Context

The @mem0/openclaw-mem0 plugin works great for single-agent setups. However, OpenClaw supports multi-agent configurations, and the plugin currently takes over the global memory slot — affecting all agents, not just the intended one.

Problem

When installing the Mem0 plugin via openclaw plugins install @mem0/openclaw-mem0, it claims the exclusive memory slot globally, disabling memory-core and memory-lancedb for every agent. There's no way to scope the plugin to a specific agent.

Example: I run three agents — a general assistant, a job search agent, and a research agent. Only the research agent should use Mem0 Cloud for memory. The other two should keep the default memory plugin.

Current workaround

Using the Mem0 Python SDK directly (MemoryClient) from the agent workspace, bypassing the plugin system entirely. This works but loses the auto-recall/auto-capture lifecycle that makes the plugin valuable.

Suggestion

Consider whether the OpenClaw plugin could:

1. Support an agent filter in config (e.g., agents: ["mem0claw"]) so it only activates for specified agents
2. Or gracefully coexist with the default memory plugin by operating as an additive memory layer rather than replacing the slot entirely

Related

• OpenClaw upstream issue for per-agent plugin slots: Feature: Per-agent plugin slot overrides for multi-agent setups openclaw/openclaw#25359

Environment

• OpenClaw with multi-agent config (3 agents)
• @mem0/openclaw-mem0 v0.1.2
• Mem0 Cloud API