update sqlite.ts to use VEC

Signed-off-by: MarcoMandar <malicemandar@gmail.com>

src/adapters/sqlite.ts

@@ -210,18 +210,16 @@ export class SqliteDatabaseAdapter extends DatabaseAdapter {
 
     // Insert the memory with the appropriate 'unique' value
     const sql = `INSERT OR REPLACE INTO memories (id, type, content, embedding, userId, roomId, \`unique\`, createdAt) VALUES (?, ?, ?, ?, ?, ?, ?, ?)`;
-    this.db
-      .prepare(sql)
-      .run(
-        memory.id ?? v4(),
-        tableName,
-        content,
-        JSON.stringify(memory.embedding ?? embeddingZeroVector),
-        memory.userId,
-        memory.roomId,
-        isUnique ? 1 : 0,
-        createdAt
-      );
+    this.db.prepare(sql).run(
+      memory.id ?? v4(),
+      tableName,
+      content,
+      new Float32Array(memory.embedding ?? embeddingZeroVector), // Store as Float32Array
+      memory.userId,
+      memory.roomId,
+      isUnique ? 1 : 0,
+      createdAt
+    );
   }
 
   async searchMemories(params: {
@@ -232,37 +230,36 @@ export class SqliteDatabaseAdapter extends DatabaseAdapter {
     match_count: number;
     unique: boolean;
   }): Promise<Memory[]> {
+    const queryParams = [
+      new Float32Array(params.embedding), // Ensure embedding is Float32Array
+      params.tableName,
+      params.roomId,
+      params.match_count,
+    ];
+
     let sql = `
-SELECT *, (1 - vss_distance_l2(embedding, ?)) AS similarity
-FROM memories
-WHERE type = ?
-AND roomId = ?`;
+      SELECT *, vec_distance_L2(embedding, ?) AS similarity
+      FROM memories
+      WHERE type = ?`;
 
     if (params.unique) {
       sql += " AND `unique` = 1";
     }
 
-    sql += ` ORDER BY similarity DESC LIMIT ?`;
-    const queryParams = [
-      JSON.stringify(params.embedding),
-      params.tableName,
-      params.roomId,
-      params.match_count,
-    ];
+    sql += ` ORDER BY similarity ASC LIMIT ?`; // ASC for lower distance
+    // Updated queryParams order matches the placeholders
 
     const memories = this.db.prepare(sql).all(...queryParams) as (Memory & {
       similarity: number;
     })[];
-    return memories.map((memory) => {
-      return {
-        ...memory,
-        createdAt:
-          typeof memory.createdAt === "string"
-            ? Date.parse(memory.createdAt as string)
-            : memory.createdAt,
-        content: JSON.parse(memory.content as unknown as string),
-      };
-    });
+    return memories.map((memory) => ({
+      ...memory,
+      createdAt:
+        typeof memory.createdAt === "string"
+          ? Date.parse(memory.createdAt as string)
+          : memory.createdAt,
+      content: JSON.parse(memory.content as unknown as string),
+    }));
   }
 
   async searchMemoriesByEmbedding(
@@ -276,15 +273,15 @@ AND roomId = ?`;
     }
   ): Promise<Memory[]> {
     const queryParams = [
-      JSON.stringify(embedding),
-      params.tableName,
       // JSON.stringify(embedding),
+      new Float32Array(embedding),
+      params.tableName,
     ];
 
     let sql = `
-      SELECT *, (1 - vss_distance_l2(embedding, ?)) AS similarity
+      SELECT *, vec_distance_L2(embedding, ?) AS similarity
       FROM memories
-      WHERE type = ?`; // AND vss_search(embedding, ?)
+      WHERE type = ?`;
 
     if (params.unique) {
       sql += " AND `unique` = 1";
@@ -330,21 +327,21 @@ AND roomId = ?`;
       SELECT *
       FROM memories
       WHERE type = ?
-      AND vss_search(${opts.query_field_name}, ?)
-      ORDER BY vss_search(${opts.query_field_name}, ?) DESC
+      AND vec_distance_L2(${opts.query_field_name}, ?) <= ?
+      ORDER BY vec_distance_L2(${opts.query_field_name}, ?) ASC
       LIMIT ?
     `;
-    const memories = this.db
-      .prepare(sql)
-      .all(
-        opts.query_table_name,
-        opts.query_input,
-        opts.query_input,
-        opts.query_match_count
-      ) as Memory[];
+    const memories = this.db.prepare(sql).all(
+      opts.query_table_name,
+      new Float32Array(opts.query_input.split(",").map(Number)), // Convert string to Float32Array
+      opts.query_input,
+      new Float32Array(opts.query_input.split(",").map(Number))
+    ) as Memory[];
 
     return memories.map((memory) => ({
-      embedding: JSON.parse(memory.embedding as unknown as string),
+      embedding: Array.from(
+        new Float32Array(memory.embedding as unknown as Buffer)
+      ), // Convert Buffer to number[]
       levenshtein_score: 0,
     }));
   }