@adminforth/completion-adapter-google-gemini 2.0.8 → 2.1.0

package/dist/index.js

@@ -38,6 +38,20 @@ function getGoogleLangChainAgentOptions(extraRequestBodyParameters) {
     }
     return options;
 }
+function extractUsedTokens(usageMetadata) {
+    var _a, _b, _c, _d, _e, _f, _g, _h;
+    if (!usageMetadata)
+        return undefined;
+    const input = (_b = (_a = usageMetadata.promptTokenCount) !== null && _a !== void 0 ? _a : usageMetadata.prompt_token_count) !== null && _b !== void 0 ? _b : 0;
+    const inputCached = (_d = (_c = usageMetadata.cachedContentTokenCount) !== null && _c !== void 0 ? _c : usageMetadata.cached_content_token_count) !== null && _d !== void 0 ? _d : 0;
+    const output = (_f = (_e = usageMetadata.candidatesTokenCount) !== null && _e !== void 0 ? _e : usageMetadata.candidates_token_count) !== null && _f !== void 0 ? _f : 0;
+    const thoughts = (_h = (_g = usageMetadata.thoughtsTokenCount) !== null && _g !== void 0 ? _g : usageMetadata.thoughts_token_count) !== null && _h !== void 0 ? _h : 0;
+    return {
+        input_uncached: Math.max(input - inputCached, 0),
+        input_cached: inputCached,
+        output: output + thoughts,
+    };
+}
 export default class CompletionAdapterGoogleGemini {
     constructor(options) {
         this.complete = (requestOrContent_1, ...args_1) => __awaiter(this, [requestOrContent_1, ...args_1], void 0, function* (requestOrContent, stopOrMaxTokens = 50, maxTokensOrOutputSchema = 50, outputSchema) {
@@ -63,17 +77,30 @@ export default class CompletionAdapterGoogleGemini {
                 try {
                     const response = yield ai.models.generateContent({
                         model: this.options.model || "gemini-3-flash-preview",
-                        contents: content,
+                        contents: [
+                            {
+                                role: "user",
+                                parts: [{ text: content }],
+                            },
+                        ],
                         config: Object.assign({ responseJsonSchema: requestOutputSchema ? requestOutputSchema : undefined, maxOutputTokens: requestMaxTokens }, this.options.extraRequestBodyParameters),
                     });
                     logger.debug(`Google Gemini SUCCESSFUL API response: ${response}`);
                     return {
                         content: response.text,
+                        used_tokens: extractUsedTokens(response.usageMetadata),
                     };
                 }
                 catch (error) {
                     logger.error(`Error during Google Gemini API call: ${error}`);
-                    throw new Error(`Error during Google Gemini API call: ${JSON.parse(error.message).error.message}`);
+                    const errorMessage = error instanceof Error ? error.message : String(error);
+                    let googleErrorMessage = errorMessage;
+                    try {
+                        googleErrorMessage = JSON.parse(errorMessage).error.message;
+                    }
+                    catch (_a) {
+                    }
+                    throw new Error(`Error during Google Gemini API call: ${googleErrorMessage}`);
                 }
             });
             const result = yield pRetry(tryToGenerate, {

package/index.ts

@@ -13,6 +13,23 @@ type CompletionRequestInput = {
 
 type AgentModelPurpose = "primary" | "summary";
 
+type GeminiUsageMetadata = {
+  promptTokenCount?: number;
+  cachedContentTokenCount?: number;
+  candidatesTokenCount?: number;
+  thoughtsTokenCount?: number;
+  prompt_token_count?: number;
+  cached_content_token_count?: number;
+  candidates_token_count?: number;
+  thoughts_token_count?: number;
+};
+
+type UsedTokens = {
+  input_uncached: number;
+  input_cached: number;
+  output: number;
+};
+
 const GOOGLE_LANGCHAIN_AGENT_OPTION_KEYS = [
   "temperature",
   "topP",
@@ -47,6 +64,30 @@ function getGoogleLangChainAgentOptions(
   return options;
 }
 
+function extractUsedTokens(usageMetadata?: GeminiUsageMetadata): UsedTokens | undefined {
+  if (!usageMetadata) return undefined;
+
+  const input = usageMetadata.promptTokenCount ?? usageMetadata.prompt_token_count ?? 0;
+  const inputCached =
+    usageMetadata.cachedContentTokenCount ??
+    usageMetadata.cached_content_token_count ??
+    0;
+  const output =
+    usageMetadata.candidatesTokenCount ??
+    usageMetadata.candidates_token_count ??
+    0;
+  const thoughts =
+    usageMetadata.thoughtsTokenCount ??
+    usageMetadata.thoughts_token_count ??
+    0;
+
+  return {
+    input_uncached: Math.max(input - inputCached, 0),
+    input_cached: inputCached,
+    output: output + thoughts,
+  };
+}
+
 export default class CompletionAdapterGoogleGemini
   implements CompletionAdapter
 {
@@ -61,7 +102,7 @@ export default class CompletionAdapterGoogleGemini
       throw new Error("geminiApiKey is required");
     }
   }
- 
+
   async measureTokensCount(content: string): Promise<number> {
     // Implement token counting logic here
     const ai = new GoogleGenAI({
@@ -82,7 +123,7 @@ export default class CompletionAdapterGoogleGemini
     const modelOptions = getGoogleLangChainAgentOptions(
       this.options.extraRequestBodyParameters,
     );
- 
+
     return {
       model: new ChatGoogleGenerativeAI({
         model: this.options.model || "gemini-3-flash-preview",
@@ -103,6 +144,7 @@ export default class CompletionAdapterGoogleGemini
     content?: string;
     finishReason?: string;
     error?: string;
+    used_tokens?: UsedTokens;
   }> => {
     const request =
       typeof requestOrContent === "string"
@@ -135,7 +177,12 @@ export default class CompletionAdapterGoogleGemini
       try {
         const response = await ai.models.generateContent({
           model: this.options.model || "gemini-3-flash-preview",
-          contents: content,
+          contents: [
+            {
+              role: "user",
+              parts: [{ text: content }],
+            },
+          ],
           config: {
             responseJsonSchema: requestOutputSchema ? requestOutputSchema : undefined,
             maxOutputTokens: requestMaxTokens,
@@ -145,10 +192,17 @@ export default class CompletionAdapterGoogleGemini
         logger.debug(`Google Gemini SUCCESSFUL API response: ${response}`);
         return {
           content: response.text,
+          used_tokens: extractUsedTokens(response.usageMetadata),
         };
       } catch (error) {
         logger.error(`Error during Google Gemini API call: ${error}`);
-        throw new Error(`Error during Google Gemini API call: ${JSON.parse(error.message).error.message}`);
+        const errorMessage = error instanceof Error ? error.message : String(error);
+        let googleErrorMessage = errorMessage;
+        try {
+          googleErrorMessage = JSON.parse(errorMessage).error.message;
+        } catch {
+        }
+        throw new Error(`Error during Google Gemini API call: ${googleErrorMessage}`);
       }
     }
     const result = await pRetry(tryToGenerate, {

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@adminforth/completion-adapter-google-gemini",
-  "version": "2.0.8",
+  "version": "2.1.0",
   "main": "dist/index.js",
   "types": "dist/index.d.ts",
   "type": "module",
@@ -16,7 +16,7 @@
   "description": "",
   "dependencies": {
     "@google/genai": "^1.35.0",
-    "@langchain/google-genai": "2.1.27",
+    "@langchain/google-genai": "2.1.28",
     "p-retry": "^7.1.1"
   },
   "peerDependencies": {