@neuromem/openclaw-neuromem

Deep behavioral memory for OpenClaw agents — facts, episodes, knowledge graph, and AI-generated trait insights.

Quick Start

1. Install the plugin:

openclaw plugins install @neuromem/openclaw-neuromem

2. Get an API key at neuromem.cloud

3. Configure in ~/.openclaw/config:

"openclaw-neuromem": {
  "enabled": true,
  "config": {
    "apiKey": "${NEUROMEM_API_KEY}"
  }
}

That's it. Your agent now has persistent memory across sessions.

What It Does

• Auto-recall — Before each response, recalls relevant memories (facts, episodes, behavioral traits) and injects them into context.
• Auto-capture — After each response, stores new information using ONE LLM mode (zero extra LLM cost).
• Auto-digest — At session end, discovers behavioral patterns and stores them as traits.

All three behaviors work out of the box with no code changes to your agent.

Configuration Reference

| Parameter | Type | Default | Description | |-----------|------|---------|-------------| | apiKey | string | required | neuromem.cloud API key (nm_sk_...) | | baseUrl | string | https://api.neuromem.cloud | API base URL | | autoRecall | boolean | true | Inject memories before responses | | autoCapture | boolean | true | Store memories after responses | | autoDigest | "off" | "session-end" | "session-end" | When to discover behavioral patterns | | topK | integer | 10 | Max memories per recall (1-50) | | includeTraits | boolean | true | Include behavioral traits in recall |

Agent Tools

The plugin registers 5 tools that your agent (or the user) can invoke directly:

| Tool | Description | |------|-------------| | neuromem_recall | Search memories with a natural language query, ranked by similarity | | neuromem_store | Store information to long-term memory (facts, episodes, relationships) | | neuromem_digest | Analyze stored memories to generate behavioral traits and patterns | | neuromem_list | Browse stored memories, optionally filtered by type | | neuromem_forget | Delete a specific memory by ID |

Why neuromem vs mem0?

| Feature | mem0 | neuromem | |---------|------|----------| | Auto-recall | Yes | Yes | | Auto-capture | Yes | Yes | | Auto-digest (traits) | No | Yes | | ONE LLM mode | No | Yes (zero extra cost) | | Knowledge graph | Beta | Built-in | | Emotion metadata | No | Yes | | Sleep-time reflection | No | Yes |

MCP Direct Connect

For advanced users who want fine-grained control, you can connect to neuromem.cloud directly via MCP without installing this plugin.

Add to ~/.openclaw/mcp.json:

{
  "mcpServers": {
    "neuromem": {
      "url": "https://api.neuromem.cloud/mcp/",
      "headers": {
        "Authorization": "Bearer nm_sk_your_key_here"
      }
    }
  }
}

This exposes 12 MCP tools:

| Tool | Description | |------|-------------| | ingest | Store raw conversation content | | recall | Search memories by natural language query | | digest | Generate behavioral traits from stored memories | | ingest_extracted | Store pre-extracted memory objects directly | | digest_extracted | Store pre-generated trait objects directly | | list_memories | List stored memories with optional filters | | update_memory | Update an existing memory | | delete_memory | Delete a single memory by ID | | delete_memories | Bulk delete memories by filter | | import_conversations | Import conversation history from external sources | | import_status | Check the status of an ongoing import | | cancel_import | Cancel a running import job |

Plugin vs MCP: When to Use Which

• Plugin — Automatic memory with zero effort. Install, configure, forget about it. Best for most users.
• MCP direct connect — Full control over when and how memories are stored and retrieved. Best for custom workflows, batch imports, or when you need access to all 12 tools.

License

MIT