agent_id filter not working in v2 API (get_all and search)

[feiwofeifeixiaowo]

Description

The agent_id filter doesn't work in v2 API for get_all and search operations, even though the documentation states it should.

Documentation reference: https://docs.mem0.ai/platform/features/v2-memory-filters#missing-results-with-agent-id

Steps to Reproduce

from mem0 import AsyncMemoryClient

client = AsyncMemoryClient(api_key="...")

# 1. Add memory with agent_id
result = await client.add(
    messages=[{"role": "user", "content": "Test memory"}],
    user_id="test_user_123",
    agent_id="test_agent_456",
    run_id="test_run_789",
    infer=False,
)
# ✅ Memory created successfully

# 2. Try to filter by agent_id
filters = {"AND": [{"user_id": "test_user_123"}, {"agent_id": "test_agent_456"}]}
results = await client.get_all(filters=filters)
# ❌ Returns 0 results

# 3. Try wildcard agent_id (per documentation)
filters = {"AND": [{"user_id": "test_user_123"}, {"agent_id": "*"}]}
results = await client.get_all(filters=filters)
# ❌ Returns 0 results

# 4. Filter by run_id works correctly
filters = {"AND": [{"user_id": "test_user_123"}, {"run_id": "test_run_789"}]}
results = await client.get_all(filters=filters)
# ✅ Returns 1 result

Expected Behavior

According to the documentation, filtering by agent_id should return memories that were added with that agent_id. The wildcard {"agent_id": "*"} should return all memories with non-null agent_id values.

Actual Behavior

• agent_id filter returns 0 results (both exact match and wildcard)
• run_id filter works correctly
• When retrieving the memory by ID, it shows agent_id is stored in metadata but not as a filterable field

Memory Structure Returned

{
    'id': '...',
    'memory': 'Test memory',
    'user_id': 'test_user_123',
    'session_id': 'test_run_789',  # run_id is mapped to session_id
    'metadata': {
        'agent_id': 'test_agent_456',  # agent_id only in metadata
        'run_id': 'test_run_789'
    },
    ...
}

Note: run_id is properly indexed as session_id, but agent_id appears to only be stored in metadata without indexing.

Environment

• mem0 SDK version: 0.1.98 (latest)
• Python version: 3.12
• API: Platform API (api.mem0.ai)

Questions

1. Is agent_id filtering currently supported in v2 API?
2. If not, is this a planned feature?
3. Should we use run_id as the primary scope identifier instead?

Thank you!