llmist 15.14.0 → 15.14.2

package/dist/index.cjs

@@ -5669,11 +5669,12 @@ var init_anthropic = __esm({
 });
 
 // src/providers/gemini-cache-manager.ts
-var import_node_crypto3, GeminiCacheManager;
+var import_node_crypto3, import_genai, GeminiCacheManager;
 var init_gemini_cache_manager = __esm({
   "src/providers/gemini-cache-manager.ts"() {
     "use strict";
     import_node_crypto3 = require("crypto");
+    import_genai = require("@google/genai");
     GeminiCacheManager = class {
       constructor(client) {
         this.client = client;
@@ -5718,7 +5719,12 @@ var init_gemini_cache_manager = __esm({
             config: {
               contents: cacheableContents,
               ttl,
-              displayName: `llmist-${scope}-${Date.now()}`
+              displayName: `llmist-${scope}-${Date.now()}`,
+              toolConfig: {
+                functionCallingConfig: {
+                  mode: import_genai.FunctionCallingConfigMode.NONE
+                }
+              }
             }
           });
           if (!response.name) {
@@ -6359,9 +6365,10 @@ function resolveGeminiThinkingConfig(reasoning, modelName) {
   if (!reasoning?.enabled) return void 0;
   const isGemini3 = modelName.includes("gemini-3");
   if (isGemini3) {
+    const levelMap = modelName.includes("pro") ? GEMINI3_PRO_THINKING_LEVEL : GEMINI3_FLASH_THINKING_LEVEL;
     return {
       thinkingConfig: {
-        thinkingLevel: GEMINI3_THINKING_LEVEL[reasoning.effort ?? "medium"]
+        thinkingLevel: levelMap[reasoning.effort ?? "medium"]
       }
     };
   }
@@ -6398,13 +6405,13 @@ function wrapPcmInWav(pcmData, sampleRate, bitsPerSample, numChannels) {
   return buffer;
 }
 function createGeminiProviderFromEnv() {
-  return createProviderFromEnv("GEMINI_API_KEY", import_genai.GoogleGenAI, GeminiGenerativeProvider);
+  return createProviderFromEnv("GEMINI_API_KEY", import_genai2.GoogleGenAI, GeminiGenerativeProvider);
 }
-var import_genai, GEMINI3_THINKING_LEVEL, GEMINI25_THINKING_BUDGET, GEMINI_ROLE_MAP, GeminiGenerativeProvider;
+var import_genai2, GEMINI3_PRO_THINKING_LEVEL, GEMINI3_FLASH_THINKING_LEVEL, GEMINI25_THINKING_BUDGET, GEMINI_ROLE_MAP, GeminiGenerativeProvider;
 var init_gemini = __esm({
   "src/providers/gemini.ts"() {
     "use strict";
-    import_genai = require("@google/genai");
+    import_genai2 = require("@google/genai");
     init_messages();
     init_base_provider();
     init_constants2();
@@ -6413,7 +6420,14 @@ var init_gemini = __esm({
     init_gemini_models();
     init_gemini_speech_models();
     init_utils();
-    GEMINI3_THINKING_LEVEL = {
+    GEMINI3_PRO_THINKING_LEVEL = {
+      none: "low",
+      low: "low",
+      medium: "high",
+      high: "high",
+      maximum: "high"
+    };
+    GEMINI3_FLASH_THINKING_LEVEL = {
       none: "minimal",
       low: "low",
       medium: "medium",
@@ -6535,7 +6549,7 @@ var init_gemini = __esm({
           model: options.model,
           contents: [{ role: "user", parts: [{ text: options.prompt }] }],
           config: {
-            responseModalities: [import_genai.Modality.IMAGE, import_genai.Modality.TEXT]
+            responseModalities: [import_genai2.Modality.IMAGE, import_genai2.Modality.TEXT]
           }
         });
         const images = [];
@@ -6583,7 +6597,7 @@ var init_gemini = __esm({
             }
           ],
           config: {
-            responseModalities: [import_genai.Modality.AUDIO],
+            responseModalities: [import_genai2.Modality.AUDIO],
             speechConfig: {
               voiceConfig: {
                 prebuiltVoiceConfig: {
@@ -6642,15 +6656,16 @@ var init_gemini = __esm({
           // Note: systemInstruction removed - it doesn't work with countTokens()
           // System messages are now included in contents as user+model exchanges
           ...generationConfig ? { ...generationConfig } : {},
-          // Explicitly disable function calling to prevent UNEXPECTED_TOOL_CALL errors
-          toolConfig: {
-            functionCallingConfig: {
-              mode: import_genai.FunctionCallingConfigMode.NONE
+          // When using cached content, toolConfig lives inside the cache resource.
+          // Gemini rejects requests that include both cachedContent and toolConfig.
+          ...cacheName ? { cachedContent: cacheName } : {
+            toolConfig: {
+              functionCallingConfig: {
+                mode: import_genai2.FunctionCallingConfigMode.NONE
+              }
             }
           },
           ...thinkingConfig ?? {},
-          // Add cache reference if available
-          ...cacheName ? { cachedContent: cacheName } : {},
           ...options.extra
         };
         return {