multi-agent-architecture

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Translation

Chinese

Multi-Agent Architecture - 多 Agent 架构

Multi-Agent Architecture

概述

Overview

OpenClaw 支持多 Agent 架构，每个 Agent 可以有不同的：

专业领域和 System Prompt
模型配置和成本策略
Channel 绑定和权限
工具集和 MCP 配置

OpenClaw supports multi-agent architecture, where each Agent can have different:

Professional domains and System Prompt
Model configurations and cost strategies
Channel bindings and permissions
Toolkits and MCP configurations

架构设计

Architecture Design

Recommended Agent Division of Labor

┌─────────────────────────────────────────────────────────────┐
│                      Main Agent (小a)                        │
│  - 主会话处理                                                 │
│  - 任务分发和协调                                             │
│  - 复杂决策和规划                                             │
│  - 模型: opus-4.5 (高质量)                                    │
└─────────────────────────────────────────────────────────────┘
                              │
          ┌───────────────────┼───────────────────┐
          ▼                   ▼                   ▼
┌─────────────────┐ ┌─────────────────┐ ┌─────────────────┐
│   News Agent    │ │   Code Agent    │ │  Research Agent │
│  - 新闻抓取     │ │  - 代码生成     │ │  - 深度研究     │
│  - 内容摘要     │ │  - Bug 修复     │ │  - 文档分析     │
│  - 定时推送     │ │  - 代码审查     │ │  - 知识整合     │
│  模型: sonnet   │ │  模型: codex    │ │  模型: opus     │
└─────────────────┘ └─────────────────┘ └─────────────────┘
          │                   │                   │
          ▼                   ▼                   ▼
┌─────────────────┐ ┌─────────────────┐ ┌─────────────────┐
│  Quick Agent    │ │  Batch Agent    │ │  Monitor Agent  │
│  - 快速问答     │ │  - 批量处理     │ │  - 系统监控     │
│  - 简单任务     │ │  - 数据处理     │ │  - 健康检查     │
│  - 低延迟响应   │ │  - 文件操作     │ │  - 告警通知     │
│  模型: flash    │ │  模型: mini     │ │  模型: mini     │
└─────────────────┘ └─────────────────┘ └─────────────────┘

┌─────────────────────────────────────────────────────────────┐
│                      Main Agent (Xiaoa)                        │
│  - Main session processing                                                 │
│  - Task distribution and coordination                                             │
│  - Complex decision-making and planning                                             │
│  - Model: opus-4.5 (High quality)                                    │
└─────────────────────────────────────────────────────────────┘
                              │
          ┌───────────────────┼───────────────────┐
          ▼                   ▼                   ▼
┌─────────────────┐ ┌─────────────────┐ ┌─────────────────┐
│   News Agent    │ │   Code Agent    │ │  Research Agent │
│  - News scraping     │ │  - Code generation     │ │  - In-depth research     │
│  - Content summarization     │ │  - Bug fixing     │ │  - Document analysis     │
│  - Scheduled push notifications     │ │  - Code review     │ │  - Knowledge integration     │
│  Model: sonnet   │ │  Model: codex    │ │  Model: opus     │
└─────────────────┘ └─────────────────┘ └─────────────────┘
          │                   │                   │
          ▼                   ▼                   ▼
┌─────────────────┐ ┌─────────────────┐ ┌─────────────────┐
│  Quick Agent    │ │  Batch Agent    │ │  Monitor Agent  │
│  - Quick Q&A     │ │  - Batch processing     │ │  - System monitoring     │
│  - Simple tasks     │ │  - Data processing     │ │  - Health check     │
│  - Low-latency response   │ │  - File operations     │ │  - Alert notifications     │
│  Model: flash    │ │  Model: mini     │ │  Model: mini     │
└─────────────────┘ └─────────────────┘ └─────────────────┘

Agent 配置

Agent Configuration

1. 创建 Agent 目录结构

1. Create Agent Directory Structure

bash

~/.openclaw/agents/
├── main/           # 主 Agent (已存在)
│   └── agent/
│       ├── AGENTS.md
│       ├── SOUL.md
│       └── ...
├── news/           # 新闻 Agent
│   └── agent/
│       ├── AGENTS.md
│       └── config.json
├── code/           # 代码 Agent
│   └── agent/
│       ├── AGENTS.md
│       └── config.json
├── research/       # 研究 Agent
│   └── agent/
│       ├── AGENTS.md
│       └── config.json
├── quick/          # 快速响应 Agent
│   └── agent/
│       └── config.json
└── batch/          # 批量处理 Agent
    └── agent/
        └── config.json

bash

~/.openclaw/agents/
├── main/           # Main Agent (Exists)
│   └── agent/
│       ├── AGENTS.md
│       ├── SOUL.md
│       └── ...
├── news/           # News Agent
│   └── agent/
│       ├── AGENTS.md
│       └── config.json
├── code/           # Code Agent
│   └── agent/
│       ├── AGENTS.md
│       └── config.json
├── research/       # Research Agent
│   └── agent/
│       ├── AGENTS.md
│       └── config.json
├── quick/          # Quick Response Agent
│   └── agent/
│       └── config.json
└── batch/          # Batch Processing Agent
    └── agent/
        └── config.json

2. Agent 配置示例

2. Agent Configuration Examples

News Agent (

~/.openclaw/agents/news/agent/config.json

)

News Agent (

~/.openclaw/agents/news/agent/config.json

)

json

{
  "model": {
    "primary": "anthropic/claude-sonnet-4-5"
  },
  "systemPrompt": "你是新闻抓取和摘要专家。专注于：\n1. 从权威来源抓取真实新闻\n2. 生成简洁准确的摘要\n3. 确保每条新闻有原文链接\n4. 按时推送到指定渠道",
  "tools": {
    "allow": ["web_fetch", "exec", "message"]
  }
}

json

{
  "model": {
    "primary": "anthropic/claude-sonnet-4-5"
  },
  "systemPrompt": "You are an expert in news scraping and summarization. Focus on:\n1. Scrape real news from authoritative sources\n2. Generate concise and accurate summaries\n3. Ensure each news item has the original link\n4. Push to specified channels on time",
  "tools": {
    "allow": ["web_fetch", "exec", "message"]
  }
}

Code Agent (

~/.openclaw/agents/code/agent/config.json

)

Code Agent (

~/.openclaw/agents/code/agent/config.json

)

json

{
  "model": {
    "primary": "openrouter-vip/gpt-5.2-codex"
  },
  "systemPrompt": "你是代码专家。专注于：\n1. 高质量代码生成\n2. Bug 分析和修复\n3. 代码审查和优化\n4. 技术文档编写",
  "tools": {
    "allow": ["read", "write", "edit", "exec"]
  }
}

json

{
  "model": {
    "primary": "openrouter-vip/gpt-5.2-codex"
  },
  "systemPrompt": "You are a code expert. Focus on:\n1. High-quality code generation\n2. Bug analysis and fixing\n3. Code review and optimization\n4. Technical documentation writing",
  "tools": {
    "allow": ["read", "write", "edit", "exec"]
  }
}

Quick Agent (

~/.openclaw/agents/quick/agent/config.json

)

Quick Agent (

~/.openclaw/agents/quick/agent/config.json

)

json

{
  "model": {
    "primary": "google/gemini-flash-latest"
  },
  "systemPrompt": "你是快速响应助手。特点：\n1. 简洁直接的回答\n2. 低延迟响应\n3. 处理简单查询\n4. 不需要深度分析的任务"
}

json

{
  "model": {
    "primary": "google/gemini-flash-latest"
  },
  "systemPrompt": "You are a quick response assistant. Features:\n1. Concise and direct answers\n2. Low-latency response\n3. Handle simple queries\n4. Tasks that do not require in-depth analysis"
}

3. 在 openclaw.json 中注册 Agent

3. Register Agents in openclaw.json

json

{
  "agents": {
    "entries": {
      "news": {
        "enabled": true,
        "allowSpawnFrom": ["main"]
      },
      "code": {
        "enabled": true,
        "allowSpawnFrom": ["main"]
      },
      "research": {
        "enabled": true,
        "allowSpawnFrom": ["main"]
      },
      "quick": {
        "enabled": true,
        "allowSpawnFrom": ["main"]
      },
      "batch": {
        "enabled": true,
        "allowSpawnFrom": ["main"]
      }
    },
    "defaults": {
      "maxConcurrent": 4,
      "subagents": {
        "maxConcurrent": 8
      }
    }
  }
}

json

{
  "agents": {
    "entries": {
      "news": {
        "enabled": true,
        "allowSpawnFrom": ["main"]
      },
      "code": {
        "enabled": true,
        "allowSpawnFrom": ["main"]
      },
      "research": {
        "enabled": true,
        "allowSpawnFrom": ["main"]
      },
      "quick": {
        "enabled": true,
        "allowSpawnFrom": ["main"]
      },
      "batch": {
        "enabled": true,
        "allowSpawnFrom": ["main"]
      }
    },
    "defaults": {
      "maxConcurrent": 4,
      "subagents": {
        "maxConcurrent": 8
      }
    }
  }
}

智能 Spawn 系统

Intelligent Spawn System

任务分类规则

Task Classification Rules

Main Agent 根据任务类型自动选择合适的 Agent：

任务类型	关键词	目标 Agent	模型
新闻抓取	news, 新闻, 早报, 推送	news	sonnet
代码任务	code, 代码, bug, 开发	code	codex
深度研究	research, 分析, 调研	research	opus
快速问答	简单, 快速, 查询	quick	flash
批量处理	batch, 批量, 文件	batch	mini
复杂任务	保留在 main	main	opus

Main Agent automatically selects the appropriate Agent based on task type:

Task Type	Keywords	Target Agent	Model
News Scraping	news, news, morning briefing, push	news	sonnet
Code Tasks	code, code, bug, development	code	codex
In-depth Research	research, analysis, research	research	opus
Quick Q&A	simple, quick, query	quick	flash
Batch Processing	batch, batch, file	batch	mini
Complex Tasks	Retained in main	main	opus

智能 Spawn 实现

Intelligent Spawn Implementation

python

undefined

python

undefined

在 AGENTS.md 中添加智能 Spawn 逻辑

Add intelligent spawn logic in AGENTS.md

🧠 智能任务分发

🧠 Intelligent Task Distribution

当收到任务时，评估以下因素：

任务复杂度
- 简单查询 → quick agent
- 中等任务 → 专业 agent
- 复杂任务 → main 处理或 research agent
任务类型
- 新闻相关 → news agent
- 代码相关 → code agent
- 研究分析 → research agent
- 批量操作 → batch agent
时间敏感度
- 需要快速响应 → quick agent
- 可以等待 → 专业 agent
资源消耗
- 高 token 消耗 → 使用便宜模型的 agent
- 需要高质量 → 使用 opus 的 agent

When receiving a task, evaluate the following factors:

Task Complexity
- Simple queries → quick agent
- Medium tasks → specialized agents
- Complex tasks → handled by main or research agent
Task Type
- News-related → news agent
- Code-related → code agent
- Research and analysis → research agent
- Batch operations → batch agent
Time Sensitivity
- Requires quick response → quick agent
- Can wait → specialized agents
Resource Consumption
- High token consumption → use agents with cheaper models
- Requires high quality → use agents with opus

Spawn 命令示例

Spawn Command Examples

python

undefined

python

undefined

新闻任务

News Task

sessions_spawn( task="抓取今日科技新闻并推送到 DailyNews 群组", agentId="news", label="news-morning" )

sessions_spawn( task="Scrape today's tech news and push to the DailyNews group", agentId="news", label="news-morning" )

代码任务

Code Task

sessions_spawn( task="修复 auth.py 中的登录 bug", agentId="code", label="fix-auth-bug" )

sessions_spawn( task="Fix the login bug in auth.py", agentId="code", label="fix-auth-bug" )

研究任务

Research Task

sessions_spawn( task="深度分析 GPT-5 的技术架构", agentId="research", label="gpt5-analysis" )

sessions_spawn( task="In-depth analysis of GPT-5's technical architecture", agentId="research", label="gpt5-analysis" )

快速查询

Quick Query

sessions_spawn( task="查询今天的天气", agentId="quick", label="weather-check" )

undefined

sessions_spawn( task="Check today's weather", agentId="quick", label="weather-check" )

undefined

并发处理

Concurrent Processing

配置并发限制

Configure Concurrency Limits

json

{
  "agents": {
    "defaults": {
      "maxConcurrent": 4,      // 主 agent 最大并发
      "subagents": {
        "maxConcurrent": 8    // 子 agent 最大并发
      }
    }
  }
}

json

{
  "agents": {
    "defaults": {
      "maxConcurrent": 4,      // Maximum concurrency for main agent
      "subagents": {
        "maxConcurrent": 8    // Maximum concurrency for sub agents
      }
    }
  }
}

并发场景

Concurrency Scenarios

用户消息 → Main Agent
              │
              ├─→ spawn(news) ──→ 抓取新闻
              │
              ├─→ spawn(code) ──→ 修复 bug
              │
              └─→ spawn(research) ──→ 深度分析
              
              ↓ (并行执行)
              
         所有任务完成后汇报

User message → Main Agent
              │
              ├─→ spawn(news) ──→ Scrape news
              │
              ├─→ spawn(code) ──→ Fix bug
              │
              └─→ spawn(research) ──→ In-depth analysis
              
              ↓ (Executed in parallel)
              
         Report after all tasks are completed

Channel 绑定

Channel Binding

不同 Channel 使用不同 Agent

Use Different Agents for Different Channels

json

{
  "channels": {
    "telegram": {
      "defaultAgent": "main"
    },
    "whatsapp": {
      "defaultAgent": "main"
    }
  },
  "agents": {
    "entries": {
      "news": {
        "channels": ["telegram-newsbot"]
      }
    }
  }
}

json

{
  "channels": {
    "telegram": {
      "defaultAgent": "main"
    },
    "whatsapp": {
      "defaultAgent": "main"
    }
  },
  "agents": {
    "entries": {
      "news": {
        "channels": ["telegram-newsbot"]
      }
    }
  }
}

监控和管理

Monitoring and Management

查看活跃 Session

View Active Sessions

bash

undefined

bash

undefined

列出所有 session

List all sessions

openclaw sessions list

查看特定 agent 的 session

View sessions for a specific agent

openclaw sessions list --agent news

undefined

openclaw sessions list --agent news

undefined

查看 Spawn 状态

View Spawn Status

python

undefined

python

undefined

在代码中

In code

sessions_list(kinds=["spawn"], limit=10)

undefined

sessions_list(kinds=["spawn"], limit=10)

undefined

最佳实践

Best Practices

1. 任务分发原则

1. Task Distribution Principles

简单任务不 spawn - 直接处理更快
耗时任务必 spawn - 不阻塞主会话
相关任务批量 spawn - 提高效率

Do not spawn for simple tasks - Direct processing is faster
Always spawn for time-consuming tasks - Do not block the main session
Spawn related tasks in batches - Improve efficiency

2. 模型选择原则

2. Model Selection Principles

质量优先 → opus
速度优先 → flash
代码任务 → codex
成本优先 → mini

Quality first → opus
Speed first → flash
Code tasks → codex
Cost first → mini

3. 错误处理

3. Error Handling

python

undefined

python

undefined

spawn 时设置超时

Set timeout when spawning

sessions_spawn( task="...", agentId="code", runTimeoutSeconds=300, # 5分钟超时 cleanup="keep" # 保留 session 用于调试 )

undefined

sessions_spawn( task="...", agentId="code", runTimeoutSeconds=300, # 5-minute timeout cleanup="keep" # Keep session for debugging )

undefined

Related Resources

由小a设计 - 实现真正的多 Agent 协作

Designed by Xiaoa - Achieve true multi-agent collaboration

multi-agent-architecture

Original

Translation

Multi-Agent Architecture - 多 Agent 架构

Multi-Agent Architecture

概述

Overview

架构设计

Architecture Design

推荐的 Agent 分工

Recommended Agent Division of Labor

Agent 配置

Agent Configuration

1. 创建 Agent 目录结构

1. Create Agent Directory Structure

2. Agent 配置示例

2. Agent Configuration Examples

News Agent (~/.openclaw/agents/news/agent/config.json)

News Agent (~/.openclaw/agents/news/agent/config.json)

Code Agent (~/.openclaw/agents/code/agent/config.json)

Code Agent (~/.openclaw/agents/code/agent/config.json)

Quick Agent (~/.openclaw/agents/quick/agent/config.json)

Quick Agent (~/.openclaw/agents/quick/agent/config.json)

3. 在 openclaw.json 中注册 Agent

3. Register Agents in openclaw.json

智能 Spawn 系统

Intelligent Spawn System

任务分类规则

Task Classification Rules

智能 Spawn 实现

Intelligent Spawn Implementation

在 AGENTS.md 中添加智能 Spawn 逻辑

Add intelligent spawn logic in AGENTS.md

🧠 智能任务分发

🧠 Intelligent Task Distribution

Spawn 命令示例

Spawn Command Examples

新闻任务

News Task

代码任务

Code Task

研究任务

Research Task

快速查询

Quick Query

并发处理

Concurrent Processing

配置并发限制

Configure Concurrency Limits

并发场景

Concurrency Scenarios

Channel 绑定

Channel Binding

不同 Channel 使用不同 Agent

Use Different Agents for Different Channels

监控和管理

Monitoring and Management

查看活跃 Session

View Active Sessions

列出所有 session

List all sessions

查看特定 agent 的 session

View sessions for a specific agent

查看 Spawn 状态

View Spawn Status

在代码中

In code

最佳实践

Best Practices

1. 任务分发原则

1. Task Distribution Principles

2. 模型选择原则

2. Model Selection Principles

3. 错误处理

3. Error Handling

spawn 时设置超时

Set timeout when spawning

相关资源

Related Resources

News Agent (
`~/.openclaw/agents/news/agent/config.json`
)

News Agent (
`~/.openclaw/agents/news/agent/config.json`
)

Code Agent (
`~/.openclaw/agents/code/agent/config.json`
)

Code Agent (
`~/.openclaw/agents/code/agent/config.json`
)

Quick Agent (
`~/.openclaw/agents/quick/agent/config.json`
)

Quick Agent (
`~/.openclaw/agents/quick/agent/config.json`
)