fix: prevent repeated memories via archival and activated-only retrieval

src/memos/graph_dbs/base.py

@@ -19,12 +19,13 @@ def add_node(self, id: str, memory: str, metadata: dict[str, Any]) -> None:
         """
 
     @abstractmethod
-    def update_node(self, id: str, fields: dict[str, Any]) -> None:
+    def update_node(self, id: str, fields: dict[str, Any], user_name: str | None = None) -> None:
         """
         Update attributes of an existing node.
         Args:
             id: Node identifier to be updated.
             fields: Dictionary of fields to update.
+            user_name: given user_name
         """
 
     @abstractmethod
@@ -70,7 +71,7 @@ def edge_exists(self, source_id: str, target_id: str, type: str) -> bool:
 
     # Graph Query & Reasoning
     @abstractmethod
-    def get_node(self, id: str, include_embedding: bool = False) -> dict[str, Any] | None:
+    def get_node(self, id: str, include_embedding: bool = False, **kwargs) -> dict[str, Any] | None:
         """
         Retrieve the metadata and content of a node.
         Args:

src/memos/mem_reader/multi_modal_struct.py

@@ -474,13 +474,12 @@ def _get_maybe_merged_memory(
             )
 
             if not search_results:
-                # No similar memories found, return original
                 return extracted_memory_dict
 
             # Get full memory details
             similar_memory_ids = [r["id"] for r in search_results if r.get("id")]
             similar_memories_list = [
-                self.graph_db.get_node(mem_id, include_embedding=False)
+                self.graph_db.get_node(mem_id, include_embedding=False, user_name=user_name)
                 for mem_id in similar_memory_ids
             ]
 
@@ -505,7 +504,6 @@ def _get_maybe_merged_memory(
             )
 
             if not filtered_similar:
-                # No valid similar memories, return original
                 return extracted_memory_dict
 
             # Create a temporary TextualMemoryItem for merge check
@@ -529,14 +527,12 @@ def _get_maybe_merged_memory(
             if merge_result:
                 # Return merged memory dict
                 merged_dict = extracted_memory_dict.copy()
-                merged_dict["value"] = merge_result.get("value", mem_text)
-                merged_dict["merged_from"] = merge_result.get("merged_from", [])
-                logger.info(
-                    f"[MultiModalFine] Merged memory with {len(merged_dict['merged_from'])} existing memories"
-                )
+                merged_content = merge_result.get("value", mem_text)
+                merged_dict["value"] = merged_content
+                merged_from_ids = merge_result.get("merged_from", [])
+                merged_dict["merged_from"] = merged_from_ids
                 return merged_dict
             else:
-                # No merge needed, return original
                 return extracted_memory_dict
 
         except Exception as e:
@@ -648,6 +644,7 @@ def _process_one_item(fast_item: TextualMemoryItem) -> list[TextualMemoryItem]:
                             extracted_memory_dict=m,
                             mem_text=m.get("value", ""),
                             sources=sources,
+                            original_query=mem_str,
                             **kwargs,
                         )
                         # Normalize memory_type (same as simple_struct)
@@ -680,6 +677,7 @@ def _process_one_item(fast_item: TextualMemoryItem) -> list[TextualMemoryItem]:
                         extracted_memory_dict=resp,
                         mem_text=resp.get("value", "").strip(),
                         sources=sources,
+                        original_query=mem_str,
                         **kwargs,
                     )
                     node = self._make_memory_item(

src/memos/mem_scheduler/general_scheduler.py

@@ -899,7 +899,7 @@ def _process_memories_with_reader(
                     )
 
                     # Mark merged_from memories as archived when provided in memory metadata
-                    if self.mem_reader and self.mem_reader.graph_db:
+                    if self.mem_reader.graph_db:
                         for memory in flattened_memories:
                             merged_from = (memory.metadata.info or {}).get("merged_from")
                             if merged_from:

src/memos/memories/textual/tree_text_memory/retrieve/recall.py

@@ -77,6 +77,7 @@ def retrieve(
                 include_embedding=self.include_embedding,
                 user_name=user_name,
                 filter=search_filter,
+                status="activated",
             )
             return [TextualMemoryItem.from_dict(record) for record in working_memories[:top_k]]
 
@@ -247,7 +248,7 @@ def process_node(node):
 
             # Load nodes and post-filter
             node_dicts = self.graph_store.get_nodes(
-                list(candidate_ids), include_embedding=self.include_embedding
+                list(candidate_ids), include_embedding=self.include_embedding, user_name=user_name
             )
 
             final_nodes = []
@@ -277,7 +278,9 @@ def process_node(node):
                     {"field": "key", "op": "in", "value": parsed_goal.keys},
                     {"field": "memory_type", "op": "=", "value": memory_scope},
                 ]
-                key_ids = self.graph_store.get_by_metadata(key_filters, user_name=user_name)
+                key_ids = self.graph_store.get_by_metadata(
+                    key_filters, user_name=user_name, status="activated"
+                )
                 candidate_ids.update(key_ids)
 
             # 2) tag-based OR branch
@@ -286,7 +289,9 @@ def process_node(node):
                     {"field": "tags", "op": "contains", "value": parsed_goal.tags},
                     {"field": "memory_type", "op": "=", "value": memory_scope},
                 ]
-                tag_ids = self.graph_store.get_by_metadata(tag_filters, user_name=user_name)
+                tag_ids = self.graph_store.get_by_metadata(
+                    tag_filters, user_name=user_name, status="activated"
+                )
                 candidate_ids.update(tag_ids)
 
             # No matches → return empty
@@ -422,9 +427,11 @@ def _bm25_recall(
                 value = search_filter[key]
                 key_filters.append({"field": key, "op": "=", "value": value})
             corpus_name += "".join(list(search_filter.values()))
-        candidate_ids = self.graph_store.get_by_metadata(key_filters, user_name=user_name)
+        candidate_ids = self.graph_store.get_by_metadata(
+            key_filters, user_name=user_name, status="activated"
+        )
         node_dicts = self.graph_store.get_nodes(
-            list(candidate_ids), include_embedding=self.include_embedding
+            list(candidate_ids), include_embedding=self.include_embedding, user_name=user_name
         )
 
         bm25_query = " ".join(list({query, *parsed_goal.keys}))

src/memos/multi_mem_cube/single_cube.py

@@ -846,7 +846,9 @@ def _process_text_mem(
                         for old_id in old_ids:
                             try:
                                 self.mem_reader.graph_db.update_node(
-                                    str(old_id), {"status": "archived"}
+                                    str(old_id),
+                                    {"status": "archived"},
+                                    user_name=user_context.mem_cube_id,
                                 )
                                 self.logger.info(
                                     f"[SingleCubeView] Archived merged_from memory: {old_id}"