← 返回 Skills 市场

多Agent记忆系统

Name: 多Agent记忆系统
Author: liucunguang

作者佐岸流年L · GitHub ↗ · v1.0.0 · MIT-0

cross-platform ⚠ suspicious

总下载

当前安装

版本数

在 OpenClaw 中安装

/install light-office-multi-agent-memory

功能描述

通用多Agent记忆系统 - 自动捕获、RRF检索、知识图谱、矛盾检测、Token追踪

使用说明 (SKILL.md)

多Agent记忆系统 🧠

功能: 通用多Agent记忆系统，支持自动捕获、RRF检索、知识图谱、矛盾检测、Token追踪

适用场景:

单Agent记忆管理
多Agent协作记忆
企业级记忆系统

核心特性:

✅ 8个Hook自动捕获
✅ RRF融合检索（BM25 + Vector + Graph）
✅ 知识图谱构建
✅ 矛盾检测集成（95.7%解决率）
✅ Token消耗追踪
✅ 检索基准测试（R@5=100%）
✅ Git快照管理
✅ 工作流引擎
✅ 实时可视化面板
✅ RRF权重优化
✅ 动态Agent管理（自动增加/减少）

快速开始

1. 安装

# 使用ClawHub安装
clawhub install multi-agent-memory

# 或手动安装
npx clawhub@latest install multi-agent-memory

2. 配置

# 运行配置向导
python3 ~/.openclaw/workspace/skills/multi-agent-memory/scripts/setup.py

3. 使用

# 运行集成测试
python3 ~/.openclaw/workspace/skills/multi-agent-memory/scripts/memory-system.py

# 查看可视化面板
open ~/.openclaw/workspace/skills/multi-agent-memory/public/dashboard.html

核心功能

1. Hook自动捕获

python3 ~/.openclaw/workspace/skills/multi-agent-memory/scripts/hook-capture.py

2. RRF融合检索

python3 ~/.openclaw/workspace/skills/multi-agent-memory/scripts/rrf-search.py

3. 知识图谱构建

python3 ~/.openclaw/workspace/skills/multi-agent-memory/scripts/graph-builder.py

4. 矛盾检测集成

python3 ~/.openclaw/workspace/skills/multi-agent-memory/scripts/conflict-detector.py

5. Token消耗追踪

python3 ~/.openclaw/workspace/skills/multi-agent-memory/scripts/token-tracker.py

6. 动态Agent管理

# 注册Agent
python3 ~/.openclaw/workspace/skills/multi-agent-memory/scripts/agent-manager.py register agent-001

# 注销Agent
python3 ~/.openclaw/workspace/skills/multi-agent-memory/scripts/agent-manager.py unregister agent-001

# 列出Agent
python3 ~/.openclaw/workspace/skills/multi-agent-memory/scripts/agent-manager.py list

# 健康检查
python3 ~/.openclaw/workspace/skills/multi-agent-memory/scripts/agent-manager.py health

# 自动扩展
python3 ~/.openclaw/workspace/skills/multi-agent-memory/scripts/agent-manager.py scale

---

## 配置说明

### 默认配置

```yaml
# config/default-config.yaml
memory:
  vector_model: "nomic-embed-text-v1.5"
  vector_dim: 768
  search_mode: "rrf"  # rrf / vector / bm25
  
hooks:
  enabled: true
  count: 8
  
rrf:
  k: 60
  weights:
    bm25: 0.2
    vector: 0.5
    graph: 0.3

环境变量

# .env
MEMORY_WORKSPACE=/path/to/workspace
MEMORY_VECTOR_MODEL=nomic-embed-text-v1.5
MEMORY_LLM_API_KEY=your-api-key

示例项目

单Agent示例

cd examples/single-agent
python3 run.py

多Agent示例

cd examples/multi-agent
python3 run.py

测试

# 运行所有测试
python3 -m pytest tests/

# 运行特定测试
python3 -m pytest tests/test-hooks.py

文档

贡献

欢迎贡献！请参阅 CONTRIBUTING.md 了解指南。

许可证

MIT License

支持

GitHub Issues: https://github.com/light-office/multi-agent-memory/issues
Discord: https://discord.gg/light-office
邮件: [email protected]

安全使用建议

Install only if you want a persistent local memory system. Use a dedicated MEMORY_WORKSPACE, avoid entering secrets into captured prompts, do not provide an LLM API key unless needed, and ask the publisher to validate agent IDs, pin dependencies, document retention/deletion controls, and clarify that the dashboard metrics are real rather than simulated.

能力标签

requires-sensitive-credentials

能力评估

ℹ Purpose & Capability

The stated purpose, code, and documentation are broadly coherent for a local multi-agent memory/RAG system, including hooks, graph building, conflict logs, token tracking, and agent registry functions.

⚠ Instruction Scope

The documented agent registration flow is not clearly scoped to safe agent names; the code uses the agent_id directly as a filesystem path segment before creating a directory.

ℹ Install Mechanism

Registry metadata says there is no install spec, while SKILL.md frontmatter declares pip installs for lancedb and numpy without version pins. This is purpose-aligned but should be made explicit and pinned.

ℹ Credentials

The skill defaults to a local /tmp memory workspace and mentions an optional LLM API key/provider host. No provided source shows credential leakage, but users should understand where memory and any provider data may go.

ℹ Persistence & Privilege

The skill is designed to persist user/agent activity summaries into local memory files for future retrieval. That is expected for a memory system, but retention, deletion, and exclusion controls are not documented.

如何使用

确保已安装 OpenClaw（本地或 Docker 部署）
在对话框中输入安装命令：/install light-office-multi-agent-memory
安装完成后，直接呼叫该 Skill 的名称或使用 /light-office-multi-agent-memory 触发
根据 Skill 的参数说明提供必要输入，即可获得结构化输出

版本历史

v1.0.0

Initial release of multi-agent-memory 1.0.0: - Provides a general-purpose multi-agent memory system with automatic capture, RRF retrieval, knowledge graph construction, contradiction detection, and token tracking. - Supports dynamic agent management, including registration, listing, health checks, and automatic scaling. - Integrates 8 hooks for memory capture, RRF fusion search (BM25 + Vector + Graph), and knowledge graph features. - Includes benchmarking tools, Git snapshot integration, workflow engine, and a real-time dashboard. - Installation and configuration instructions provided, with example projects and test suite included.

元数据

Slug light-office-multi-agent-memory

版本 1.0.0

许可证 MIT-0

累计安装 0

当前安装数 0

历史版本数 1

常见问题