Multi-Agent Routing

Using multiple specialized agents.

Overview

Multi-agent routing allows different agents to handle different types of requests:

┌─────────────────────────────────────────┐
│                 Router                   │
│  ┌─────────┐  ┌─────────┐  ┌─────────┐ │
│  │ Coding  │  │ Writing │  │ Research│ │
│  │ Agent   │  │  Agent  │  │  Agent  │ │
│  └─────────┘  └─────────┘  └─────────┘ │
└─────────────────────────────────────────┘

Creating Agents

# Coding agent
openclaw agents create coding \
  --model claude-3-5-sonnet \
  --prompt "You are a coding assistant..."

# Writing agent
openclaw agents create writing \
  --model gpt-4 \
  --prompt "You are a creative writer..."

# Research agent
openclaw agents create research \
  --model claude-3-opus \
  --prompt "You are a research assistant..."

Routing Rules

Configure how messages are routed:

{
  routing: {
    rules: [
      {
        pattern: "code|programming|debug",
        agent: "coding"
      },
      {
        pattern: "write|essay|story",
        agent: "writing"
      }
    ],
    default: "general"
  }
}

Routing Methods

Pattern Matching

Route based on message content patterns.

Channel-Based

Route based on source channel:

{
  routing: {
    channels: {
      "discord-coding": "coding",
      "telegram-personal": "general"
    }
  }
}

AI-Based

Let an AI decide which agent to use:

{
  routing: {
    method: "ai",
    routerModel: "claude-3-haiku"
  }
}

Manual Switching

Users can manually switch agents:

/agent coding

See Also

• Agent
• CLI: agents