[Bug] Mem0 2.0 API breaking change causes Mem0LongTermMemory to fail in AgentScope

[Baochr]

[Bug] Mem0AI 2.0 API breaking change causes Mem0LongTermMemory to fail in AgentScope

🐛 Description

After upgrading mem0ai from 1.x to 2.0, the Mem0LongTermMemory class in AgentScope no longer works properly. The root cause is that mem0ai 2.0 introduced a breaking change to the search() method: it no longer supports passing user_id, agent_id, run_id as top-level parameters. Instead, these must be provided within a filters dictionary.

🔍 Steps to Reproduce

1. Install mem0ai 2.0 or later:

   pip install mem0ai>=2.0.0

2. Use Mem0LongTermMemory in AgentScope:

   from agentscope.memory import Mem0LongTermMemory
   
   memory = Mem0LongTermMemory(
       user_name="test_user",
       agent_name="test_agent",
       # ... other configs
   )
   
   # Execute a retrieval operation
   await memory.retrieve(query_msg)

3. Observe the error:

   "text": "Error retrieving memory: Top-level entity parameters frozenset({'user_id', 'run_id', 'agent_id'}) are not supported in search(). Use filters={'user_id': '...'} instead."
   

📌 Expected Behavior

Mem0LongTermMemory should work seamlessly with mem0ai 2.0's updated API and successfully execute retrieval operations.

🛠️ Root Cause Analysis

According to mem0ai 2.0 documentation, the API has changed as follows:

Version	Call Style
v1.x	client.search(query, user_id="alice")
v2.0	client.search(query, filters={"user_id": "alice"})

For scenarios with multiple entity parameters, v2.0 requires using AND/OR logical operators:

filters = {
    "AND": [
        {"user_id": "traveler_cam"},
        {"agent_id": "travel_planner"},
        {"run_id": "tokyo-2025-weekend"}
    ]
}

AgentScope's Mem0LongTermMemory implementation currently uses the v1.x parameter passing style and needs to be adapted.

📁 Affected Files

Based on AgentScope's code structure, the issue likely exists in the retrieve() method of the Mem0LongTermMemory class, where it internally calls mem0ai's search() operation.

💡 Proposed Fix

Modify the search() call in Mem0LongTermMemory to convert top-level parameters into a filters dictionary:

Before (v1.x style):

# Current implementation likely resembles
result = self.mem0_client.search(query, user_id=user_id, agent_id=agent_id, run_id=run_id)

After (v2.0 style):

# Build filters dictionary
filters_conditions = []
if user_id:
    filters_conditions.append({"user_id": user_id})
if agent_id:
    filters_conditions.append({"agent_id": agent_id})
if run_id:
    filters_conditions.append({"run_id": run_id})

if len(filters_conditions) == 1:
    filters = filters_conditions[0]
elif len(filters_conditions) > 1:
    filters = {"AND": filters_conditions}
else:
    filters = None

result = self.mem0_client.search(query, filters=filters)

🔄 Compatibility Consideration

Since some users may still be using mem0ai 1.x, consider adding version detection for backward compatibility:

from packaging import version
import mem0ai

if version.parse(mem0ai.__version__) >= version.parse("2.0.0"):
    # Use new filters approach
    ...
else:
    # Use old top-level parameters approach
    ...

📋 Environment

• AgentScope version: latest
• mem0ai version: 2.0.0+
• Python version: 3.x

🔗 References

• Mem0AI Entity Partitioning Documentation: Partition Memories by Entity
• AgentScope Mem0LongTermMemory Integration: Long-Term Memory Integration