memento-mcp-server 0.1.0 → 1.7.1

package/scripts/debug-embeddings.js

@@ -1,184 +1,184 @@
-#!/usr/bin/env node
-
-/**
- * 임베딩 디버깅 스크립트
- * 현재 데이터베이스의 임베딩 상태를 상세히 분석
- */
-
-import Database from 'better-sqlite3';
-import path from 'path';
-import { fileURLToPath } from 'url';
-
-const __filename = fileURLToPath(import.meta.url);
-const __dirname = path.dirname(__filename);
-
-// 데이터베이스 경로 설정
-const dbPath = process.env.DB_PATH || path.join(__dirname, '..', 'data', 'memory.db');
-
-async function debugEmbeddings() {
-  console.log('🔍 임베딩 상태 디버깅 시작...');
-  
-  // 데이터베이스 연결
-  const db = new Database(dbPath);
-  
-  try {
-    // 1. 전체 임베딩 통계
-    console.log('\n📊 전체 임베딩 통계:');
-    const totalStats = db.prepare(`
-      SELECT 
-        COUNT(*) as total_count,
-        AVG(dim) as avg_dim,
-        MIN(dim) as min_dim,
-        MAX(dim) as max_dim,
-        COUNT(DISTINCT dim) as unique_dims
-      FROM memory_embedding
-    `).get();
-
-    console.log(`- 총 임베딩 개수: ${totalStats.total_count}`);
-    console.log(`- 평균 차원: ${totalStats.avg_dim?.toFixed(1) || 'N/A'}`);
-    console.log(`- 최소 차원: ${totalStats.min_dim || 'N/A'}`);
-    console.log(`- 최대 차원: ${totalStats.max_dim || 'N/A'}`);
-    console.log(`- 고유 차원 수: ${totalStats.unique_dims}`);
-
-    // 2. 차원별 분포
-    console.log('\n📈 차원별 분포:');
-    const dimensionStats = db.prepare(`
-      SELECT 
-        dim,
-        COUNT(*) as count,
-        GROUP_CONCAT(memory_id) as memory_ids
-      FROM memory_embedding
-      GROUP BY dim
-      ORDER BY dim
-    `).all();
-
-    dimensionStats.forEach(stat => {
-      console.log(`- ${stat.dim}차원: ${stat.count}개`);
-      if (stat.count <= 5) {
-        console.log(`  메모리 ID: ${stat.memory_ids}`);
-      }
-    });
-
-    // 3. 모델별 분포
-    console.log('\n🤖 모델별 분포:');
-    const modelStats = db.prepare(`
-      SELECT 
-        model,
-        COUNT(*) as count,
-        AVG(dim) as avg_dim
-      FROM memory_embedding
-      GROUP BY model
-      ORDER BY count DESC
-    `).all();
-
-    modelStats.forEach(stat => {
-      console.log(`- ${stat.model || 'NULL'}: ${stat.count}개 (평균 ${stat.avg_dim?.toFixed(1)}차원)`);
-    });
-
-    // 4. 최근 생성된 임베딩들
-    console.log('\n🕒 최근 생성된 임베딩 (최대 10개):');
-    const recentEmbeddings = db.prepare(`
-      SELECT 
-        memory_id,
-        dim,
-        model,
-        created_at,
-        LENGTH(embedding) as embedding_length
-      FROM memory_embedding
-      ORDER BY created_at DESC
-      LIMIT 10
-    `).all();
-
-    recentEmbeddings.forEach(emb => {
-      console.log(`- ${emb.memory_id}: ${emb.dim}차원, ${emb.model}, ${emb.created_at}`);
-    });
-
-    // 5. 문제가 있는 임베딩 찾기
-    console.log('\n⚠️ 문제가 있을 수 있는 임베딩들:');
-    
-    // 차원이 0인 경우
-    const zeroDim = db.prepare(`
-      SELECT memory_id, dim, model FROM memory_embedding WHERE dim = 0
-    `).all();
-    
-    if (zeroDim.length > 0) {
-      console.log(`- 차원이 0인 임베딩: ${zeroDim.length}개`);
-      zeroDim.forEach(emb => {
-        console.log(`  ${emb.memory_id} (${emb.model})`);
-      });
-    }
-
-    // 차원이 매우 큰 경우 (1536보다 큰 경우)
-    const largeDim = db.prepare(`
-      SELECT memory_id, dim, model FROM memory_embedding WHERE dim > 1536
-    `).all();
-    
-    if (largeDim.length > 0) {
-      console.log(`- 차원이 1536보다 큰 임베딩: ${largeDim.length}개`);
-      largeDim.forEach(emb => {
-        console.log(`  ${emb.memory_id}: ${emb.dim}차원 (${emb.model})`);
-      });
-    }
-
-    // 6. 임베딩 데이터 샘플 확인
-    console.log('\n🔬 임베딩 데이터 샘플 (첫 3개):');
-    const samples = db.prepare(`
-      SELECT 
-        memory_id,
-        dim,
-        model,
-        SUBSTR(embedding, 1, 100) as embedding_preview
-      FROM memory_embedding
-      LIMIT 3
-    `).all();
-
-    samples.forEach((sample, index) => {
-      console.log(`\n샘플 ${index + 1}:`);
-      console.log(`- 메모리 ID: ${sample.memory_id}`);
-      console.log(`- 차원: ${sample.dim}`);
-      console.log(`- 모델: ${sample.model}`);
-      console.log(`- 임베딩 미리보기: ${sample.embedding_preview}...`);
-      
-      // 실제 벡터 길이 확인
-      try {
-        const fullEmbedding = db.prepare(`
-          SELECT embedding FROM memory_embedding WHERE memory_id = ?
-        `).get(sample.memory_id);
-        
-        const vector = JSON.parse(fullEmbedding.embedding);
-        console.log(`- 실제 벡터 길이: ${vector.length}`);
-        console.log(`- 첫 5개 값: [${vector.slice(0, 5).join(', ')}...]`);
-      } catch (error) {
-        console.log(`- 벡터 파싱 오류: ${error.message}`);
-      }
-    });
-
-    // 7. 메모리 아이템과의 연결 상태 확인
-    console.log('\n🔗 메모리 아이템 연결 상태:');
-    const connectionStats = db.prepare(`
-      SELECT 
-        (SELECT COUNT(*) FROM memory_item) as total_memories,
-        (SELECT COUNT(*) FROM memory_embedding) as total_embeddings,
-        (SELECT COUNT(*) FROM memory_item mi 
-         LEFT JOIN memory_embedding me ON mi.id = me.memory_id 
-         WHERE me.memory_id IS NULL) as memories_without_embedding
-    `).get();
-
-    console.log(`- 총 메모리 개수: ${connectionStats.total_memories}`);
-    console.log(`- 총 임베딩 개수: ${connectionStats.total_embeddings}`);
-    console.log(`- 임베딩이 없는 메모리: ${connectionStats.memories_without_embedding}`);
-
-  } catch (error) {
-    console.error('❌ 디버깅 실패:', error);
-    process.exit(1);
-  } finally {
-    db.close();
-  }
-}
-
-// 스크립트 실행
-if (import.meta.url === `file://${process.argv[1]}`) {
-  debugEmbeddings().catch(console.error);
-}
-
-export { debugEmbeddings };
+#!/usr/bin/env node
+
+/**
+ * 임베딩 디버깅 스크립트
+ * 현재 데이터베이스의 임베딩 상태를 상세히 분석
+ */
+
+import Database from 'better-sqlite3';
+import path from 'path';
+import { fileURLToPath } from 'url';
+
+const __filename = fileURLToPath(import.meta.url);
+const __dirname = path.dirname(__filename);
+
+// 데이터베이스 경로 설정
+const dbPath = process.env.DB_PATH || path.join(__dirname, '..', 'data', 'memory.db');
+
+async function debugEmbeddings() {
+  console.log('🔍 임베딩 상태 디버깅 시작...');
+  
+  // 데이터베이스 연결
+  const db = new Database(dbPath);
+  
+  try {
+    // 1. 전체 임베딩 통계
+    console.log('\n📊 전체 임베딩 통계:');
+    const totalStats = db.prepare(`
+      SELECT 
+        COUNT(*) as total_count,
+        AVG(dim) as avg_dim,
+        MIN(dim) as min_dim,
+        MAX(dim) as max_dim,
+        COUNT(DISTINCT dim) as unique_dims
+      FROM memory_embedding
+    `).get();
+
+    console.log(`- 총 임베딩 개수: ${totalStats.total_count}`);
+    console.log(`- 평균 차원: ${totalStats.avg_dim?.toFixed(1) || 'N/A'}`);
+    console.log(`- 최소 차원: ${totalStats.min_dim || 'N/A'}`);
+    console.log(`- 최대 차원: ${totalStats.max_dim || 'N/A'}`);
+    console.log(`- 고유 차원 수: ${totalStats.unique_dims}`);
+
+    // 2. 차원별 분포
+    console.log('\n📈 차원별 분포:');
+    const dimensionStats = db.prepare(`
+      SELECT 
+        dim,
+        COUNT(*) as count,
+        GROUP_CONCAT(memory_id) as memory_ids
+      FROM memory_embedding
+      GROUP BY dim
+      ORDER BY dim
+    `).all();
+
+    dimensionStats.forEach(stat => {
+      console.log(`- ${stat.dim}차원: ${stat.count}개`);
+      if (stat.count <= 5) {
+        console.log(`  메모리 ID: ${stat.memory_ids}`);
+      }
+    });
+
+    // 3. 모델별 분포
+    console.log('\n🤖 모델별 분포:');
+    const modelStats = db.prepare(`
+      SELECT 
+        model,
+        COUNT(*) as count,
+        AVG(dim) as avg_dim
+      FROM memory_embedding
+      GROUP BY model
+      ORDER BY count DESC
+    `).all();
+
+    modelStats.forEach(stat => {
+      console.log(`- ${stat.model || 'NULL'}: ${stat.count}개 (평균 ${stat.avg_dim?.toFixed(1)}차원)`);
+    });
+
+    // 4. 최근 생성된 임베딩들
+    console.log('\n🕒 최근 생성된 임베딩 (최대 10개):');
+    const recentEmbeddings = db.prepare(`
+      SELECT 
+        memory_id,
+        dim,
+        model,
+        created_at,
+        LENGTH(embedding) as embedding_length
+      FROM memory_embedding
+      ORDER BY created_at DESC
+      LIMIT 10
+    `).all();
+
+    recentEmbeddings.forEach(emb => {
+      console.log(`- ${emb.memory_id}: ${emb.dim}차원, ${emb.model}, ${emb.created_at}`);
+    });
+
+    // 5. 문제가 있는 임베딩 찾기
+    console.log('\n⚠️ 문제가 있을 수 있는 임베딩들:');
+    
+    // 차원이 0인 경우
+    const zeroDim = db.prepare(`
+      SELECT memory_id, dim, model FROM memory_embedding WHERE dim = 0
+    `).all();
+    
+    if (zeroDim.length > 0) {
+      console.log(`- 차원이 0인 임베딩: ${zeroDim.length}개`);
+      zeroDim.forEach(emb => {
+        console.log(`  ${emb.memory_id} (${emb.model})`);
+      });
+    }
+
+    // 차원이 매우 큰 경우 (1536보다 큰 경우)
+    const largeDim = db.prepare(`
+      SELECT memory_id, dim, model FROM memory_embedding WHERE dim > 1536
+    `).all();
+    
+    if (largeDim.length > 0) {
+      console.log(`- 차원이 1536보다 큰 임베딩: ${largeDim.length}개`);
+      largeDim.forEach(emb => {
+        console.log(`  ${emb.memory_id}: ${emb.dim}차원 (${emb.model})`);
+      });
+    }
+
+    // 6. 임베딩 데이터 샘플 확인
+    console.log('\n🔬 임베딩 데이터 샘플 (첫 3개):');
+    const samples = db.prepare(`
+      SELECT 
+        memory_id,
+        dim,
+        model,
+        SUBSTR(embedding, 1, 100) as embedding_preview
+      FROM memory_embedding
+      LIMIT 3
+    `).all();
+
+    samples.forEach((sample, index) => {
+      console.log(`\n샘플 ${index + 1}:`);
+      console.log(`- 메모리 ID: ${sample.memory_id}`);
+      console.log(`- 차원: ${sample.dim}`);
+      console.log(`- 모델: ${sample.model}`);
+      console.log(`- 임베딩 미리보기: ${sample.embedding_preview}...`);
+      
+      // 실제 벡터 길이 확인
+      try {
+        const fullEmbedding = db.prepare(`
+          SELECT embedding FROM memory_embedding WHERE memory_id = ?
+        `).get(sample.memory_id);
+        
+        const vector = JSON.parse(fullEmbedding.embedding);
+        console.log(`- 실제 벡터 길이: ${vector.length}`);
+        console.log(`- 첫 5개 값: [${vector.slice(0, 5).join(', ')}...]`);
+      } catch (error) {
+        console.log(`- 벡터 파싱 오류: ${error.message}`);
+      }
+    });
+
+    // 7. 메모리 아이템과의 연결 상태 확인
+    console.log('\n🔗 메모리 아이템 연결 상태:');
+    const connectionStats = db.prepare(`
+      SELECT 
+        (SELECT COUNT(*) FROM memory_item) as total_memories,
+        (SELECT COUNT(*) FROM memory_embedding) as total_embeddings,
+        (SELECT COUNT(*) FROM memory_item mi 
+         LEFT JOIN memory_embedding me ON mi.id = me.memory_id 
+         WHERE me.memory_id IS NULL) as memories_without_embedding
+    `).get();
+
+    console.log(`- 총 메모리 개수: ${connectionStats.total_memories}`);
+    console.log(`- 총 임베딩 개수: ${connectionStats.total_embeddings}`);
+    console.log(`- 임베딩이 없는 메모리: ${connectionStats.memories_without_embedding}`);
+
+  } catch (error) {
+    console.error('❌ 디버깅 실패:', error);
+    process.exit(1);
+  } finally {
+    db.close();
+  }
+}
+
+// 스크립트 실행
+if (import.meta.url === `file://${process.argv[1]}`) {
+  debugEmbeddings().catch(console.error);
+}
+
+export { debugEmbeddings };

package/scripts/direct-sql-migration.sql

@@ -1,36 +1,36 @@
--- 임베딩 데이터 마이그레이션 SQL
--- 기존 데이터에 메타데이터 추가
-
--- 1. 기존 데이터 업데이트
-UPDATE memory_embedding 
-SET 
-  embedding_provider = CASE 
-    WHEN model = 'lightweight-hybrid' THEN 'tfidf'
-    WHEN model IS NULL OR model = '' THEN 'tfidf'
-    ELSE 'unknown'
-  END,
-  dimensions = dim,
-  created_by = 'legacy'
-WHERE embedding_provider IS NULL;
-
--- 2. 인덱스 생성
-CREATE INDEX IF NOT EXISTS idx_memory_embedding_provider ON memory_embedding(embedding_provider);
-CREATE INDEX IF NOT EXISTS idx_memory_embedding_dimensions ON memory_embedding(dimensions);
-CREATE INDEX IF NOT EXISTS idx_memory_embedding_created_by ON memory_embedding(created_by);
-
--- 3. 검증 쿼리
-SELECT 
-  COUNT(*) as total,
-  COUNT(CASE WHEN embedding_provider IS NOT NULL THEN 1 END) as with_provider,
-  COUNT(CASE WHEN dimensions IS NOT NULL THEN 1 END) as with_dimensions,
-  COUNT(CASE WHEN created_by IS NOT NULL THEN 1 END) as with_created_by
-FROM memory_embedding;
-
--- 4. 최종 데이터 분포 확인
-SELECT 
-  embedding_provider,
-  dimensions,
-  COUNT(*) as count
-FROM memory_embedding 
-GROUP BY embedding_provider, dimensions
-ORDER BY count DESC;
+-- 임베딩 데이터 마이그레이션 SQL
+-- 기존 데이터에 메타데이터 추가
+
+-- 1. 기존 데이터 업데이트
+UPDATE memory_embedding 
+SET 
+  embedding_provider = CASE 
+    WHEN model = 'lightweight-hybrid' THEN 'tfidf'
+    WHEN model IS NULL OR model = '' THEN 'tfidf'
+    ELSE 'unknown'
+  END,
+  dimensions = dim,
+  created_by = 'legacy'
+WHERE embedding_provider IS NULL;
+
+-- 2. 인덱스 생성
+CREATE INDEX IF NOT EXISTS idx_memory_embedding_provider ON memory_embedding(embedding_provider);
+CREATE INDEX IF NOT EXISTS idx_memory_embedding_dimensions ON memory_embedding(dimensions);
+CREATE INDEX IF NOT EXISTS idx_memory_embedding_created_by ON memory_embedding(created_by);
+
+-- 3. 검증 쿼리
+SELECT 
+  COUNT(*) as total,
+  COUNT(CASE WHEN embedding_provider IS NOT NULL THEN 1 END) as with_provider,
+  COUNT(CASE WHEN dimensions IS NOT NULL THEN 1 END) as with_dimensions,
+  COUNT(CASE WHEN created_by IS NOT NULL THEN 1 END) as with_created_by
+FROM memory_embedding;
+
+-- 4. 최종 데이터 분포 확인
+SELECT 
+  embedding_provider,
+  dimensions,
+  COUNT(*) as count
+FROM memory_embedding 
+GROUP BY embedding_provider, dimensions
+ORDER BY count DESC;

package/scripts/docker-migration.sh

@@ -1,105 +1,105 @@
-#!/bin/bash
-
-echo "🐳 도커 환경에서 임베딩 데이터 마이그레이션 시작..."
-
-# 1. 백업 생성
-echo "💾 백업 생성 중..."
-cp /app/data/memory.db /app/data/memory-backup-$(date +%s).db
-
-# 2. sqlite-vec 없이 새로운 데이터베이스 생성
-echo "📝 새로운 데이터베이스 생성 중..."
-sqlite3 /app/data/memory-new.db << 'EOF'
--- 메인 기억 테이블
-CREATE TABLE memory_item (
-  id TEXT PRIMARY KEY,
-  type TEXT CHECK (type IN ('working','episodic','semantic','procedural')) NOT NULL,
-  content TEXT NOT NULL,
-  importance REAL CHECK (importance >= 0 AND importance <= 1) DEFAULT 0.5,
-  privacy_scope TEXT CHECK (privacy_scope IN ('private','team','public')) DEFAULT 'private',
-  created_at TIMESTAMP DEFAULT CURRENT_TIMESTAMP,
-  last_accessed TIMESTAMP,
-  pinned BOOLEAN DEFAULT FALSE,
-  tags TEXT,
-  source TEXT,
-  view_count INTEGER DEFAULT 0,
-  cite_count INTEGER DEFAULT 0,
-  edit_count INTEGER DEFAULT 0
-);
-
--- 임베딩 저장 테이블 (새로운 컬럼 포함)
-CREATE TABLE memory_embedding (
-  id INTEGER PRIMARY KEY AUTOINCREMENT,
-  memory_id TEXT NOT NULL,
-  embedding TEXT NOT NULL,
-  dim INTEGER NOT NULL,
-  model TEXT,
-  created_at TIMESTAMP DEFAULT CURRENT_TIMESTAMP,
-  embedding_provider TEXT,
-  dimensions INTEGER,
-  created_by TEXT DEFAULT 'migration',
-  FOREIGN KEY (memory_id) REFERENCES memory_item(id) ON DELETE CASCADE,
-  UNIQUE(memory_id)
-);
-
--- 인덱스 생성
-CREATE INDEX idx_memory_embedding_provider ON memory_embedding(embedding_provider);
-CREATE INDEX idx_memory_embedding_dimensions ON memory_embedding(dimensions);
-CREATE INDEX idx_memory_embedding_created_by ON memory_embedding(created_by);
-EOF
-
-# 3. 데이터 복사
-echo "📊 데이터 복사 중..."
-sqlite3 /app/data/memory-new.db << 'EOF'
-ATTACH DATABASE '/app/data/memory.db' AS source;
-
--- memory_item 데이터 복사
-INSERT INTO memory_item 
-SELECT * FROM source.memory_item;
-
--- memory_embedding 데이터 복사 및 메타데이터 추가
-INSERT INTO memory_embedding (memory_id, embedding, dim, model, created_at, embedding_provider, dimensions, created_by)
-SELECT 
-  memory_id,
-  embedding,
-  dim,
-  model,
-  created_at,
-  CASE 
-    WHEN model = 'lightweight-hybrid' THEN 'tfidf'
-    WHEN model IS NULL OR model = '' THEN 'tfidf'
-    ELSE 'unknown'
-  END as embedding_provider,
-  dim as dimensions,
-  'legacy' as created_by
-FROM source.memory_embedding;
-
-DETACH DATABASE source;
-EOF
-
-# 4. 검증
-echo "🔍 마이그레이션 검증 중..."
-sqlite3 /app/data/memory-new.db << 'EOF'
-SELECT 
-  COUNT(*) as total,
-  COUNT(CASE WHEN embedding_provider IS NOT NULL THEN 1 END) as with_provider,
-  COUNT(CASE WHEN dimensions IS NOT NULL THEN 1 END) as with_dimensions,
-  COUNT(CASE WHEN created_by IS NOT NULL THEN 1 END) as with_created_by
-FROM memory_embedding;
-
-SELECT 
-  embedding_provider,
-  dimensions,
-  COUNT(*) as count
-FROM memory_embedding 
-GROUP BY embedding_provider, dimensions
-ORDER BY count DESC;
-EOF
-
-# 5. 원본 데이터베이스 교체
-echo "🔄 데이터베이스 교체 중..."
-mv /app/data/memory.db /app/data/memory-old.db
-mv /app/data/memory-new.db /app/data/memory.db
-
-echo "✅ 마이그레이션 완료!"
-echo "💾 백업 파일: /app/data/memory-backup-*.db"
-echo "🔄 롤백이 필요한 경우: mv /app/data/memory-old.db /app/data/memory.db"
+#!/bin/bash
+
+echo "🐳 도커 환경에서 임베딩 데이터 마이그레이션 시작..."
+
+# 1. 백업 생성
+echo "💾 백업 생성 중..."
+cp /app/data/memory.db /app/data/memory-backup-$(date +%s).db
+
+# 2. sqlite-vec 없이 새로운 데이터베이스 생성
+echo "📝 새로운 데이터베이스 생성 중..."
+sqlite3 /app/data/memory-new.db << 'EOF'
+-- 메인 기억 테이블
+CREATE TABLE memory_item (
+  id TEXT PRIMARY KEY,
+  type TEXT CHECK (type IN ('working','episodic','semantic','procedural')) NOT NULL,
+  content TEXT NOT NULL,
+  importance REAL CHECK (importance >= 0 AND importance <= 1) DEFAULT 0.5,
+  privacy_scope TEXT CHECK (privacy_scope IN ('private','team','public')) DEFAULT 'private',
+  created_at TIMESTAMP DEFAULT CURRENT_TIMESTAMP,
+  last_accessed TIMESTAMP,
+  pinned BOOLEAN DEFAULT FALSE,
+  tags TEXT,
+  source TEXT,
+  view_count INTEGER DEFAULT 0,
+  cite_count INTEGER DEFAULT 0,
+  edit_count INTEGER DEFAULT 0
+);
+
+-- 임베딩 저장 테이블 (새로운 컬럼 포함)
+CREATE TABLE memory_embedding (
+  id INTEGER PRIMARY KEY AUTOINCREMENT,
+  memory_id TEXT NOT NULL,
+  embedding TEXT NOT NULL,
+  dim INTEGER NOT NULL,
+  model TEXT,
+  created_at TIMESTAMP DEFAULT CURRENT_TIMESTAMP,
+  embedding_provider TEXT,
+  dimensions INTEGER,
+  created_by TEXT DEFAULT 'migration',
+  FOREIGN KEY (memory_id) REFERENCES memory_item(id) ON DELETE CASCADE,
+  UNIQUE(memory_id)
+);
+
+-- 인덱스 생성
+CREATE INDEX idx_memory_embedding_provider ON memory_embedding(embedding_provider);
+CREATE INDEX idx_memory_embedding_dimensions ON memory_embedding(dimensions);
+CREATE INDEX idx_memory_embedding_created_by ON memory_embedding(created_by);
+EOF
+
+# 3. 데이터 복사
+echo "📊 데이터 복사 중..."
+sqlite3 /app/data/memory-new.db << 'EOF'
+ATTACH DATABASE '/app/data/memory.db' AS source;
+
+-- memory_item 데이터 복사
+INSERT INTO memory_item 
+SELECT * FROM source.memory_item;
+
+-- memory_embedding 데이터 복사 및 메타데이터 추가
+INSERT INTO memory_embedding (memory_id, embedding, dim, model, created_at, embedding_provider, dimensions, created_by)
+SELECT 
+  memory_id,
+  embedding,
+  dim,
+  model,
+  created_at,
+  CASE 
+    WHEN model = 'lightweight-hybrid' THEN 'tfidf'
+    WHEN model IS NULL OR model = '' THEN 'tfidf'
+    ELSE 'unknown'
+  END as embedding_provider,
+  dim as dimensions,
+  'legacy' as created_by
+FROM source.memory_embedding;
+
+DETACH DATABASE source;
+EOF
+
+# 4. 검증
+echo "🔍 마이그레이션 검증 중..."
+sqlite3 /app/data/memory-new.db << 'EOF'
+SELECT 
+  COUNT(*) as total,
+  COUNT(CASE WHEN embedding_provider IS NOT NULL THEN 1 END) as with_provider,
+  COUNT(CASE WHEN dimensions IS NOT NULL THEN 1 END) as with_dimensions,
+  COUNT(CASE WHEN created_by IS NOT NULL THEN 1 END) as with_created_by
+FROM memory_embedding;
+
+SELECT 
+  embedding_provider,
+  dimensions,
+  COUNT(*) as count
+FROM memory_embedding 
+GROUP BY embedding_provider, dimensions
+ORDER BY count DESC;
+EOF
+
+# 5. 원본 데이터베이스 교체
+echo "🔄 데이터베이스 교체 중..."
+mv /app/data/memory.db /app/data/memory-old.db
+mv /app/data/memory-new.db /app/data/memory.db
+
+echo "✅ 마이그레이션 완료!"
+echo "💾 백업 파일: /app/data/memory-backup-*.db"
+echo "🔄 롤백이 필요한 경우: mv /app/data/memory-old.db /app/data/memory.db"