@peopl-health/nexus 2.4.3 → 2.4.4

package/lib/helpers/assistantHelper.js

@@ -111,6 +111,7 @@ const runAssistantWithRetries = async (thread, assistant, runConfig, patientRepl
     assistant.setReplies(patientReply);
   }
 
+  const startTime = Date.now();
   let run, output, completed;
   let retries = 0;
   const maxRetries = DEFAULT_MAX_RETRIES;
@@ -131,11 +132,14 @@ const runAssistantWithRetries = async (thread, assistant, runConfig, patientRepl
     if (retries < maxRetries) await new Promise(resolve => setTimeout(resolve, 2000));
   } while (retries < maxRetries && (!completed || !output));
 
+  const predictionTimeMs = Date.now() - startTime;
+
   if (run?.last_error) console.log('[runAssistantWithRetries] RUN LAST ERROR:', run.last_error);
   console.log('[runAssistantWithRetries] RUN STATUS', completed);
   console.log('[runAssistantWithRetries] OUTPUT', output);
+  console.log('[runAssistantWithRetries] TIMING', { predictionTimeMs, retries });
   
-  return { run, output, completed, retries };
+  return { run, output, completed, retries, predictionTimeMs };
 };
 
 module.exports = {

package/lib/models/predictionMetricsModel.js

@@ -0,0 +1,20 @@
+const mongoose = require('mongoose');
+
+const predictionMetricsSchema = new mongoose.Schema({
+  message_id: { type: String, required: true, index: true },
+  numero: { type: String, required: true, index: true },
+  assistant_id: { type: String, required: true, index: true },
+  thread_id: { type: String, required: true },
+  prediction_time_ms: { type: Number, required: true },
+  retry_count: { type: Number, required: true, default: 1 },
+  completed: { type: Boolean, default: true },
+  error: { type: String, default: null }
+}, { timestamps: true });
+
+predictionMetricsSchema.index({ createdAt: -1 });
+predictionMetricsSchema.index({ assistant_id: 1, createdAt: -1 });
+predictionMetricsSchema.index({ numero: 1, createdAt: -1 });
+
+const PredictionMetrics = mongoose.model('PredictionMetrics', predictionMetricsSchema);
+
+module.exports = { PredictionMetrics, predictionMetricsSchema };

package/lib/services/assistantService.js

@@ -6,6 +6,7 @@ const { BaseAssistant } = require('../assistants/BaseAssistant');
 const { createProvider } = require('../providers/createProvider');
 
 const { Thread } = require('../models/threadModel.js');
+const { PredictionMetrics } = require('../models/predictionMetricsModel');
 
 const { getCurRow } = require('../helpers/assistantHelper.js');
 const { runAssistantAndWait, runAssistantWithRetries } = require('../helpers/assistantHelper.js');
@@ -316,7 +317,7 @@ const replyAssistantCore = async (code, message_ = null, thread_ = null, runOpti
   if (!patientMsg || thread.stopped) return null;
 
   const assistant = getAssistantById(thread.getAssistantId(), thread);
-  const { run, output, completed, retries } = await withTracing(
+  const { run, output, completed, retries, predictionTimeMs } = await withTracing(
     runAssistantWithRetries,
     'run_assistant_with_retries',
     (thread, assistant, runConfig, patientReply) => ({
@@ -326,6 +327,19 @@ const replyAssistantCore = async (code, message_ = null, thread_ = null, runOpti
     })
   )(thread, assistant, runOptions, patientReply);
 
+  // Store prediction metrics
+  if (output && predictionTimeMs) {
+    await PredictionMetrics.create({
+      message_id: `${code}-${Date.now()}`,
+      numero: code,
+      assistant_id: thread.getAssistantId(),
+      thread_id: thread.getConversationId(),
+      prediction_time_ms: predictionTimeMs,
+      retry_count: retries,
+      completed: completed
+    }).catch(err => console.error('[replyAssistantCore] Failed to store metrics:', err));
+  }
+
   return output;
 };
 

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@peopl-health/nexus",
-  "version": "2.4.3",
+  "version": "2.4.4",
   "description": "Core messaging and assistant library for WhatsApp communication platforms",
   "keywords": [
     "whatsapp",