nexus/AI/openclaw/OpenClaw 多 Agent 系统设计.md

#openclaw #agent #telegram 
```table-of-contents
```

---

# OpenClaw 多 Agent 自动化系统架构笔记 (Advanced)

Author: Billy  
Purpose: 构建基于 OpenClaw + Telegram + n8n 的 AI 自动化控制系统  

---

# 1 系统总体架构

**目标：**

构建一个 **AI Automation Control Center**

**核心组件：**

- OpenClaw Gateway
- 多 Agent
- Telegram Interface
- n8n Workflow Engine

**系统架构：**

```
            Telegram
               │
               ▼
        Telegram Bot
               │
               ▼
             n8n
        (Command Router)
               │
               ▼
        OpenClaw Gateway
               │
               ▼
      ┌────────┬────────┬────────┐
      ▼        ▼        ▼        ▼
   Router    Dev      Research    Ops
    Agent    Agent      Agent     Agent
```

**系统功能：**

| 组件 | 作用 |
|------|------|
| Telegram | 用户交互 |
| n8n | 任务路由 |
| OpenClaw | AI Agent Runtime |
| Agents | 任务执行 |

---

# 2 OpenClaw 基础目录结构

**默认路径：** `~/.openclaw/`

**结构：**

```
.openclaw/
│
├── agents/
│   │
│   ├── main/
│   │   └── agent/
│   │       ├── authprofile.json
│   │       └── models.json
│   │
│   └── research/
│       └── agent/
│           ├── authprofile.json
│           └── models.json
│
└── workspace/
    │
    ├── skills/
    ├── memory/
    ├── identity/
    │   └── Identity.md
    ├── logs/
    ├── devices/
    ├── completions/
    └── canvas/
```

**说明：**

| 目录 | 作用 |
|------|------|
| agents | Agent Profile |
| workspace | 运行数据 |
| skills | 技能插件 |
| memory | AI记忆 |
| identity | System Prompt |
| logs | 日志 |

---

# 3 多 Agent 设计

**创建 Agent：**

```bash
openclaw agent create dev
openclaw agent create research
openclaw agent create ops
openclaw agent create orchestrator
```

**推荐 Agent 列表：**

| Agent | 职责 |
|-------|------|
| orchestrator | 任务调度 |
| dev | 编程 |
| research | 搜索分析 |
| ops | 服务器运维 |
| data | 数据处理 |

**推荐结构：**

```
agents/
│
├── orchestrator/
├── dev/
├── research/
├── ops/
└── data/
```

---

# 4 Agent Identity 设计

Identity 代表：System Prompt

**建议放在：** `agents/dev/agent/identity.md`

**示例：**

```
You are DevAgent.

Responsibilities:
- write code
- debug programs
- generate scripts
- create docker configurations
```

**Research Agent：**

```
You are ResearchAgent.

Responsibilities:
- research information
- summarize technical topics
- analyze documents
```

---

# 5 Skills 机制

**Skills 目录：** `workspace/skills`

**示例：**

```
skills/
├── browser
├── filesystem
├── python
├── telegram
└── self-improving-agent
```

**特点：**

- 所有 Agent 共享
- 类似插件

---

# 6 Memory 设计

**默认 memory：** `workspace/memory`

默认共享。

**memory 类型：**

```
memory/
├── episodic
├── semantic
└── vector
```

**推荐 Memory 分层：**

```
memory/
├── system
├── dev
├── research
└── ops
```

这样避免：memory pollution

**Agent 指定 namespace：**

```json
{
  "memoryNamespace": "dev"
}
```

---

# 7 模型配置

每个 Agent 独立：`agents/<agent>/agent/models.json`

**示例：**

```json
{
  "default": "gpt-4o-mini",
  "models": {
    "gpt-4o-mini": {
      "provider": "openai",
      "model": "gpt-4o-mini"
    },
    "deepseek-coder": {
      "provider": "openai-compatible",
      "base_url": "https://api.deepseek.com/v1",
      "model": "deepseek-coder"
    }
  }
}
```

**推荐模型分配：**

| Agent | Model |
|-------|-------|
| orchestrator | gpt-4o-mini |
| dev | deepseek-coder |
| research | gpt-4o |
| ops | claude-sonnet |

**优点：**

- 降低成本
- 提升能力

---

# 8 Agent 切换

**切换 active agent：**

```bash
openclaw agent use research
```

**行为：**

```
current_agent = research
```

下一条消息：

```
Telegram → research agent
```

不需要重启 Gateway。

---

# 9 Telegram 集成

OpenClaw 支持：Telegram Bot

**结构：**

```
Telegram
   │
   ▼
OpenClaw Gateway
```

**默认：** 所有消息 → active agent

---

# 10 Telegram 指令体系

**推荐命令：**

```
/dev
/research
/ops
/data
/help
```

**示例：**

```
/dev 写一个python脚本
/research 查一下OpenClaw架构
/ops restart docker
```

在 BotFather 设置：`/setcommands`

**示例：**

```
dev - coding tasks
research - research tasks
ops - server tasks
```

---

# 11 n8n Command Router

推荐使用：n8n 进行任务路由。

**架构：**

```
Telegram
   │
   ▼
   n8n
   │
   ▼
OpenClaw
```

**Workflow：**

```
Telegram Trigger
   │
   ▼
Parse Command
   │
   ▼
Agent Router
   │
   ▼
HTTP → OpenClaw
```

**解析示例：**

```javascript
const msg = $json.message.text

if(msg.startsWith("/dev")){
  return {agent:"dev"}
}

if(msg.startsWith("/research")){
  return {agent:"research"}
}
```

---

# 12 Router Agent

Router Agent 负责：Task Distribution

**Prompt 示例：**

```
You are a routing agent.

Rules:
/dev → DevAgent
/research → ResearchAgent
/ops → OpsAgent
```

---

# 13 推荐系统架构

Mac Mini AI Control Center：

```
Mac Mini
│
├─ OpenClaw Gateway
├─ n8n
└─ Telegram Bot

Agents
│
├─ orchestrator
├─ dev
├─ research
├─ ops
└─ data
```

**数据流：**

```
Telegram
   ↓
n8n Router
   ↓
OpenClaw
   ↓
  Agents
```

---

# 14 高级玩法

**支持：**

### 多服务器 Agent

```
Mac Mini
  orchestrator

Ubuntu Server
  dev agent

NAS
  data agent
```

**通信：**

```
HTTP
Redis
MQ
```

---

# 15 最佳实践

### 1 Agent 专业化

**正确：**

```
dev-agent
ops-agent
research-agent
```

**错误：**

```
python-agent
browser-agent
```

---

### 2 模型分工

不要所有 agent 使用同一个模型。

**否则：**

```
multi-agent ≈ single-agent
```

---

### 3 Memory 分层

**推荐：**

```
memory/
├── system
├── dev
├── research
└── ops
```

---

### 4 Router 统一入口

**不要：**

```
Telegram → 各 agent
```

**应该：**

```
Telegram → Router → Agents
```

---

# 16 最终架构总结

**最终系统：**

```
Telegram
   │
   ▼
 n8n Router
   │
   ▼
OpenClaw Gateway
   │
   ▼
┌──────────────┬──────────────┬──────────────┐
   DevAgent  ResearchAgent    OpsAgent
```

**系统特点：**

- 多 Agent
- 多模型
- 自动任务分发
- Telegram 控制
- 自动化工作流

---

# 17 未来扩展

**未来可扩展：**

- AI 自动任务
- 自动服务器运维
- 自动代码生成
- 自动数据分析
- 自动监控告警

**目标：** 构建 Personal AI Operations Center