_error" : "server_error",
-        isQuotaError || isContextError ? err.code : void 0
+        isQuotaError ? "quota_error" : isContextError || isFileError ? "invalid_request_error" : "server_error",
+        isQuotaError || isContextError || isFileError ? err.code : void 0
       );
     }
   }
@@ -252,18 +238,25 @@ async function handleNonStreamingCompletion(ctx, chatModel, messages, completion
     const textPart = result.content.find((c) => c.type === "text");
     content = (textPart == null ? void 0 : textPart.text) || JSON.stringify(result.content);
   }
-  const usage = (0, import_usage.setAiApiUsageResult)(ctx, result.usage_metadata, {
-    gatewayResponseId: completionId,
-    providerRequestId: (0, import_usage.extractProviderRequestId)(result)
-  });
+  const usage = (0, import_usage.setAiApiUsageResult)(
+    ctx,
+    result.usage_metadata,
+    {
+      gatewayResponseId: completionId,
+      providerRequestId: (0, import_usage.extractProviderRequestId)(result)
+    },
+    result.response_metadata
+  );
   ctx.status = 200;
   const toolCalls = normalizeToolCalls(result.tool_calls);
+  const providerFinishReason = extractFinishReason(result);
   ctx.body = (0, import_openai_format.toOpenAIResponse)({
     id: completionId,
     model: modelName,
     content,
     usage: usage ?? { prompt_tokens: 0, completion_tokens: 0, total_tokens: 0 },
-    toolCalls
+    toolCalls,
+    ...providerFinishReason ? { finishReason: providerFinishReason } : {}
   });
 }
 async function handleStreamingCompletion(ctx, chatModel, messages, completionId, modelName, providerRequestParameters) {
@@ -287,8 +280,10 @@ async function handleStreamingCompletion(ctx, chatModel, messages, completionId,
   );
   const requestAbort = (0, import_streaming.createRequestAbortController)(ctx);
   let usage;
+  let usageResponseMetadata;
   let providerRequestId;
-  let finishReason = "stop";
+  let providerFinishReason;
+  let sawToolCalls = false;
   try {
     const stream = await chatModel.stream(messages, { ...providerRequestParameters, signal: requestAbort.signal });
     for await (const chunk of stream) {
@@ -316,7 +311,7 @@ async function handleStreamingCompletion(ctx, chatModel, messages, completionId,
       const toolCallChunks = normalizeToolCallChunks(chunk.tool_call_chunks);
       if (toolCallChunks.length) {
         (0, import_app_observability.markAiApiFirstProviderOutput)(ctx);
-        finishReason = "tool_calls";
+        sawToolCalls = true;
         await (0, import_streaming.writeResponse)(
           ctx,
           (0, import_openai_format.formatSSE)(
@@ -329,8 +324,14 @@ async function handleStreamingCompletion(ctx, chatModel, messages, completionId,
         );
       }
       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 = extractFinishReason(chunk);
+      if (chunkFinishReason) providerFinishReason = chunkFinishReason;
       providerRequestId = providerRequestId ?? (0, import_usage.extractProviderRequestId)(chunk);
     }
     await (0, import_streaming.writeResponse)(
@@ -340,7 +341,7 @@ async function handleStreamingCompletion(ctx, chatModel, messages, completionId,
           id: completionId,
           model: modelName,
           delta: {},
-          finishReason
+          finishReason: providerFinishReason ?? (sawToolCalls ? "tool_calls" : "stop")
         })
       )
     );
@@ -357,7 +358,7 @@ async function handleStreamingCompletion(ctx, chatModel, messages, completionId,
       );
     }
     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);
@@ -373,7 +374,7 @@ async function handleStreamingCompletion(ctx, chatModel, messages, completionId,
         })
       );
     }
-    (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,
@@ -387,8 +388,20 @@ async function handleStreamingCompletion(ctx, chatModel, messages, completionId,
 function getErrorMessage(error, fallback) {
   return error instanceof Error && error.message ? error.message : fallback;
 }
-const SUPPORTED_CONTENT_BLOCK_TYPES = /* @__PURE__ */ new Set(["text", "image_url"]);
+function extractFinishReason(value) {
+  if (!isRecord(value)) return void 0;
+  const candidates = [
+    isRecord(value.response_metadata) ? value.response_metadata.finish_reason : void 0,
+    isRecord(value.additional_kwargs) ? value.additional_kwargs.finish_reason : void 0
+  ];
+  for (const candidate of candidates) {
+    if (typeof candidate === "string" && candidate.length > 0) return candidate;
+  }
+  return void 0;
+}
+const SUPPORTED_CONTENT_BLOCK_TYPES = /* @__PURE__ */ new Set(["text", "image_url", "file", "file_url"]);
 const BASE64_DATA_URL_PATTERN = /^data:(\w+\/\w+);base64,([A-Za-z0-9+/]+=*)$/;
+const FILE_BASE64_DATA_URL_PATTERN = /^data:([^;\s]+);base64,([A-Za-z0-9+/]+=*)$/;
 function isDecodableBase64(payload) {
   try {
     return Buffer.from(payload, "base64").toString("base64") === payload;
@@ -440,13 +453,50 @@ function describeContentBlockProblem(block) {
   const type = typeof block.type === "string" ? block.type : void 0;
   if (!type) return "each content block requires a 'type' field";
   if (!SUPPORTED_CONTENT_BLOCK_TYPES.has(type)) {
-    return `content block type '${type}' is not supported \u2014 this gateway forwards 'text' and 'image_url' only. Send documents as text, or inline them as an 'image_url' data URL if the model reads images`;
+    return `content block type '${type}' is not supported \u2014 this gateway forwards 'text', 'image_url', 'file', and 'file_url' only. Send documents as text, or inline them as a 'file' / 'file_url' block`;
   }
   if (type === "text") {
     return typeof block.text === "string" ? void 0 : "a 'text' block requires a string 'text' field";
   }
+  if (type === "file") {
+    return describeFileProblem(block.file);
+  }
+  if (type === "file_url") {
+    return describeFileUrlProblem(block.file_url);
+  }
   return describeImageUrlProblem(block.image_url);
 }
+function describeFileProblem(file) {
+  if (!isRecord(file)) return "a 'file' block requires an object 'file' field";
+  const fileData = typeof file.file_data === "string" ? file.file_data : void 0;
+  if (!fileData) return "a 'file' block requires a string 'file.file_data' field";
+  if (!fileData.startsWith("data:")) {
+    return "a 'file' block's 'file_data' must be a base64 data URL starting with 'data:'";
+  }
+  const match = FILE_BASE64_DATA_URL_PATTERN.exec(fileData);
+  if (!match || !match[1].includes("/")) {
+    return `malformed base64 data URL. Expected 'data:<mime-type>;base64,<base64>' with a valid type/subtype and standard base64 (no whitespace or URL-safe characters)`;
+  }
+  if (!isDecodableBase64(match[2])) {
+    return `base64 payload is not decodable. Check the padding and length \u2014 the data must be a multiple of 4 characters with at most two trailing '='`;
+  }
+  return void 0;
+}
+function describeFileUrlProblem(fileUrl) {
+  if (!isRecord(fileUrl)) return "a 'file_url' block requires an object 'file_url' field";
+  const url = typeof fileUrl.url === "string" ? fileUrl.url : void 0;
+  if (!url) return "a 'file_url' block requires a string 'file_url.url' field";
+  let protocol;
+  try {
+    protocol = new URL(url).protocol;
+  } catch {
+    return `'${url}' is not a valid URL. Use an http(s) URL`;
+  }
+  if (protocol !== "http:" && protocol !== "https:") {
+    return `URL protocol '${protocol}' is not supported. Use an http(s) URL`;
+  }
+  return void 0;
+}
 function describeImageUrlProblem(imageUrl) {
   const url = typeof imageUrl === "string" ? imageUrl : isRecord(imageUrl) ? imageUrl.url : void 0;
   if (typeof url !== "string" || url === "") {
@@ -492,7 +542,35 @@ function normalizeMessageContent(content) {
   if (content === null || content === void 0) return "";
   return JSON.stringify(content);
 }
-const GATEWAY_MANAGED_PARAMETERS = /* @__PURE__ */ new Set(["model", "messages", "tools", "tool_choice", "stream", "n"]);
+async function processMessageContentFileBlocks(content, ctx, plugin) {
+  if (!Array.isArray(content)) return content;
+  const processed = [];
+  for (const block of content) {
+    processed.push(...await processFileBlockChain(block, ctx, plugin, 0));
+  }
+  return processed;
+}
+const MAX_FILE_PROCESSOR_CHAIN_DEPTH = 3;
+async function processFileBlockChain(block, ctx, plugin, depth) {
+  if (!isRecord(block) || block.type !== "file" && block.type !== "file_url") {
+    return [block];
+  }
+  if (depth > MAX_FILE_PROCESSOR_CHAIN_DEPTH) {
+    return [block];
+  }
+  const result = await plugin.fileProcessorService.process(block, { ctx });
+  const results = Array.isArray(result) ? result : [result];
+  const next = [];
+  for (const item of results) {
+    if (isRecord(item) && (item.type === "file" || item.type === "file_url")) {
+      next.push(...await processFileBlockChain(item, ctx, plugin, depth + 1));
+    } else {
+      next.push(item);
+    }
+  }
+  return next;
+}
+const GATEWAY_MANAGED_PARAMETERS = /* @__PURE__ */ new Set(["model", "messages", "prompt", "tools", "tool_choice", "stream", "n"]);
 function getProviderRequestParameters(body) {
   return Object.fromEntries(
     Object.entries(body).filter(([name, value]) => !GATEWAY_MANAGED_PARAMETERS.has(name) && value !== void 0)
@@ -562,6 +640,7 @@ function serializeToolArguments(value) {
 // Annotate the CommonJS export names for ESM import in node:
 0 && (module.exports = {
   applyProviderRequestParameters,
+  extractFinishReason,
   findContentBlockProblem,
   findMessageProblem,
   getProviderRequestParameters,

package/dist/server/routes/completions.js

@@ -33,6 +33,8 @@ 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_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");
@@ -90,7 +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;
     }
@@ -109,19 +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,
@@ -141,6 +131,8 @@ async function handleCompletions(ctx, plugin) {
     ]);
     const completionId = (0, import_openai_format.generateCompletionId)().replace("chatcmpl-", "cmpl-");
     const chatModel = provider.createModel();
+    const providerRequestParameters = (0, import_chat_completions.getProviderRequestParameters)(body);
+    (0, import_chat_completions.applyProviderRequestParameters)(chatModel, providerRequestParameters);
     (0, import_billing.markLlmProviderAttempted)(ctx);
     if (stream) {
       await handleStreamingTextCompletion(
@@ -149,10 +141,18 @@ async function handleCompletions(ctx, plugin) {
         langchainMessages,
         completionId,
         body.model,
-        body.stream_options
+        body.stream_options,
+        providerRequestParameters
       );
     } else {
-      await handleNonStreamingTextCompletion(ctx, chatModel, langchainMessages, completionId, body.model);
+      await handleNonStreamingTextCompletion(
+        ctx,
+        chatModel,
+        langchainMessages,
+        completionId,
+        body.model,
+        providerRequestParameters
+      );
     }
   } catch (err) {
     ctx.log.error("AI API completions error:", err);
@@ -170,8 +170,8 @@ async function handleCompletions(ctx, plugin) {
     }
   }
 }
-async function handleNonStreamingTextCompletion(ctx, chatModel, messages, completionId, modelName) {
-  const result = await chatModel.invoke(messages);
+async function handleNonStreamingTextCompletion(ctx, chatModel, messages, completionId, modelName, providerRequestParameters) {
+  const result = await chatModel.invoke(messages, providerRequestParameters);
   let text = "";
   if (typeof result.content === "string") {
     text = result.content;
@@ -179,10 +179,15 @@ async function handleNonStreamingTextCompletion(ctx, chatModel, messages, comple
     const textPart = result.content.find((c) => c.type === "text");
     text = (textPart == null ? void 0 : textPart.text) || JSON.stringify(result.content);
   }
-  const usage = (0, import_usage.setAiApiUsageResult)(ctx, result.usage_metadata, {
-    gatewayResponseId: completionId,
-    providerRequestId: (0, import_usage.extractProviderRequestId)(result)
-  });
+  const usage = (0, import_usage.setAiApiUsageResult)(
+    ctx,
+    result.usage_metadata,
+    {
+      gatewayResponseId: completionId,
+      providerRequestId: (0, import_usage.extractProviderRequestId)(result)
+    },
+    result.response_metadata
+  );
   ctx.status = 200;
   ctx.body = {
     id: completionId,
@@ -195,13 +200,18 @@ 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 ?? { prompt_tokens: 0, completion_tokens: 0, total_tokens: 0 }
+    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 }
+    } : { prompt_tokens: 0, completion_tokens: 0, total_tokens: 0, prompt_tokens_details: { cached_tokens: null } }
   };
 }
-async function handleStreamingTextCompletion(ctx, chatModel, messages, completionId, modelName, streamOptions) {
+async function handleStreamingTextCompletion(ctx, chatModel, messages, completionId, modelName, streamOptions, providerRequestParameters) {
   ctx.set({
     "Content-Type": "text/event-stream",
     "Cache-Control": "no-cache",
@@ -211,9 +221,12 @@ 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,
       stream_options: { ...streamOptions, include_usage: true },
       signal: requestAbort.signal
     });
@@ -249,8 +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)(
@@ -266,7 +285,7 @@ async function handleStreamingTextCompletion(ctx, chatModel, messages, completio
             text: "",
             index: 0,
             logprobs: null,
-            finish_reason: "stop"
+            finish_reason: providerFinishReason ?? "stop"
           }
         ],
         usage: null
@@ -286,7 +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);
@@ -302,7 +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") {