[eric7h]

Summary

Mem0LTMBackend logged self.index (from LongTermMemory.app_name) but never passed it to Mem0 add/search. Apps that share one Mem0 project therefore shared memories for the same user_id, unlike VikingDB (collection=index), Redis/OpenSearch (index name includes index), OpenViking (app_name), and TOS ContextBucket (bucket=index).

Also align save_memory(user_id, event_strings, ...) with BaseLongTermMemoryBackend — the previous signature swapped the first two parameters, so a positional call treated the user id as the event list and failed.

Fix

• Pass agent_id=self.index on Mem0 add and search (Mem0 1.x kwargs API that veadk pins via mem0ai>=1.0,<2).
• Restore base-class parameter order for save_memory / search_memory.

How to reproduce

# With a fake MemoryClient that stores by (user_id, agent_id):
alpha = Mem0LTMBackend(index="app_alpha", mem0_config=...)
beta = Mem0LTMBackend(index="app_beta", mem0_config=...)
alpha.save_memory(user_id="u1", event_strings=["alpha-only"])
beta.save_memory(user_id="u1", event_strings=["beta-only"])
# Before: both searches returned both memories (agent_id was None).
# After: each search returns only its own index's memories.

New tests in tests/memory/long_term_memory_backends/test_mem0_ltm_backend.py fail on main and pass with this change (3 passed). Related memory suite: 43 passed.

Test plan

• pytest tests/memory/long_term_memory_backends/test_mem0_ltm_backend.py — 3 passed
• pytest tests/test_long_term_memory.py tests/memory/ (excluding postgres endpoint) — 43 passed
• pre-commit (ruff check/format, gitleaks) on touched files