🐛 Describe the bug
#3966
Mem0 Update Payload Replacement Issue
Problem
Mem0's MongoDB vector store update() method replaces the entire payload field when updating a memory, causing complete loss of metadata if not handled correctly.
Root Cause
Mem0's vector store implementation does:
def update(self, vector_id: str, vector: Optional[List[float]] = None, payload: Optional[Dict] = None) -> None:
update_fields = {}
if vector is not None:
update_fields["embedding"] = vector
if payload is not None:
update_fields["payload"] = payload # ⚠️ REPLACES ENTIRE PAYLOAD
if update_fields:
self.collection.update_one({"_id": vector_id}, {"$set": update_fields})When Mem0's memory.update() is called, it internally calls the vector store's update() method with a new payload, which results in:
// MongoDB operation
{"$set": {"payload": new_payload}} // ⚠️ This REPLACES the entire payload objectResult: All existing metadata fields (bucket_id, context_id, category, source, etc.) are completely lost unless explicitly preserved.
Impact
- Metadata Loss: All custom metadata fields are wiped when updating memory content
- Data Integrity: Critical fields like
bucket_id, context_id, category disappear
- User Experience: UI shows "Ungrouped" memories, filters break, context is lost
Update memory content (without metadata preservation)
Without our fix, if you call Mem0 directly:
# ⚠️ THIS LOSES METADATA
from mem0 import Memory
mem0 = Memory.from_config({...})
mem0.update(memory_id=memory_id, data="User loves chocolate and vanilla")
# Check metadata - ALL GONE!
updated = memory_service.get(memory_id=memory_id, user_id="user123")
print(f"Metadata after update: {updated['metadata']}")
# Output: {} # ⚠️ EMPTY! All metadata lost!3. Verify metadata loss
# Fetch the memory
memory = memory_service.get(memory_id=memory_id, user_id="user123")
# Check metadata
assert "bucket_id" not in memory.get("metadata", {}), "Metadata should be lost!"
assert "source" not in memory.get("metadata", {}), "Metadata should be lost!"
# These assertions will PASS (metadata is gone)Potential Future Fixes
Option 1: Patch Mem0's Vector Store (Upstream Fix)
Ideal Solution: Modify Mem0's MongoDB vector store to merge payload instead of replacing:
# In mem0/vector_stores/mongodb.py
def update(self, vector_id: str, vector: Optional[List[float]] = None, payload: Optional[Dict] = None) -> None:
update_fields = {}
if vector is not None:
update_fields["embedding"] = vector
if payload is not None:
# ✅ MERGE instead of REPLACE
for k, v in payload.items():
update_fields[f"payload.{k}"] = v
# Instead of: update_fields["payload"] = payload
if update_fields:
self.collection.update_one({"_id": vector_id}, {"$set": update_fields})Pros:
- Fixes the issue at the source
- Benefits all Mem0 users
- Cleaner architecture
🐛 Describe the bug
#3966
Mem0 Update Payload Replacement Issue
Problem
Mem0's MongoDB vector store
update()method replaces the entirepayloadfield when updating a memory, causing complete loss of metadata if not handled correctly.Root Cause
Mem0's vector store implementation does:
When Mem0's
memory.update()is called, it internally calls the vector store'supdate()method with a new payload, which results in:Result: All existing metadata fields (
bucket_id,context_id,category,source, etc.) are completely lost unless explicitly preserved.Impact
bucket_id,context_id,categorydisappearUpdate memory content (without metadata preservation)
Without our fix, if you call Mem0 directly:
3. Verify metadata loss
Potential Future Fixes
Option 1: Patch Mem0's Vector Store (Upstream Fix)
Ideal Solution: Modify Mem0's MongoDB vector store to merge payload instead of replacing:
Pros: