MemoryX OpenClaw Plugin

Official MemoryX plugin for OpenClaw. Enables long-term memory for agents by recalling context before execution and saving conversations after each run.

Features

• Auto Recall: before_agent_start → semantic search for relevant memories
• Auto Capture: message_received + assistant_response → buffer and flush to MemoryX API
• Function Calling: LLM can actively search, list, and delete memories
• Auto Registration: Agents auto-register with machine fingerprint
• Conversation Buffer: Smart buffering with token counting and round-based flushing

Function Calling Tools

The plugin registers three tools that LLM can call during conversations:

| Tool | Description | When to Use | |------|-------------|-------------| | memoryx_recall | Search through long-term memories | User asks "do you remember X?" or needs context | | memoryx_store | Save important information to memory | User says "remember this" or you identify important info | | memoryx_list | List all stored memories | User asks "what do you remember about me?" | | memoryx_forget | Delete a specific memory | User asks to forget/remove something |

memoryx_store Parameters

| Parameter | Type | Required | Description | |-----------|------|----------|-------------| | content | string | Yes | The information to remember (server will auto-categorize) |

Example Usage

User: "What do you remember about my preferences?" LLM: calls memoryx_list → Returns list of stored memories

User: "Did I ever mention my favorite color?" LLM: calls memoryx_recall with query="favorite color" → Searches and returns relevant memories

User: "Please forget about my old address" LLM: calls memoryx_forget with memory_id → Deletes the memory

User: "Remember that my favorite color is blue" LLM: calls memoryx_store with content="User's favorite color is blue" → Stores the memory

Install

openclaw plugins install @t0ken.ai/memoryx-openclaw-plugin
openclaw gateway restart

Update

openclaw plugins update @t0ken.ai/memoryx-openclaw-plugin
openclaw gateway restart

Or update all plugins:

openclaw plugins update --all
openclaw gateway restart

Configuration

Edit ~/.openclaw/openclaw.json:

{
  "plugins": {
    "slots": {
      "memory": "memoryx-openclaw-plugin"
    },
    "entries": {
      "memoryx-openclaw-plugin": {
        "enabled": true,
        "config": {
          "apiBaseUrl": "https://t0ken.ai/api"
        }
      },
      "memory-core": {
        "enabled": false
      },
      "memory-lancedb": {
        "enabled": false
      }
    }
  }
}

For self-hosted MemoryX:

{
  "plugins": {
    "slots": {
      "memory": "memoryx-openclaw-plugin"
    },
    "entries": {
      "memoryx-openclaw-plugin": {
        "enabled": true,
        "config": {
          "apiBaseUrl": "http://192.168.31.65:8000/api"
        }
      },
      "memory-core": {
        "enabled": false
      },
      "memory-lancedb": {
        "enabled": false
      }
    }
  }
}

Restart the gateway after config changes:

openclaw gateway restart

How it Works

Powered by @t0ken.ai/memoryx-sdk with conversation preset.

Recall (before_agent_start)

• Builds a /v1/memories/search request using the current prompt
• Injects relevant memories via prependContext:

  [Relevant Memories]
  - [preference] User prefers dark mode
  - [fact] User's timezone is UTC+8
  [End of memories]

Add (message_received + assistant_response)

• Buffers messages with precise token counting
• Flushes to /v1/conversations/flush when:
  • 30k tokens reached
  • 5 minutes idle timeout
• Server extracts entities, facts, and preferences automatically

Auto Registration

On first run, the plugin:

1. Generates a machine fingerprint
2. Calls /agents/auto-register to get API key
3. Stores credentials locally for future sessions

Memory Categories

Memories are categorized by the server:

• preference: User preferences and settings
• fact: Factual information about the user
• plan: Future plans and goals
• experience: Past experiences
• opinion: User opinions and views

Notes

• Conversation buffer uses cl100k_base encoding (GPT-4 compatible)
• Maximum 8000 tokens per message
• Minimum 2 characters per message
• Short messages like "ok", "thanks" are skipped

How the proxy works

Current behavior: The plugin starts a local Sidecar (port 3335) and tries to redirect your existing provider’s baseUrl to it (so you keep using e.g. zai/glm-5). Whether that works depends on how OpenClaw resolves the model (config vs. registry). See OPENCLAW-BASEURL-FLOW.md for details.

SlimClaw-style (recommended, no file changes): Like SlimClaw, you can use a virtual provider so all traffic goes through the Sidecar without the plugin mutating config or files. Add this to ~/.openclaw/openclaw.json under models.providers:

"memoryx-gateway": {
  "baseUrl": "http://localhost:3335/v1",
  "apiKey": "placeholder",
  "models": [{ "id": "auto" }]
}

Then add an auth profile for memoryx-gateway (e.g. in ~/.openclaw/agents/main/agent/auth-profiles.json or global auth) with any placeholder key. In the UI, choose model MemoryX Gateway (id: memoryx-gateway/auto). Requests will go to the Sidecar, which forwards to the MemoryX server with your real provider credentials (you configure which provider/model to use via the plugin/Sidecar logic). The plugin does not modify your config or models.json; it only runs the Sidecar.

License

MIT