#bm25 #multi-agent #memory-system #recall

OpenClaw Memory System

Persistent memory system for OpenClaw multi-agent workflows — keyword indexing, BM25 semantic search, DAG linking, and session hooks.

Jialin@0xcjl

Install

openclaw plugins install clawhub:openclaw-memory-plugin

OpenClaw Memory System Plugin :root { --bg: #0d1117; --text: #c9d1d9; --muted: #8b949e; --border: #30363d; --accent: #58a6ff; --tab-bg: #161b22; --tab-active: #21262d; --code-bg: #161b22; } @media (prefers-color-scheme: light) { :root { --bg: #ffffff; --text: #24292f; --muted: #57606a; --border: #d0d7de; --accent: #0969da; --tab-bg: #f6f8fa; --tab-active: #ffffff; --code-bg: #f6f8fa; } } * { box-sizing: border-box; } body { font-family: -apple-system, BlinkMacSystemFont, "Segoe UI", Helvetica, Arial, sans-serif; background: var(--bg); color: var(--text); line-height: 1.6; max-width: 900px; margin: 0 auto; padding: 2rem 1.5rem; } a { color: var(--accent); text-decoration: none; } a:hover { text-decoration: underline; } h1, h2, h3 { border-bottom: 1px solid var(--border); padding-bottom: 0.3em; } h1 { font-size: 2em; margin-top: 0; } h2 { font-size: 1.5em; margin-top: 1.8em; } h3 { font-size: 1.2em; } code, pre { background: var(--code-bg); border: 1px solid var(--border); border-radius: 6px; font-size: 0.9em; font-family: "SFMono-Regular", Consolas, "Liberation Mono", Menlo, monospace; } pre { padding: 1em; overflow-x: auto; } :not(pre) > code { padding: 0.2em 0.4em; } pre code { border: none; padding: 0; background: none; } .lang-tabs { display: flex; gap: 0; margin: 1.5em 0 1em; border-bottom: 2px solid var(--border); list-style: none; padding: 0; } .lang-tabs li { margin: 0; padding: 0; } .lang-tabs button { background: var(--tab-bg); border: 1px solid var(--border); border-bottom: none; border-radius: 6px 6px 0 0; color: var(--muted); cursor: pointer; font-size: 0.9em; font-weight: 600; padding: 0.5em 1.2em; margin-right: 2px; transition: color 0.15s, background 0.15s; } .lang-tabs button:hover { color: var(--text); background: var(--tab-active); } .lang-tabs button.active { background: var(--tab-active); color: var(--accent); border-bottom: 2px solid var(--tab-active); margin-bottom: -2px; } .lang-section { display: none; } .lang-section.active { display: block; } .badge-row { display: flex; gap: 0.5em; flex-wrap: wrap; margin: 1em 0; } .badge { background: var(--tab-active); border: 1px solid var(--border); border-radius: 2em; color: var(--muted); font-size: 0.75em; font-weight: 600; padding: 0.2em 0.8em; } .install-cmd { background: var(--code-bg); border: 1px solid var(--border); border-radius: 6px; padding: 1em 1.2em; font-family: monospace; overflow-x: auto; margin: 1em 0; } .feature-grid { display: grid; grid-template-columns: repeat(auto-fill, minmax(240px, 1fr)); gap: 1em; margin: 1em 0; } .feature-card { background: var(--tab-bg); border: 1px solid var(--border); border-radius: 8px; padding: 1em; } .feature-card h3 { border: none; margin: 0 0 0.3em; font-size: 1em; } .feature-card p { margin: 0; color: var(--muted); font-size: 0.9em; } hr { border: none; border-top: 1px solid var(--border); margin: 2em 0; } table { border-collapse: collapse; width: 100%; margin: 1em 0; font-size: 0.9em; } th, td { border: 1px solid var(--border); padding: 0.5em 0.8em; text-align: left; } th { background: var(--tab-bg); } tr:nth-child(even) td { background: var(--code-bg); }

🇺🇸 English
🇨🇳 中文
🇯🇵 日本語

OpenClaw Memory System Plugin

openclaw openclaw-plugin memory ai-agent bm25 multi-agent

Persistent memory system for OpenClaw multi-agent workflows — zero external dependencies, pure Python BM25 + keyword indexing, DAG association graph, and session lifecycle hooks.

✨ Features

🔍 BM25 Recall

Pure-Python Okapi BM25 semantic search. No pip packages — stdlib only.

🔑 Keyword Index

Fast token-based indexing with shared标记 and agent isolation.

🪝 Lifecycle Hooks

before-task recall + after-task save CLI for session memory management.

🔗 DAG Association

Typed directed links between memory entries. BFS path finding.

⚡ WAL Snapshot

Non-blocking index rebuild after writes. Sub-500ms recall target.

🌐 Multi-Agent

Agent-scoped memory isolation with shared cross-agent memories.

📦 Installation

bash

# From GitHub (once published to ClawHub/npm)
openclaw plugins install @0xcjl/openclaw-memory-plugin

# From local directory
git clone https://github.com/0xcjl/openclaw-memory-plugin.git
openclaw plugins install ./openclaw-memory-plugin

Then restart the gateway:

bash

openclaw gateway restart

🚀 Quick Start

CLI Usage

bash

# Install dependencies (none required — pure Python/stdlib)
# No pip install needed!

# Build indexes
./scripts/memory-hook.sh build

# Recall memories before a task (< 500ms target)
./scripts/memory-hook.sh before-task "optimize Vue rendering performance"

# Save after task completion
./scripts/memory-hook.sh after-task "completed Vue optimization" /tmp/result.md

# Search memories
python3 scripts/bm25_search.py --search "Docker CI/CD pipeline" --top 5 --json

# Build DAG auto-links (shared tags → edges)
python3 scripts/dag-builder.py build

# Find shortest path between memory entries
python3 scripts/dag-builder.py paths abc123 def456

Plugin Tools (via OpenClaw)

Once installed, these tools are automatically registered:

Tool	Description
`memory_recall`	Recall top-N memories before a task
`memory_save`	Save task output to memory
`memory_search`	BM25 full-text search
`memory_build`	Rebuild all indexes
`memory_dag_link`	Link two memory entries

Example conversation with the agent:

"recall memories about Vue performance optimization" "save this task: completed the refactor" "search memories for Docker deployment issues"

🗂️ Project Structure

text

openclaw-memory-plugin/
├── openclaw.plugin.json   # Plugin manifest
├── index.ts               # TypeScript entry (registers tools)
├── package.json           # npm package
├── scripts/
│   ├── memory-indexer.py  # Keyword index builder + search
│   ├── bm25_search.py     # Pure-Python BM25 implementation
│   ├── memory-hook.sh     # Session lifecycle CLI
│   ├── dag-builder.py     # Memory DAG builder + path queries
│   └── wal-snapshot.sh    # WAL snapshot / background rebuild
├── skills/
│   └── memory-recall/
│       └── SKILL.md       # Skill interface documentation
└── docs/
    └── ARCHITECTURE.md    # System architecture details

⚙️ Configuration

In ~/.openclaw/config.yaml:

yaml

plugins:
  entries:
    openclaw-memory-system:
      enabled: true
      config:
        agentId: dev          # default agent ID
        workspaceRoot: ~/.openclaw/workspace-dev
        topN: 3               # memories per recall

📊 Scoring Formula

text

combined_score = bm25_norm × 0.6 + weight × 0.2 + keyword_overlap × 0.2

Component	Range	Weight
`bm25_norm`	0–1	60%
`weight`	0–1	20%
`keyword_overlap`	0–1	20%

🌳 Memory TTL Classes

Class	Meaning	Auto-expiry
P0	Core identity, key configs	Never
P1	Project decisions, progress	~90 days
P2	Debug, temporary	~30 days

🐳 Requirements

Python 3 (standard library only — no pip packages)
Bash
OpenClaw gateway

📄 License

MIT

🔗 Related

OpenClaw 记忆系统插件

openclaw openclaw-plugin memory ai-agent bm25 multi-agent

OpenClaw 多智能体工作流的持久化记忆系统 — 零外部依赖，纯 Python BM25 + 关键词索引，DAG 关联图谱，以及会话生命周期钩子。

✨ 功能特性

🔍 BM25 语义搜索

纯 Python Okapi BM25 算法实现，无需任何 pip 包。

🔑 关键词索引

快速基于词的索引，支持 shared 标记和 Agent 隔离。

🪝 生命周期钩子

before-task 召回 + after-task 保存 CLI，管理会话记忆。

🔗 DAG 关联图谱

记忆节点间带类型的双向链接，支持 BFS 路径查找。

⚡ WAL 快照

写入后非阻塞索引重建，召回延迟目标 < 500ms。

🌐 多 Agent 支持

Agent 作用域隔离记忆 + 跨 Agent 共享记忆。

📦 安装

bash

# 从 ClawHub / npm 安装（发布后）
openclaw plugins install @0xcjl/openclaw-memory-plugin

# 从本地目录安装
git clone https://github.com/0xcjl/openclaw-memory-plugin.git
openclaw plugins install ./openclaw-memory-plugin

然后重启 Gateway：

bash

openclaw gateway restart

🚀 快速上手

CLI 使用

bash

# 构建索引
./scripts/memory-hook.sh build

# 任务前召回（目标 < 500ms）
./scripts/memory-hook.sh before-task "优化 Vue 渲染性能"

# 任务后保存
./scripts/memory-hook.sh after-task "完成 Vue 优化" /tmp/result.md

# 记忆搜索
python3 scripts/bm25_search.py --search "Docker CI/CD" --top 5 --json

# 构建 DAG 自动链接（共享标签 → 边）
python3 scripts/dag-builder.py build

# 查找两条记忆间的最短路径
python3 scripts/dag-builder.py paths abc123 def456

插件工具（通过 OpenClaw）

安装后自动注册以下工具：

工具	说明
`memory_recall`	任务前召回 Top-N 记忆
`memory_save`	将任务输出保存到记忆
`memory_search`	BM25 全文搜索
`memory_build`	重建所有索引
`memory_dag_link`	链接两条记忆

Agent 对话示例

"召回关于 Vue 性能优化的记忆" "保存这个任务：完成了重构" "搜索 Docker 部署相关的记忆"

🗂️ 项目结构

text

openclaw-memory-plugin/
├── openclaw.plugin.json   # 插件清单
├── index.ts               # TypeScript 入口（注册工具）
├── package.json           # npm 包配置
├── scripts/
│   ├── memory-indexer.py  # 关键词索引构建 + 搜索
│   ├── bm25_search.py     # 纯 Python BM25 实现
│   ├── memory-hook.sh     # 会话生命周期 CLI
│   ├── dag-builder.py     # 记忆 DAG 构建 + 路径查询
│   └── wal-snapshot.sh    # WAL 快照 / 后台重建
├── skills/
│   └── memory-recall/
│       └── SKILL.md       # 技能接口文档
└── docs/
    └── ARCHITECTURE.md    # 系统架构说明

⚙️ 配置

在 ~/.openclaw/config.yaml 中：

yaml

plugins:
  entries:
    openclaw-memory-system:
      enabled: true
      config:
        agentId: dev          # 默认 Agent ID
        workspaceRoot: ~/.openclaw/workspace-dev
        topN: 3               # 每次召回的记忆数量

📊 评分公式

text

combined_score = bm25_norm × 0.6 + weight × 0.2 + keyword_overlap × 0.2

组成部分	范围	权重
`bm25_norm`	0–1	60%
`weight`	0–1	20%
`keyword_overlap`	0–1	20%

🌳 记忆 TTL 分类

类别	含义	自动过期
P0	核心身份、关键配置	永不
P1	项目决策、进展	约 90 天
P2	调试、临时	约 30 天

🐳 环境要求

Python 3（标准库 — 无需任何 pip 包）
Bash
OpenClaw Gateway

📄 许可证

MIT

🔗 相关链接

OpenClaw メモリシステムプラグイン

openclaw openclaw-plugin memory ai-agent bm25 multi-agent

OpenClaw マルチエージェントワークフロー向け永続メモリシステム — 外部依存ゼロ、ピュアPython BM25 + キーワードインデックス、DAG連想グラフ、セッションライフサイクルフック。

✨ 機能

🔍 BM25 リ:call

ピュアPython Okapi BM25実装。pipパッケージ不要。

🔑 キーワードインデックス

sharedフラグとエージェント分離に対応する高速トークンインデックス。

🪝 ライフサイクルフック

before-task recall + after-task save CLI。

🔗 DAG連想グラフ

メモリエントリ間の型付き有向リンク。BFS経路検索。

⚡ WALスナップショット

書き込み後のノンブロッキングインデックス再構築。

🌐 マルチエージェント

エージェントスコープのメモリ分離と共有クロスエージェントメモリ。

📦 インストール

bash

# ClawHub / npm から（公開後）
openclaw plugins install @0xcjl/openclaw-memory-plugin

# ローカルディレクトリから
git clone https://github.com/0xcjl/openclaw-memory-plugin.git
openclaw plugins install ./openclaw-memory-plugin

Gatewayを再起動：

bash

openclaw gateway restart

🚀 クイックスタート

bash

# インデックスのビルド
./scripts/memory-hook.sh build

# タスク前のrecall（目標 < 500ms）
./scripts/memory-hook.sh before-task "Vueパフォーマンス最適化"

# タスク後の保存
./scripts/memory-hook.sh after-task "Vue最適化完了" /tmp/result.md

# メモリ検索
python3 scripts/bm25_search.py --search "Docker CI/CD" --top 5 --json

# DAG自動リンクビルド
python3 scripts/dag-builder.py build

⚙️ 設定

~/.openclaw/config.yaml:

yaml

plugins:
  entries:
    openclaw-memory-system:
      enabled: true
      config:
        agentId: dev
        workspaceRoot: ~/.openclaw/workspace-dev
        topN: 3

🐳 必要環境

Python 3（標準ライブラリのみ — pipパッケージ不要）
Bash
OpenClaw Gateway

📄 ライセンス

MIT