#long-term-memory #enhancement #bge-m3 #add-memories

Memory Enhancement

BGE-M3记忆增强插件，为OpenClaw添加长期记忆功能

Install

openclaw plugins install clawhub:@cv-forever/memory-enhancement

Memory Enhancement Plugin

BGE-M3 based long-term memory enhancement plugin for OpenClaw.
Automatically retrieves relevant historical memories and injects them into conversation context.

Privacy

Data Flow: This plugin sends user conversation content to a configured BGE-M3 API server for embedding and retrieves relevant historical memories, which are then injected into the agent's context.

Default Configuration (Local-Only): By default, the plugin connects to 127.0.0.1:8080 — a local API server running on your own machine. Your data stays on your device and never leaves your network.

What you should know:

Conversation content is sent to the BGE-M3 API server for embedding computation
Generated embeddings and memory data are persisted on the server
If you override apiHost to a remote server, your conversation data will be sent to that server
Logging: by default (logDetailedContent: false), the plugin only logs metadata (query length, status codes, memory count). Set logDetailedContent: true to enable detailed query content logging

Controls:

Set enabled: false to completely disable memory retrieval
Set logDetailedContent: false (default) to prevent conversation content from appearing in logs
Run the BGE-M3 API server locally (default) to keep all data on your machine

Features

Automatic memory retrieval — On each session start, queries the BGE-M3 API server for memories relevant to the current context, injects them into before_prompt_build
Long-term memory storage — add_memories tool to save important conversations, decisions, and discoveries
Memory search — retrieve_memories tool to query stored memories on demand
Configurable — threshold, topK, cache, debug logging, all tunable via OpenClaw plugin config

How it works

text

Session Start
  │
  ▼
before_prompt_build hook
  │
  ├── Extract recent conversation text
  ├── Query BGE-M3 API (POST /retrieve)
  ├── Inject top-K relevant memories into prompt
  │
  ▼
Agent response with memory context

Installation

bash

openclaw plugins install @cv/memory-enhancement
openclaw gateway restart

Configuration

Field	Default	Description
enabled	true	Enable/disable the plugin
apiHost	127.0.0.1	BGE-M3 API server host
apiPort	8080	BGE-M3 API server port
threshold	0.5	Similarity threshold (0.0–1.0)
topK	3	Max memories to inject
timeout	15000	API call timeout (ms)
maxRetries	2	Max retries on failure
cacheEnabled	true	Enable query cache
cacheTTL	300	Cache TTL (seconds)
debug	false	Verbose debug logging
logDetailedContent	false	Log detailed query content (default: false, metadata only)

Prerequisites

BGE-M3 API server running on the configured host:port
OpenClaw >= 2026.4.0

Server Setup

Memory Enhancement 插件需要一个 BGE-M3 API 服务端来提供记忆检索能力。

服务端代码和部署说明在仓库的 server/ 目录中：

bash

cd server
pip install -r requirements.txt
python3 memory_api_server_mac.py --port 8080

详见 server/README.md。

如果需要训练自己的 LoRA 适配器，请参考原始项目仓库（包含完整训练脚本、数据预处理流程和评估工具）。

Version

1.1.0 — Privacy improvements: added Privacy section to README, logDetailedContent config, privacy warning on plugin load

Author

小爪 🐾