From 7e50e82d8cf8cd351809c9cb28b7ce0e5b3cda9a Mon Sep 17 00:00:00 2001
From: Storme-bit <tk.stomre@gmail.com>
Date: Fri, 17 Apr 2026 06:46:26 -0700
Subject: [PATCH] fix entity duplication glitch

---
 .../memory-service/src/entities/extraction.js | 25 ++++++++++++++-----
 1 file changed, 19 insertions(+), 6 deletions(-)

diff --git a/packages/memory-service/src/entities/extraction.js b/packages/memory-service/src/entities/extraction.js
index 048b848..499dee8 100644
--- a/packages/memory-service/src/entities/extraction.js
+++ b/packages/memory-service/src/entities/extraction.js
@@ -8,7 +8,15 @@ const EMBEDDING_SERVICE_URL = getEnv('EMBEDDING_SERVICE_URL', SERVICES.EMBEDDING
 
 const ENTITY_TYPES = ['person', 'place', 'project', 'technology', 'concept', 'organization'];
 
-function buildExtractionPrompt(userMessage, aiResponse) {
+function buildExtractionPrompt(userMessage, aiResponse, knownEntities = []) {
+    const knownBlock = knownEntities.length > 0
+        ? [
+            'Already known entities (use these exact name and type values if the same entity appears):',
+            ...knownEntities.map(e => `- "${e.name}" (${e.type})`),
+            '',
+          ].join('\n')
+        : '';
+
     return [
         '<|im_start|>system',
         'You are an entity extraction assistant. You extract named entities from conversations and return them as a JSON array. You never return empty arrays when entities are present. You only output valid JSON.<|im_end|>',
@@ -17,6 +25,7 @@ function buildExtractionPrompt(userMessage, aiResponse) {
         '',
         `Valid types: ${ENTITY_TYPES.join(', ')}`,
         '',
+        knownBlock,
         'Examples of entities to extract:',
         '- People: names of individuals',
         '- Projects: software projects, systems, tools being built',
@@ -25,8 +34,8 @@ function buildExtractionPrompt(userMessage, aiResponse) {
         '- Places: locations, servers, infrastructure',
         '',
         'Return a JSON array where each item has:',
-        '  "name": the entity name',
-        '  "type": one of the valid types above',
+        '  "name": the entity name (match exactly if already known)',
+        '  "type": one of the valid types above (match exactly if already known)',
         '  "notes": one sentence describing this entity based on the conversation',
         '',
         '### Conversation:',
@@ -58,15 +67,19 @@ async function embedEntity(entity) {
 async function extractAndStoreEntities(userMessage, aiResponse) {
     console.log('[entities] Extraction triggered')
     try {
+        // Fetch existing entities to guide the model toward consistent name/type pairs
+        const db = require('../db').getDB();
+        const knownEntities = db.prepare(`SELECT name, type FROM entities ORDER BY name`).all();
+
         const res = await fetch(`${EXTRACTION_URL}/api/generate`, {
             method: 'POST',
             headers: { 'Content-Type': 'application/json' },
             body: JSON.stringify({
                 model: EXTRACTION_MODEL,
-                prompt: buildExtractionPrompt(userMessage, aiResponse),
-                stream: false,      // we want the complete response, not a stream
+                prompt: buildExtractionPrompt(userMessage, aiResponse, knownEntities),
+                stream: false,
                 options: {
-                    temperature: 0.1,   // low temp for deterministic structured output
+                    temperature: 0.1,
                     num_predict: 512,
                 },
             }),