@tiledesk/tiledesk-tybot-connector 2.0.45 → 2.0.47

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@tiledesk/tiledesk-tybot-connector",
-  "version": "2.0.45",
+  "version": "2.0.47",
   "description": "Tiledesk Tybot connector",
   "main": "index.js",
   "scripts": {
@@ -32,6 +32,7 @@
     "mqtt": "^5.10.4",
     "multer": "^1.4.5-lts.1",
     "nanoid": "^3.1.25",
+    "path": "^0.12.7",
     "redis": "^4.7.0",
     "string-argv": "^0.3.2",
     "uuid": "^3.3.3",

package/services/KbService.js

@@ -73,9 +73,7 @@ class KBService {
     })
   }
 
-  async addUnansweredQuestion(id_project, namespace, question, token) {
-    
-    const json = { namespace, question };
+  async addUnansweredQuestion(id_project, data, token) {
     
     return new Promise((resolve, reject) => {
       const http_request = {
@@ -85,7 +83,7 @@ class KBService {
           'Authorization': 'JWT ' + token
         },
         method: "POST",
-        json: json
+        json: data
       }
       winston.debug("Kb HttpRequest", http_request);
 
@@ -98,6 +96,28 @@ class KBService {
       });
     });
   }
+
+  async addAnsweredQuestion(id_project, data, token) {
+    return new Promise((resolve, reject) => {
+      const http_request = {
+        url: API_ENDPOINT + "/" + id_project + "/kb/answered/",
+        headers: {
+          'Content-Type': 'application/json',
+          'Authorization': 'JWT ' + token
+        },
+        json: data,
+        method: "POST"
+      }
+      winston.debug("Kb HttpRequest", http_request);
+      httpUtils.request(http_request, (err, response) => {
+        if (err) {
+          reject(err);
+        } else {
+          resolve(response);
+        }
+      });
+    });
+  }
 }
 
 const kbService = new KBService();

package/tiledeskChatbotPlugs/directives/DirAskGPTV2.js

@@ -2,6 +2,7 @@ const axios = require("axios").default;
 const { TiledeskChatbot } = require('../../engine/TiledeskChatbot');
 const { Filler } = require('../Filler');
 let https = require("https");
+var path = require('path');
 const { DirIntent } = require("./DirIntent");
 const { TiledeskChatbotConst } = require("../../engine/TiledeskChatbotConst");
 const { TiledeskChatbotUtil } = require("../../utils/TiledeskChatbotUtil");
@@ -17,6 +18,40 @@ const aiController = require("../../services/AIController");
 const default_engine = require('../../config/kb/engine');
 const default_engine_hybrid = require('../../config/kb/engine.hybrid');
 const default_embedding = require("../../config/kb/embedding");
+const PromptManager = require('../../config/kb/prompt/rag/PromptManager');
+const { MODELS_MULTIPLIER } = require("../../utils/aiUtils");
+
+//const ragPromptManager = new PromptManager(path.join(__dirname, '../../config/kb/prompt/rag'));
+const ragPromptManager = new PromptManager(path.join(__dirname, '../../config/kb/prompt/rag'));
+
+const RAG_CONTEXT_ENV_OVERRIDES = {
+  "gpt-3.5-turbo":       process.env.GPT_3_5_CONTEXT,
+  "gpt-4":               process.env.GPT_4_CONTEXT,
+  "gpt-4-turbo-preview": process.env.GPT_4T_CONTEXT,
+  "gpt-4o":              process.env.GPT_4O_CONTEXT,
+  "gpt-4o-mini":         process.env.GPT_4O_MINI_CONTEXT,
+  "gpt-4.1":             process.env.GPT_4_1_CONTEXT,
+  "gpt-4.1-mini":        process.env.GPT_4_1_MINI_CONTEXT,
+  "gpt-4.1-nano":        process.env.GPT_4_1_NANO_CONTEXT,
+  "gpt-5":               process.env.GPT_5_CONTEXT,
+  "gpt-5-mini":          process.env.GPT_5_MINI_CONTEXT,
+  "gpt-5-nano":          process.env.GPT_5_NANO_CONTEXT,
+  "general":             process.env.GENERAL_CONTEXT
+};
+
+/** RAG system prompt per modello: file in config/kb/prompt/rag, sovrascrivibili via env (come prima). */
+function getRagContextTemplate(modelName) {
+  const envOverride = RAG_CONTEXT_ENV_OVERRIDES[modelName];
+  if (envOverride) {
+    return envOverride;
+  }
+  if (!PromptManager.modelMap[modelName] && process.env.GENERAL_CONTEXT) {
+    return process.env.GENERAL_CONTEXT;
+  }
+  return ragPromptManager.getPrompt(modelName);
+}
+
+const PINECONE_RERANKING = process.env.PINECONE_RERANKING === true || process.env.PINECONE_RERANKING === "true";
 
 class DirAskGPTV2 {
 
@@ -77,9 +112,9 @@ class DirAskGPTV2 {
     let namespace = this.context.projectId;
     let llm = "openai";
     let model;
-    let temperature;
-    let max_tokens;
-    let top_k;
+    let temperature = 0.7;
+    let max_tokens = 256;
+    let top_k = 4;
     let alpha;
     let transcript;
     let citations = false;
@@ -87,24 +122,10 @@ class DirAskGPTV2 {
     let engine;
     let embedding;
     let reranking;
+    let reranking_multiplier;
     let skip_unanswered = false;
-
-    let contexts = {
-      "gpt-3.5-turbo":        process.env.GPT_3_5_CONTEXT       || "You are an helpful assistant for question-answering tasks.\nUse ONLY the pieces of retrieved context delimited by #### and the chat history to answer the question.\nIf you don't know the answer, just say: \"I don't know<NOANS>\"\n\n####{context}####",
-      "gpt-4":                process.env.GPT_4_CONTEXT         || "You are an helpful assistant for question-answering tasks.\nUse ONLY the pieces of retrieved context delimited by #### and the chat history to answer the question.\nIf you don't know the answer, just say that you don't know.\nIf and only if none of the retrieved context is useful for your task, add this word to the end <NOANS>\n\n####{context}####",
-      "gpt-4-turbo-preview":  process.env.GPT_4T_CONTEXT        || "You are an helpful assistant for question-answering tasks.\nUse ONLY the pieces of retrieved context delimited by #### and the chat history to answer the question.\nIf you don't know the answer, just say that you don't know.\nIf and only if none of the retrieved context is useful for your task, add this word to the end <NOANS>\n\n####{context}####",
-      "gpt-4o":               process.env.GPT_4O_CONTEXT        || "You are an helpful assistant for question-answering tasks. Follow these steps carefully:\n1. Answer in the same language of the user question, regardless of the retrieved context language\n2. Use ONLY the pieces of the retrieved context and the chat history to answer the question.\n3. If the retrieved context does not contain sufficient information to generate an accurate and informative answer, return <NOANS>\n\n==Retrieved context start==\n{context}\n==Retrieved context end==",
-      "gpt-4o-mini":          process.env.GPT_4O_MINI_CONTEXT   || "You are an helpful assistant for question-answering tasks. Follow these steps carefully:\n1. Answer in the same language of the user question, regardless of the retrieved context language\n2. Use ONLY the pieces of the retrieved context and the chat history to answer the question.\n3. If the retrieved context does not contain sufficient information to generate an accurate and informative answer, return <NOANS>\n\n==Retrieved context start==\n{context}\n==Retrieved context end==",
-      "gpt-4.1":              process.env.GPT_4_1_CONTEXT       || "You are an helpful assistant for question-answering tasks. Follow these steps carefully:\n1. Answer in the same language of the user question, regardless of the retrieved context language\n2. Use ONLY the pieces of the retrieved context and the chat history to answer the question.\n3. If the retrieved context does not contain sufficient information to generate an accurate and informative answer, append <NOANS> at the end of the answer\n\n==Retrieved context start==\n{context}\n==Retrieved context end==",
-      "gpt-4.1-mini":         process.env.GPT_4_1_MINI_CONTEXT  || "You are an helpful assistant for question-answering tasks. Follow these steps carefully:\n1. Answer in the same language of the user question, regardless of the retrieved context language\n2. Use ONLY the pieces of the retrieved context and the chat history to answer the question.\n3. If the retrieved context does not contain sufficient information to generate an accurate and informative answer, append <NOANS> at the end of the answer\n\n==Retrieved context start==\n{context}\n==Retrieved context end==",
-      "gpt-4.1-nano":         process.env.GPT_4_1_NANO_CONTEXT  || "You are an helpful assistant for question-answering tasks. Follow these steps carefully:\n1. Answer in the same language of the user question, regardless of the retrieved context language\n2. Use ONLY the pieces of the retrieved context and the chat history to answer the question.\n3. If the retrieved context does not contain sufficient information to generate an accurate and informative answer, append <NOANS> at the end of the answer\n\n==Retrieved context start==\n{context}\n==Retrieved context end==",
-      "gpt-5":                process.env.GPT_5_CONTEXT         || "You are an helpful assistant for question-answering tasks. Follow these steps carefully:\n1. Answer in the same language of the user question, regardless of the retrieved context language\n2. Use ONLY the pieces of the retrieved context and the chat history to answer the question.\n3. If the retrieved context does not contain sufficient information to generate an accurate and informative answer, append <NOANS> at the end of the answer\n\n==Retrieved context start==\n{context}\n==Retrieved context end==",
-      "gpt-5-mini":           process.env.GPT_5_MINI_CONTEXT    || "You are an helpful assistant for question-answering tasks. Follow these steps carefully:\n1. Answer in the same language of the user question, regardless of the retrieved context language\n2. Use ONLY the pieces of the retrieved context and the chat history to answer the question.\n3. If the retrieved context does not contain sufficient information to generate an accurate and informative answer, append <NOANS> at the end of the answer\n\n==Retrieved context start==\n{context}\n==Retrieved context end==",
-      "gpt-5-nano":           process.env.GPT_5_NANO_CONTEXT    || "You are an helpful assistant for question-answering tasks. Follow these steps carefully:\n1. Answer in the same language of the user question, regardless of the retrieved context language\n2. Use ONLY the pieces of the retrieved context and the chat history to answer the question.\n3. If the retrieved context does not contain sufficient information to generate an accurate and informative answer, append <NOANS> at the end of the answer\n\n==Retrieved context start==\n{context}\n==Retrieved context end==",
-      "general":              process.env.GENERAL_CONTEXT       || "You are an helpful assistant for question-answering tasks. Follow these steps carefully:\n1. Answer in the same language of the user question, regardless of the retrieved context language\n2. Use ONLY the pieces of the retrieved context and the chat history to answer the question.\n3. If the retrieved context does not contain sufficient information to generate an accurate and informative answer, append <NOANS> at the end of the answer\n\n==Retrieved context start==\n{context}\n==Retrieved context end=="
-    }
-
     let source = null;
+
     if (!action.llm) {
       action.llm = "openai";
     }
@@ -152,6 +173,9 @@ class DirAskGPTV2 {
     if (action.reranking) {
       reranking = action.reranking;
     }
+    if (action.reranking_multiplier) {
+      reranking_multiplier = action.reranking_multiplier;
+    }
     if (action.skip_unanswered) {
       skip_unanswered = action.skip_unanswered;
     }
@@ -321,13 +345,43 @@ class DirAskGPTV2 {
       
       if (reranking === true) {
         json.reranking = true;
-        json.reranking_multiplier = 3;
+        json.reranking_multiplier = reranking_multiplier || 3;
         json.reranker_model = "cross-encoder/ms-marco-MiniLM-L-6-v2";
+
+        if ((top_k * reranking_multiplier) > 100) {
+          // Find the largest integer reranking_multiplier so that top_k * reranking_multiplier <= 100
+          let calculatedRerankingMultiplier = Math.floor(100 / top_k);
+          // At least 1 is required
+          if (calculatedRerankingMultiplier < 1) {
+            calculatedRerankingMultiplier = 1;
+          }
+          json.reranking_multiplier = calculatedRerankingMultiplier;
+        }
+      }
+    }
+
+    if (!ns.hybrid && reranking === true && PINECONE_RERANKING) {
+      json.reranking = {
+        "provider": "pinecone",
+        "api_key": process.env.PINECONE_API_KEY,
+        "model": process.env.PINECONE_RERANKING_MODEL || process.env.RERANKING_MODEL || "bge-reranker-v2-m3"
+      }
+
+      json.reranking_multiplier = reranking_multiplier || 3;
+      if ((top_k * reranking_multiplier) > 100) {
+        // Find the largest integer reranking_multiplier so that top_k * reranking_multiplier <= 100
+        let calculatedRerankingMultiplier = Math.floor(100 / top_k);
+        // At least 1 is required
+        if (calculatedRerankingMultiplier < 1) {
+          calculatedRerankingMultiplier = 1;
+        }
+        json.reranking_multiplier = calculatedRerankingMultiplier;
       }
     }
 
     if (!action.advancedPrompt) {
-      const contextTemplate = contexts[model.name] || contexts["general"];
+      const contextTemplate = getRagContextTemplate(model.name);
+      
       if (filled_context) {
         json.system_context = filled_context + "\n" + contextTemplate;
       } else {
@@ -398,15 +452,31 @@ class DirAskGPTV2 {
 
           } else {
             await this.#assignAttributes(action, resbody.answer, resbody.source, resbody.content_chunks);
+            let tokens = resbody.prompt_token_size;
             if (publicKey === true && !chunks_only) {
+
               let tokens_usage = {
                 tokens: resbody.prompt_token_size,
                 model: json.model
               }
+
+              let multiplier = MODELS_MULTIPLIER[json.model.name] ?? 1;
+              tokens = tokens * multiplier;
               quotasService.updateQuote(this.projectId, this.token, tokens_usage).catch((err) => {
                 winston.error("Error updating quota: ", err);
               })
             }
+            
+            const data = {
+              namespace: json.namespace,
+              question: json.question,
+              answer: resbody.answer,
+              request_id: this.requestId,
+              tokens: tokens
+            }
+            kbService.addAnsweredQuestion(this.projectId, data, this.token).catch((err) => {
+              winston.error("Error adding answered question: ", err);
+            })
   
             if (trueIntent) {
               await this.#executeCondition(true, trueIntent, trueIntentAttributes, falseIntent, falseIntentAttributes);
@@ -417,9 +487,16 @@ class DirAskGPTV2 {
             return;
           }
         } else {
+          winston.info("DirAskGPTV2 resbody else case: ", resbody);
           await this.#assignAttributes(action, answer, source);
           if (!skip_unanswered) {
-            kbService.addUnansweredQuestion(this.projectId, json.namespace, json.question, this.token).catch((err) => {
+            // console.log("this.context", JSON.stringify(this.context, null, 2));
+            const data = {
+              namespace: json.namespace,
+              question: json.question,
+              request_id: this.requestId
+            }
+            kbService.addUnansweredQuestion(this.projectId, data, this.token).catch((err) => {
               winston.error("DirAskGPTV2 - Error adding unanswered question: ", {
                 status: err.response?.status,
                 statusText: err.response?.statusText,

package/tiledeskChatbotPlugs/directives/DirIteration.js

@@ -22,7 +22,6 @@ class DirIteration {
 
   execute(directive, callback) {
     winston.verbose("Execute Iteration directive");
-    console.log("directive: ", directive);
     let action;
     if (directive.action) {
       action = directive.action;
@@ -83,7 +82,6 @@ class DirIteration {
    */
   async #initializeIteration(action, actionId, callback) {
     winston.debug("[Iteration] Initializing iteration state");
-    console.log("[Iteration] Initializing iteration state");
     
     const iterable = action.iterable;
     const goToIntent = action.goToIntent;
@@ -96,7 +94,6 @@ class DirIteration {
 
     if (!iterableValue) {
       winston.verbose("[Iteration] Iterable object is undefined");
-      console.log("[Iteration] Iterable object is undefined");
       this.logger.warn("[Iteration] Iterable object is undefined");
       callback(true);
       return;
@@ -107,7 +104,6 @@ class DirIteration {
     
     if (!iterableArray) {
       winston.verbose("[Iteration] Could not convert iterable to array");
-      console.log("[Iteration] Could not convert iterable to array");
       this.logger.error(`[Iteration] Could not convert iterable '${iterable}' to array | type: ${typeof iterableValue}`);
       callback(true);
       return;
@@ -115,7 +111,6 @@ class DirIteration {
 
     if (iterableArray.length === 0) {
       winston.verbose("[Iteration] Iterable array is empty. Exit...")
-      console.log("[Iteration] Iterable array is empty. Exit...")
       this.logger.warn("[Iteration] Iterable array is empty. Exit...")
       callback(true);
       return;
@@ -133,7 +128,6 @@ class DirIteration {
     
     await this.#saveIterationState(actionId, iterationState);
     this.logger.native(`[Iteration] Initialized iteration with ${iterableArray.length} items`);
-    console.log(`[Iteration] Initialized iteration with ${iterableArray.length} items`);
 
     // Process first item
     await this.#processCurrentItem(iterationState, actionId, goToIntent, output, callback);
@@ -141,7 +135,6 @@ class DirIteration {
 
   async #continueIteration(iterationState, actionId, callback) {
     winston.debug(`[Iteration] Continuing iteration from index ${iterationState.currentIndex}`);
-    console.log(`[Iteration] Continuing iteration from index ${iterationState.currentIndex}`);
     
     // Increment index
     iterationState.currentIndex += 1;
@@ -150,7 +143,6 @@ class DirIteration {
     if (iterationState.currentIndex >= iterationState.totalItems) {
       this.logger.native("[Iteration] Iteration completed");
       await this.#clearIterationState(actionId);
-      console.log("[Iteration] Iteration completed");
       callback(false);
       return;
     }

package/utils/aiUtils.js

@@ -0,0 +1,43 @@
+const winston = require('./winston');
+
+// MODELS_MULTIPLIER = {
+//     "gpt-3.5-turbo": 0.6,
+//     "gpt-4": 25,
+//     "gpt-4-turbo-preview": 12
+// }
+
+loadMultiplier();
+function loadMultiplier() {
+    
+
+    let models_string = process.env.AI_MODELS;
+    winston.debug("(loadMultiplier) models_string: ", models_string)
+    let models = {};
+
+    if (!models_string) {
+        winston.info("AI_MODELS not defined");
+        winston.info("AI Models: ", models)
+        return models;
+    }
+
+    let models_string_trimmed = models_string.replace(/ /g,'');
+    winston.debug("(loadMultiplier) models_string_trimmed: ", models_string_trimmed)
+
+    let splitted_string = models_string_trimmed.split(";");
+    winston.debug("splitted_string: ", splitted_string)
+
+    splitted_string.forEach(m => {
+        m_split = m.split(":");
+        if (!m_split[1]) {
+            multiplier = null;
+        } else {
+            multiplier = Number(m_split[1]);;
+        }
+        models[m_split[0]] = multiplier;
+    })
+
+    winston.info("AI Models: ", models)
+    return models;
+}
+
+module.exports = { MODELS_MULTIPLIER: loadMultiplier() }