From e9244efb053e5a4bbd002edf818ddd3b88daeb66 Mon Sep 17 00:00:00 2001
From: parastejpal987-cmyk <parastejpal987@gmail.com>
Date: Sat, 1 Aug 2026 08:57:58 +0530
Subject: [PATCH 1/2] feat(khoj): Implement PgVectorStore to satisfy issue #287

---
 src/khoj/database/adapters/vector_store.py | 67 ++++++++++++++++++++++
 1 file changed, 67 insertions(+)
 create mode 100644 src/khoj/database/adapters/vector_store.py

diff --git a/src/khoj/database/adapters/vector_store.py b/src/khoj/database/adapters/vector_store.py
new file mode 100644
index 000000000..5b89a2276
--- /dev/null
+++ b/src/khoj/database/adapters/vector_store.py
@@ -0,0 +1,67 @@
+import math
+from typing import Optional, List
+
+from django.db.models import QuerySet, Q
+from pgvector.django import CosineDistance
+from torch import Tensor
+
+from khoj.database.models import KhojUser, Agent, UserMemory
+
+
+class PgVectorStore:
+    "\""
+    A PostgreSQL/pgvector backed store for managing and querying vector embeddings.
+    Provides a standardized interface for interacting with vector fields.
+    "\""
+
+    def __init__(self, user: KhojUser, agent: Optional[Agent] = None):
+        "\""
+        Initialize the vector store for a specific user and optionally a specific agent.
+        "\""
+        self.user = user
+        self.agent = agent
+
+    def add(self, raw_text: str, embedding: Tensor, search_model=None) -> UserMemory:
+        "\""
+        Add a new memory with its embedding to the store.
+        "\""
+        memory = UserMemory.objects.create(
+            user=self.user,
+            agent=self.agent,
+            raw=raw_text,
+            embeddings=embedding,
+            search_model=search_model,
+        )
+        return memory
+
+    def search(
+        self, query_embedding: Tensor, top_k: int = 10, max_distance: float = math.inf
+    ) -> QuerySet[UserMemory]:
+        "\""
+        Search the vector store for the closest embeddings using Cosine Distance.
+        "\""
+        # Base filter by user
+        owner_filter = Q(user=self.user)
+        
+        # If agent is specified, narrow it down
+        if self.agent:
+            owner_filter &= Q(agent=self.agent)
+
+        relevant_memories = (
+            UserMemory.objects.filter(owner_filter)
+            .annotate(distance=CosineDistance("embeddings", query_embedding))
+            .filter(distance__lte=max_distance)
+            .order_by("distance")
+        )
+
+        return relevant_memories[:top_k]
+
+    def delete(self, memory_id: int) -> bool:
+        "\""
+        Delete a specific memory from the store by ID.
+        Returns True if deleted, False if not found or unauthorized.
+        "\""
+        deleted_count, _ = UserMemory.objects.filter(
+            id=memory_id, user=self.user
+        ).delete()
+        return deleted_count > 0

From abf582177e47bd7f0178416c6d801b769f767c67 Mon Sep 17 00:00:00 2001
From: Owner <Owner@Indias-iMac.local>
Date: Sun, 16 Aug 2026 17:08:26 +0530
Subject: [PATCH 2/2] fix: resolve syntax errors caused by invalid docstring
 quotes

---
 src/khoj/database/adapters/vector_store.py | 26 +++++++++++-----------
 1 file changed, 13 insertions(+), 13 deletions(-)

diff --git a/src/khoj/database/adapters/vector_store.py b/src/khoj/database/adapters/vector_store.py
index 5b89a2276..03b12ccf0 100644
--- a/src/khoj/database/adapters/vector_store.py
+++ b/src/khoj/database/adapters/vector_store.py
@@ -1,30 +1,30 @@
 import math
-from typing import Optional, List
+from typing import Optional
 
-from django.db.models import QuerySet, Q
+from django.db.models import Q, QuerySet
 from pgvector.django import CosineDistance
 from torch import Tensor
 
-from khoj.database.models import KhojUser, Agent, UserMemory
+from khoj.database.models import Agent, KhojUser, UserMemory
 
 
 class PgVectorStore:
-    "\""
+    """
     A PostgreSQL/pgvector backed store for managing and querying vector embeddings.
     Provides a standardized interface for interacting with vector fields.
-    "\""
+    """
 
     def __init__(self, user: KhojUser, agent: Optional[Agent] = None):
-        "\""
+        """
         Initialize the vector store for a specific user and optionally a specific agent.
-        "\""
+        """
         self.user = user
         self.agent = agent
 
     def add(self, raw_text: str, embedding: Tensor, search_model=None) -> UserMemory:
-        "\""
+        """
         Add a new memory with its embedding to the store.
-        "\""
+        """
         memory = UserMemory.objects.create(
             user=self.user,
             agent=self.agent,
@@ -37,9 +37,9 @@ def add(self, raw_text: str, embedding: Tensor, search_model=None) -> UserMemory
     def search(
         self, query_embedding: Tensor, top_k: int = 10, max_distance: float = math.inf
     ) -> QuerySet[UserMemory]:
-        "\""
+        """
         Search the vector store for the closest embeddings using Cosine Distance.
-        "\""
+        """
         # Base filter by user
         owner_filter = Q(user=self.user)
         
@@ -57,10 +57,10 @@ def search(
         return relevant_memories[:top_k]
 
     def delete(self, memory_id: int) -> bool:
-        "\""
+        """
         Delete a specific memory from the store by ID.
         Returns True if deleted, False if not found or unauthorized.
-        "\""
+        """
         deleted_count, _ = UserMemory.objects.filter(
             id=memory_id, user=self.user
         ).delete()