fix: delete Qdrant points by ID so memory deletions don't orphan vectors (#25495)

The Qdrant backends implemented delete(ids=...) as a payload filter on
metadata.id, but points are stored with the item id as the Qdrant point id
(see _create_points), and not every point carries an id in its payload.
Memory points store only {created_at} in metadata (KB metadata embeddings
likewise), so deleting a single memory matched nothing and left an orphaned
vector that kept being injected into RAG context.

Delete by point id instead: PointIdsList for the standard backend, and a
tenant-scoped HasIdCondition for multitenancy (point ids are unique, so tenant
isolation is preserved). Filter-based deletion is unchanged.

Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

backend/open_webui/retrieval/vector/dbs/qdrant.py

@@ -216,28 +216,23 @@ class QdrantClient(VectorDBBase):
         ids: Optional[list[str]] = None,
         filter: Optional[dict] = None,
     ):
-        # Delete the items from the collection based on the ids.
+        # Delete by point ID: the point ID is the item's id (see _create_points).
-        field_conditions = []
+        # Filtering on metadata.id silently misses points whose payload omits an
-
+        # id (e.g. memories), leaving orphaned vectors behind.
         if ids:
-            for id_value in ids:
+            return self.client.delete(
-                (
+                collection_name=f'{self.collection_prefix}_{collection_name}',
-                    field_conditions.append(
+                points_selector=models.PointIdsList(points=ids),
-                        models.FieldCondition(
+            )
-                            key='metadata.id',
+
-                            match=models.MatchValue(value=id_value),
+        field_conditions = []
-                        ),
+        if filter:
-                    ),
-                )
-        elif filter:
             for key, value in filter.items():
-                (
+                field_conditions.append(
-                    field_conditions.append(
+                    models.FieldCondition(
-                        models.FieldCondition(
+                        key=f'metadata.{key}',
-                            key=f'metadata.{key}',
+                        match=models.MatchValue(value=value),
-                            match=models.MatchValue(value=value),
+                    )
-                        ),
-                    ),
                 )
 
         return self.client.delete(

backend/open_webui/retrieval/vector/dbs/qdrant_multitenancy.py

@@ -228,15 +228,17 @@ class QdrantClient(VectorDBBase):
             return None
 
         must_conditions = [_tenant_filter(tenant_id)]
-        should_conditions = []
         if ids:
-            should_conditions = [_metadata_filter('id', id_value) for id_value in ids]
+            # Delete by point ID within the tenant. The point ID is the item's id
+            # (see _create_points); filtering on metadata.id silently misses points
+            # whose payload omits an id (e.g. memories), leaving orphaned vectors.
+            must_conditions.append(models.HasIdCondition(has_id=ids))
         elif filter:
             must_conditions += [_metadata_filter(k, v) for k, v in filter.items()]
 
         return self.client.delete(
             collection_name=mt_collection,
-            points_selector=models.FilterSelector(filter=models.Filter(must=must_conditions, should=should_conditions)),
+            points_selector=models.FilterSelector(filter=models.Filter(must=must_conditions)),
         )
 
     def search(