@contentgrowth/llm-service 1.0.3 → 1.0.5

package/dist/index.d.cts

@@ -200,6 +200,15 @@ declare class GeminiProvider extends BaseLLMProvider {
         done: boolean;
         error: any;
     }>;
+    /**
+     * Extract structured data from a file (PDF, Image, etc.) using Gemini Multimodal capabilities.
+     * @param {Buffer|string} fileData - Base64 string or Buffer of the file
+     * @param {string} mimeType - Mime type (e.g., 'application/pdf', 'image/png')
+     * @param {string} prompt - Extraction prompt
+     * @param {Object} schema - JSON schema for the output
+     * @param {Object} options - Additional options
+     */
+    extractWithLLM(fileData: Buffer | string, mimeType: string, prompt: string, schema?: any, options?: any): Promise<any>;
 }
 
 /**
@@ -331,6 +340,16 @@ declare class LLMService {
         content?: string;
         citations?: any[];
     }>;
+    /**
+     * Extract structured data from a file (Multimodal)
+     * @param {Buffer|string} fileData - File content (base64 or buffer)
+     * @param {string} mimeType - File mime type (e.g. 'application/pdf')
+     * @param {string} prompt - Extraction instructions
+     * @param {string} tenantId - Tenant ID
+     * @param {Object} schema - Optional JSON schema
+     * @param {Object} options - Options
+     */
+    extractWithLLM(fileData: Buffer | string, mimeType: string, prompt: string, tenantId: string, schema?: any, options?: any): Promise<any>;
 }
 declare class LLMServiceException extends Error {
     constructor(message: any, statusCode?: number, details?: any);

package/dist/index.d.ts

@@ -200,6 +200,15 @@ declare class GeminiProvider extends BaseLLMProvider {
         done: boolean;
         error: any;
     }>;
+    /**
+     * Extract structured data from a file (PDF, Image, etc.) using Gemini Multimodal capabilities.
+     * @param {Buffer|string} fileData - Base64 string or Buffer of the file
+     * @param {string} mimeType - Mime type (e.g., 'application/pdf', 'image/png')
+     * @param {string} prompt - Extraction prompt
+     * @param {Object} schema - JSON schema for the output
+     * @param {Object} options - Additional options
+     */
+    extractWithLLM(fileData: Buffer | string, mimeType: string, prompt: string, schema?: any, options?: any): Promise<any>;
 }
 
 /**
@@ -331,6 +340,16 @@ declare class LLMService {
         content?: string;
         citations?: any[];
     }>;
+    /**
+     * Extract structured data from a file (Multimodal)
+     * @param {Buffer|string} fileData - File content (base64 or buffer)
+     * @param {string} mimeType - File mime type (e.g. 'application/pdf')
+     * @param {string} prompt - Extraction instructions
+     * @param {string} tenantId - Tenant ID
+     * @param {Object} schema - Optional JSON schema
+     * @param {Object} options - Options
+     */
+    extractWithLLM(fileData: Buffer | string, mimeType: string, prompt: string, tenantId: string, schema?: any, options?: any): Promise<any>;
 }
 declare class LLMServiceException extends Error {
     constructor(message: any, statusCode?: number, details?: any);

package/dist/index.js

@@ -554,11 +554,13 @@ var GeminiProvider = class extends BaseLLMProvider {
     return { text: response.content };
   }
   async chatCompletion(messages, systemPrompt, tools = null, options = {}) {
+    const tier = options.tier || "default";
+    const modelName = this.models[tier] || this.defaultModel;
     return this._chatCompletionWithModel(
       messages,
       systemPrompt,
       tools,
-      this.defaultModel,
+      modelName,
       this.config.maxTokens,
       this.config.temperature,
       options
@@ -1020,6 +1022,65 @@ ${prompt}` : prompt;
       throw error;
     }
   }
+  /**
+   * Extract structured data from a file (PDF, Image, etc.) using Gemini Multimodal capabilities.
+   * @param {Buffer|string} fileData - Base64 string or Buffer of the file
+   * @param {string} mimeType - Mime type (e.g., 'application/pdf', 'image/png')
+   * @param {string} prompt - Extraction prompt
+   * @param {Object} schema - JSON schema for the output
+   * @param {Object} options - Additional options
+   */
+  async extractWithLLM(fileData, mimeType, prompt, schema = null, options = {}) {
+    var _a, _b, _c, _d;
+    const tier = options.tier || "default";
+    const modelName = this.models[tier] || this.defaultModel;
+    const maxTokens = options.maxTokens || this.config.maxTokens;
+    const temperature = options.temperature !== void 0 ? options.temperature : this.config.temperature;
+    const generationConfig = this._buildGenerationConfig(
+      { ...options, responseFormat: schema ? { type: "json_schema", schema } : "json" },
+      maxTokens,
+      temperature
+    );
+    const parts = [
+      { text: prompt }
+    ];
+    let base64Data = fileData;
+    if (typeof fileData !== "string") {
+      try {
+        base64Data = Buffer.from(fileData).toString("base64");
+      } catch (e) {
+      }
+    }
+    parts.push({
+      inlineData: {
+        data: base64Data,
+        mimeType
+      }
+    });
+    const requestOptions = {
+      model: modelName,
+      contents: [{
+        role: "user",
+        parts
+      }],
+      config: generationConfig
+    };
+    try {
+      const response = await this.client.models.generateContent(requestOptions);
+      const candidate = (_a = response.candidates) == null ? void 0 : _a[0];
+      if (!candidate) {
+        throw new LLMServiceException("No candidates returned from model during extraction", 500);
+      }
+      const textContent = ((_d = (_c = (_b = candidate.content) == null ? void 0 : _b.parts) == null ? void 0 : _c[0]) == null ? void 0 : _d.text) || "";
+      if (this._shouldAutoParse(options)) {
+        return this._safeJsonParse(textContent);
+      }
+      return textContent;
+    } catch (error) {
+      console.error(`[GeminiProvider] extractWithLLM failed (API Key: ${this._getMaskedApiKey()}):`, error);
+      throw error;
+    }
+  }
 };
 
 // src/llm-service.js
@@ -1247,6 +1308,22 @@ var LLMService = class {
     const provider = await this._getProvider(tenantId);
     return provider.getDeepResearchStatus(operationId);
   }
+  /**
+   * Extract structured data from a file (Multimodal)
+   * @param {Buffer|string} fileData - File content (base64 or buffer)
+   * @param {string} mimeType - File mime type (e.g. 'application/pdf')
+   * @param {string} prompt - Extraction instructions
+   * @param {string} tenantId - Tenant ID
+   * @param {Object} schema - Optional JSON schema
+   * @param {Object} options - Options
+   */
+  async extractWithLLM(fileData, mimeType, prompt, tenantId, schema = null, options = {}) {
+    const provider = await this._getProvider(tenantId);
+    if (!provider.extractWithLLM) {
+      throw new LLMServiceException(`Provider ${provider.config.provider} does not support extractWithLLM (Multimodal extraction)`, 400);
+    }
+    return provider.extractWithLLM(fileData, mimeType, prompt, schema, options);
+  }
 };
 var LLMServiceException = class extends Error {
   constructor(message, statusCode = 500, details = null) {
@@ -1495,9 +1572,17 @@ var TranscriptionService = class {
       const apiEndpoint = location === "global" ? "https://speech.googleapis.com" : `https://${location}-speech.googleapis.com`;
       const url = `${apiEndpoint}/v2/${recognizer}:recognize`;
       const languageCodes = language ? [language] : ["en-US"];
+      let decodingConfig = { autoDecodingConfig: {} };
+      if (audioFile.type && audioFile.type.includes("webm")) {
+        decodingConfig = {
+          explicitDecodingConfig: {
+            encoding: "WEBM_OPUS"
+          }
+        };
+      }
       const body = {
         config: {
-          autoDecodingConfig: {},
+          ...decodingConfig,
           languageCodes,
           // Sanitize model: Strict allowlist for Google v2 models
           model: model || "chirp_3"