plugin-ai-api 1.1.1 → 1.1.2

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

@@ -27,16 +27,21 @@ var __toCommonJS = (mod) => __copyProps(__defProp({}, "__esModule", { value: tru
 var chat_completions_exports = {};
 __export(chat_completions_exports, {
   applyProviderRequestParameters: () => applyProviderRequestParameters,
+  bindRequestTools: () => bindRequestTools,
   extractFinishReason: () => extractFinishReason,
   findContentBlockProblem: () => findContentBlockProblem,
   findMessageProblem: () => findMessageProblem,
   getProviderRequestParameters: () => getProviderRequestParameters,
   handleChatCompletions: () => handleChatCompletions,
-  normalizeMessageContent: () => normalizeMessageContent
+  normalizeMessageContent: () => normalizeMessageContent,
+  normalizeToolCallChunks: () => normalizeToolCallChunks,
+  normalizeToolCalls: () => normalizeToolCalls,
+  processMessageContentFileBlocks: () => processMessageContentFileBlocks
 });
 module.exports = __toCommonJS(chat_completions_exports);
 var import_openai_format = require("../utils/openai-format");
 var import_resolve_service = require("../utils/resolve-service");
+var import_virtual_models = require("../utils/virtual-models");
 var import_streaming = require("../utils/streaming");
 var import_user_permissions = require("../utils/user-permissions");
 var import_request_cache = require("../utils/request-cache");
@@ -91,7 +96,16 @@ async function handleChatCompletions(ctx, plugin) {
     return;
   }
   const stream = (0, import_streaming.isStreamingRequested)(body.stream);
-  const resolved = await (0, import_resolve_service.resolveModelString)(ctx, body.model);
+  const virtual = await (0, import_virtual_models.resolveVirtualModel)(ctx, body.model, body, "chat");
+  if ((virtual == null ? void 0 : virtual.status) === "unavailable") {
+    (0, import_virtual_models.respondVirtualModelUnavailable)(ctx, virtual);
+    return;
+  }
+  if ((virtual == null ? void 0 : virtual.status) === "resolved") {
+    ctx.state.aiApiVirtualModel = virtual.virtualModel;
+    ctx.state.aiApiRoutingReason = virtual.reason;
+  }
+  const resolved = (virtual == null ? void 0 : virtual.resolved) ?? await (0, import_resolve_service.resolveModelString)(ctx, body.model);
   if (!resolved) {
     ctx.status = 404;
     ctx.body = (0, import_openai_format.toOpenAIError)(
@@ -199,7 +213,7 @@ async function handleChatCompletions(ctx, plugin) {
         chatModel,
         langchainMessages,
         completionId,
-        body.model,
+        `${service.name}/${modelId}`,
         providerRequestParameters
       );
     } else {
@@ -208,7 +222,7 @@ async function handleChatCompletions(ctx, plugin) {
         chatModel,
         langchainMessages,
         completionId,
-        body.model,
+        `${service.name}/${modelId}`,
         providerRequestParameters
       );
     }
@@ -570,7 +584,21 @@ async function processFileBlockChain(block, ctx, plugin, depth) {
   }
   return next;
 }
-const GATEWAY_MANAGED_PARAMETERS = /* @__PURE__ */ new Set(["model", "messages", "prompt", "tools", "tool_choice", "stream", "n"]);
+const GATEWAY_MANAGED_PARAMETERS = /* @__PURE__ */ new Set([
+  "model",
+  "messages",
+  "prompt",
+  "tools",
+  "tool_choice",
+  "stream",
+  "n",
+  // Responses API specific fields that should not be passed to providers
+  "input",
+  "previous_response_id",
+  "store",
+  "truncation",
+  "metadata"
+]);
 function getProviderRequestParameters(body) {
   return Object.fromEntries(
     Object.entries(body).filter(([name, value]) => !GATEWAY_MANAGED_PARAMETERS.has(name) && value !== void 0)
@@ -640,10 +668,14 @@ function serializeToolArguments(value) {
 // Annotate the CommonJS export names for ESM import in node:
 0 && (module.exports = {
   applyProviderRequestParameters,
+  bindRequestTools,
   extractFinishReason,
   findContentBlockProblem,
   findMessageProblem,
   getProviderRequestParameters,
   handleChatCompletions,
-  normalizeMessageContent
+  normalizeMessageContent,
+  normalizeToolCallChunks,
+  normalizeToolCalls,
+  processMessageContentFileBlocks
 });

package/dist/server/routes/completions.js

@@ -31,6 +31,7 @@ __export(completions_exports, {
 module.exports = __toCommonJS(completions_exports);
 var import_openai_format = require("../utils/openai-format");
 var import_resolve_service = require("../utils/resolve-service");
+var import_virtual_models = require("../utils/virtual-models");
 var import_user_permissions = require("../utils/user-permissions");
 var import_streaming = require("../utils/streaming");
 var import_chat_completions = require("./chat-completions");
@@ -63,7 +64,16 @@ async function handleCompletions(ctx, plugin) {
     return;
   }
   const stream = (0, import_streaming.isStreamingRequested)(body.stream);
-  const resolved = await (0, import_resolve_service.resolveModelString)(ctx, body.model);
+  const virtual = await (0, import_virtual_models.resolveVirtualModel)(ctx, body.model, body, "chat");
+  if ((virtual == null ? void 0 : virtual.status) === "unavailable") {
+    (0, import_virtual_models.respondVirtualModelUnavailable)(ctx, virtual);
+    return;
+  }
+  if ((virtual == null ? void 0 : virtual.status) === "resolved") {
+    ctx.state.aiApiVirtualModel = virtual.virtualModel;
+    ctx.state.aiApiRoutingReason = virtual.reason;
+  }
+  const resolved = (virtual == null ? void 0 : virtual.resolved) ?? await (0, import_resolve_service.resolveModelString)(ctx, body.model);
   if (!resolved) {
     ctx.status = 404;
     ctx.body = (0, import_openai_format.toOpenAIError)(
@@ -108,7 +118,8 @@ async function handleCompletions(ctx, plugin) {
     };
     if (body.temperature !== void 0) modelOptions.temperature = body.temperature;
     if (body.top_p !== void 0) modelOptions.topP = body.top_p;
-    if (body.max_tokens !== void 0) modelOptions.maxTokens = body.max_tokens;
+    if (body.max_completion_tokens !== void 0) modelOptions.maxTokens = body.max_completion_tokens;
+    else 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 = [{ role: "user", content: prompt }];
@@ -116,6 +127,7 @@ async function handleCompletions(ctx, plugin) {
       serviceName: service.name,
       modelId,
       messages,
+      maxCompletionTokens: body.max_completion_tokens,
       maxTokens: body.max_tokens
     });
     await (0, import_billing.prepareLlmBilling)(ctx, resolved);
@@ -140,7 +152,7 @@ async function handleCompletions(ctx, plugin) {
         chatModel,
         langchainMessages,
         completionId,
-        body.model,
+        `${service.name}/${modelId}`,
         body.stream_options,
         providerRequestParameters
       );
@@ -150,7 +162,7 @@ async function handleCompletions(ctx, plugin) {
         chatModel,
         langchainMessages,
         completionId,
-        body.model,
+        `${service.name}/${modelId}`,
         providerRequestParameters
       );
     }

package/dist/server/routes/embeddings.js

@@ -31,6 +31,7 @@ __export(embeddings_exports, {
 module.exports = __toCommonJS(embeddings_exports);
 var import_openai_format = require("../utils/openai-format");
 var import_resolve_service = require("../utils/resolve-service");
+var import_virtual_models = require("../utils/virtual-models");
 var import_user_permissions = require("../utils/user-permissions");
 var import_request_cache = require("../utils/request-cache");
 var import_usage = require("../usage");
@@ -82,7 +83,16 @@ async function handleEmbeddings(ctx, plugin) {
     ctx.body = (0, import_openai_format.toOpenAIError)(400, "'input' must be a string or array of strings", "invalid_request_error");
     return;
   }
-  const resolved = await (0, import_resolve_service.resolveModelString)(ctx, body.model);
+  const virtual = await (0, import_virtual_models.resolveVirtualModel)(ctx, body.model, body, "embedding");
+  if ((virtual == null ? void 0 : virtual.status) === "unavailable") {
+    (0, import_virtual_models.respondVirtualModelUnavailable)(ctx, virtual);
+    return;
+  }
+  if ((virtual == null ? void 0 : virtual.status) === "resolved") {
+    ctx.state.aiApiVirtualModel = virtual.virtualModel;
+    ctx.state.aiApiRoutingReason = virtual.reason;
+  }
+  const resolved = (virtual == null ? void 0 : virtual.resolved) ?? await (0, import_resolve_service.resolveModelString)(ctx, body.model);
   if (!resolved) {
     ctx.status = 404;
     ctx.body = (0, import_openai_format.toOpenAIError)(
@@ -94,7 +104,6 @@ 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)(
@@ -137,6 +146,7 @@ async function handleEmbeddings(ctx, plugin) {
     return;
   }
   try {
+    await (0, import_billing.prepareLlmBilling)(ctx, resolved);
     const EmbeddingClass = providerMeta.embedding;
     const embeddingProvider = new EmbeddingClass({
       app: ctx.app,
@@ -155,11 +165,10 @@ async function handleEmbeddings(ctx, plugin) {
       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");
     ctx.body = (0, import_openai_format.toOpenAIEmbeddingsResponse)({
-      model: body.model,
+      model: `${service.name}/${modelId}`,
       embeddings: vectors,
       // LangChain's EmbeddingsInterface does not expose token counts.
       promptTokens: null
@@ -167,11 +176,21 @@ async function handleEmbeddings(ctx, plugin) {
   } catch (err) {
     ctx.log.error("AI API embeddings error:", err);
     if (!ctx.res.headersSent) {
-      ctx.status = 500;
-      ctx.body = (0, import_openai_format.toOpenAIError)(500, err.message || "Failed to generate embeddings", "server_error");
+      const isQuotaError = err instanceof import_billing.AiApiQuotaError;
+      ctx.status = isQuotaError ? 429 : 500;
+      if (isQuotaError) ctx.set("X-RateLimit-Reason", err.code);
+      ctx.body = (0, import_openai_format.toOpenAIError)(
+        ctx.status,
+        getErrorMessage(err, "Failed to generate embeddings"),
+        isQuotaError ? "quota_error" : "server_error",
+        isQuotaError ? err.code : void 0
+      );
     }
   }
 }
+function getErrorMessage(error, fallback) {
+  return error instanceof Error && error.message ? error.message : fallback;
+}
 // Annotate the CommonJS export names for ESM import in node:
 0 && (module.exports = {
   handleEmbeddings

package/dist/server/routes/models.js

@@ -27,6 +27,7 @@ var __toCommonJS = (mod) => __copyProps(__defProp({}, "__esModule", { value: tru
 var models_exports = {};
 __export(models_exports, {
   buildModelObject: () => buildModelObject,
+  buildVirtualModelObject: () => buildVirtualModelObject,
   handleGetModel: () => handleGetModel,
   handleListModels: () => handleListModels
 });
@@ -34,6 +35,7 @@ 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");
+var import_virtual_models = require("../utils/virtual-models");
 async function handleListModels(ctx, plugin) {
   var _a;
   try {
@@ -73,6 +75,13 @@ async function handleListModels(ctx, plugin) {
         models.push(buildModelObject(fullId, now, serviceLabel, meta));
       }
     }
+    try {
+      const virtualModels = await (0, import_virtual_models.listAccessibleVirtualModels)(ctx, scope, config == null ? void 0 : config.enabledLlmServices);
+      for (const virtualModel of virtualModels) {
+        models.push(buildVirtualModelObject(virtualModel, now));
+      }
+    } catch {
+    }
     ctx.status = 200;
     ctx.body = {
       object: "list",
@@ -121,6 +130,14 @@ async function handleGetModel(ctx, modelId, plugin) {
       }
       if (found) break;
     }
+    if (!found) {
+      try {
+        const virtualModels = await (0, import_virtual_models.listAccessibleVirtualModels)(ctx, scope, config == null ? void 0 : config.enabledLlmServices);
+        const virtualModel = virtualModels.find((candidate) => candidate.name === modelId);
+        if (virtualModel) found = buildVirtualModelObject(virtualModel, now);
+      } catch {
+      }
+    }
     if (!found) {
       ctx.status = 404;
       ctx.body = (0, import_openai_format.toOpenAIError)(404, `Model '${modelId}' not found`, "invalid_request_error", "model_not_found");
@@ -194,6 +211,17 @@ function buildModelObject(fullId, created, serviceLabel, meta) {
   }
   return model;
 }
+function buildVirtualModelObject(virtualModel, created) {
+  return {
+    id: virtualModel.name,
+    object: "model",
+    created,
+    owned_by: "ai-api-gateway",
+    virtual: true,
+    mode: virtualModel.mode,
+    description: `Virtual ${virtualModel.mode} alias routed to a concrete model based on the request shape.`
+  };
+}
 function toPositiveInt(value) {
   const n = Number(value);
   return Number.isSafeInteger(n) && n > 0 ? n : null;
@@ -235,6 +263,7 @@ function getRecommendedModelsForProvider(provider) {
 // Annotate the CommonJS export names for ESM import in node:
 0 && (module.exports = {
   buildModelObject,
+  buildVirtualModelObject,
   handleGetModel,
   handleListModels
 });