fix(memory): centralize entity cleanup and skip malformed LLM relation dicts (#4515)

mem0/graphs/neptune/base.py

@@ -1,7 +1,7 @@
 import logging
 from abc import ABC, abstractmethod
 
-from mem0.memory.utils import format_entities
+from mem0.memory.utils import format_entities, remove_spaces_from_entities
 
 try:
     from rank_bm25 import BM25Okapi
@@ -151,11 +151,7 @@ class NeptuneBase(ABC):
         return entities
 
     def _remove_spaces_from_entities(self, entity_list):
-        for item in entity_list:
-            item["source"] = item["source"].lower().replace(" ", "_")
-            item["relationship"] = item["relationship"].lower().replace(" ", "_")
-            item["destination"] = item["destination"].lower().replace(" ", "_")
-        return entity_list
+        return remove_spaces_from_entities(entity_list, sanitize_relationship=False)
 
     def _get_delete_entities_from_search_output(self, search_output, data, filters):
         """

mem0/memory/graph_memory.py

@@ -1,6 +1,6 @@
 import logging
 
-from mem0.memory.utils import format_entities, sanitize_relationship_for_cypher
+from mem0.memory.utils import format_entities, remove_spaces_from_entities
 
 try:
     from langchain_neo4j import Neo4jGraph
@@ -657,12 +657,7 @@ class MemoryGraph:
         return results
 
     def _remove_spaces_from_entities(self, entity_list):
-        for item in entity_list:
-            item["source"] = item["source"].lower().replace(" ", "_")
-            # Use the sanitization function for relationships to handle special characters
-            item["relationship"] = sanitize_relationship_for_cypher(item["relationship"].lower().replace(" ", "_"))
-            item["destination"] = item["destination"].lower().replace(" ", "_")
-        return entity_list
+        return remove_spaces_from_entities(entity_list, sanitize_relationship=True)
 
     def _search_source_node(self, source_embedding, filters, threshold=0.9):
         # Build WHERE conditions

mem0/memory/kuzu_memory.py

@@ -1,6 +1,6 @@
 import logging
 
-from mem0.memory.utils import format_entities
+from mem0.memory.utils import format_entities, remove_spaces_from_entities
 
 try:
     import kuzu
@@ -654,11 +654,7 @@ class MemoryGraph:
         return results
 
     def _remove_spaces_from_entities(self, entity_list):
-        for item in entity_list:
-            item["source"] = item["source"].lower().replace(" ", "_")
-            item["relationship"] = item["relationship"].lower().replace(" ", "_")
-            item["destination"] = item["destination"].lower().replace(" ", "_")
-        return entity_list
+        return remove_spaces_from_entities(entity_list, sanitize_relationship=False)
 
     def _search_source_node(self, source_embedding, filters, threshold=0.9):
         params = {

mem0/memory/memgraph_memory.py

@@ -1,6 +1,6 @@
 import logging
 
-from mem0.memory.utils import format_entities, sanitize_relationship_for_cypher
+from mem0.memory.utils import format_entities, remove_spaces_from_entities
 
 try:
     from langchain_memgraph.graphs.memgraph import Memgraph
@@ -550,12 +550,7 @@ class MemoryGraph:
         return results
 
     def _remove_spaces_from_entities(self, entity_list):
-        for item in entity_list:
-            item["source"] = item["source"].lower().replace(" ", "_")
-            # Use the sanitization function for relationships to handle special characters
-            item["relationship"] = sanitize_relationship_for_cypher(item["relationship"].lower().replace(" ", "_"))
-            item["destination"] = item["destination"].lower().replace(" ", "_")
-        return entity_list
+        return remove_spaces_from_entities(entity_list, sanitize_relationship=True)
 
     def _search_source_node(self, source_embedding, filters, threshold=0.9):
         """Search for source nodes with similar embeddings."""

mem0/memory/utils.py

@@ -1,6 +1,7 @@
 import hashlib
 import logging
 import re
+from typing import Any, Dict, List
 
 from mem0.configs.prompts import (
     AGENT_MEMORY_EXTRACTION_PROMPT,
@@ -265,3 +266,30 @@ def sanitize_relationship_for_cypher(relationship) -> str:
 
     return re.sub(r"_+", "_", sanitized).strip("_")
 
+
+def remove_spaces_from_entities(
+    entity_list: List[Any],
+    *,
+    sanitize_relationship: bool = True,
+) -> List[Dict[str, Any]]:
+    """
+    Normalize entity relation dicts from LLM/tool output: lowercase, spaces to underscores.
+
+    Skips entries that are not non-empty dicts or that lack any of
+    ``source``, ``relationship``, or ``destination`` (avoids KeyError on ``[{}]``
+    or partial dicts).
+    """
+    required = ("source", "relationship", "destination")
+    cleaned: List[Dict[str, Any]] = []
+    for item in entity_list:
+        if not isinstance(item, dict) or not item:
+            continue
+        if not all(key in item for key in required):
+            continue
+        item["source"] = item["source"].lower().replace(" ", "_")
+        rel = item["relationship"].lower().replace(" ", "_")
+        item["relationship"] = sanitize_relationship_for_cypher(rel) if sanitize_relationship else rel
+        item["destination"] = item["destination"].lower().replace(" ", "_")
+        cleaned.append(item)
+    return cleaned
+