openclaw-memory-milvus

A robust, Milvus-backed long-term memory plugin for OpenClaw.

This plugin provides persistent vector memory for OpenClaw agents, including features like auto-recall (injecting relevant memories into the context before the agent speaks) and auto-capture (automatically extracting and storing important information from user messages).

It uses standard OpenAI-compatible /v1/embeddings endpoints for vector generation, supporting models from OpenAI, Voyage AI, Mistral, Volcengine (Doubao), and any custom endpoint.

Features

• 🧠 Auto-Recall: Automatically searches Milvus and injects relevant context before the agent generates a response.
• 💾 Auto-Capture: Analyzes user messages, automatically categorizing and storing important preferences, facts, and entities.
• 🛠️ Memory Tools: Provides memory_store, memory_recall, and memory_forget tools directly to the AI agent.
• 🔄 Safe Model Migration: Built-in CLI tool to safely re-embed your entire memory database to a new dimensionality if you change your embedding model.
• 🔒 GDPR Compliant: memory_forget provides the ability to selectively delete memories.

Installation

You can install this plugin in OpenClaw using the CLI:

openclaw plugins install @volcengine/memory-milvus

Alternatively, if you are developing the plugin locally, you can link it:

cd /path/to/openclaw-memory-milvus && npm install && npm run build
openclaw plugins install -l /path/to/openclaw-memory-milvus

Quickstart & Setup

After installing the plugin, run the interactive setup wizard to configure your Milvus connection and Embedding provider:

openclaw milvus setup

The wizard will guide you through:

1. Embedding Provider: Select your provider (OpenAI, Voyage AI, Volcengine, Mistral, or Custom) and API key.
2. Milvus Connection: Enter your Milvus host (e.g., localhost:19530) and authentication details.
3. Behavior Settings: Enable Auto-Recall and Auto-Capture.

Once the wizard completes, restart your OpenClaw gateway to apply the configuration:

openclaw gateway restart

CLI Commands

This plugin registers several CLI commands under openclaw milvus to help you manage your agent's memory database:

• openclaw milvus setup: Run the interactive configuration wizard.
• openclaw milvus config: Display your current configuration (with masked sensitive credentials).
• openclaw milvus search "<query>": Perform a manual semantic search against your memory collection.
• openclaw milvus stats: View total memory count and collection statistics.
• openclaw milvus migrate: Safely migrate existing memories if you change your embedding model (e.g., switching from text-embedding-3-small (1536 dims) to a Voyage model (1024 dims)).

Manual Configuration (Advanced)

If you prefer to edit your openclaw.json config manually, you can set the following parameters under plugins.entries.memory-milvus.config:

{
  "plugins": {
    "slots": {
      "memory": "memory-milvus"
    },
    "entries": {
      "memory-milvus": {
        "enabled": true,
        "config": {
          "embedding": {
            "apiKey": "sk-...",
            "model": "text-embedding-3-small",
            "baseUrl": "https://api.openai.com/v1"
          },
          "milvus": {
            "address": "localhost:19530",
            "token": "optional-token",
            "username": "optional-username",
            "password": "optional-password",
            "collectionName": "openclaw_memories"
          },
          "autoRecall": true,
          "autoCapture": true
        }
      }
    }
  }
}

Note on Authentication: If both token and username/password are provided, the token takes precedence. If neither is provided, the plugin will attempt to connect without authentication.

License

MIT