openclaw agents: Multi-Agent Management Guide

Command Overview

The openclaw agents command is used to manage multiple AI agents. Each agent can have a different persona, model, skill set, and behavior configuration, making it ideal for scenarios that require multi-role AI.

View Agent List

openclaw agents list

Output:

Agents:
  default        gpt-4o-mini    Active    General assistant
  code-helper    gpt-4o         Active    Programming assistant
  translator     gpt-4o-mini    Active    Translation specialist
  customer-svc   claude-sonnet  Inactive  Customer service bot

Create a New Agent

openclaw agents create assistant-cn

Interactive creation:

Creating new agent: assistant-cn

Display name: Chinese Assistant
Description: AI assistant focused on Chinese conversation
Model [gpt-4o-mini]: qwen-72b
System prompt: You are a professional Chinese AI assistant, skilled in Chinese writing, translation, and knowledge Q&A. Always respond in Chinese.

Agent 'assistant-cn' created successfully.

Non-interactive creation:

openclaw agents create code-expert \
  --display-name "Code Expert" \
  --model gpt-4o \
  --system "You are a senior full-stack engineer, skilled in code review, architecture design, and technical documentation." \
  --tools code_exec,web_search \
  --temperature 0.3

View Agent Details

openclaw agents show code-expert

Output:

Agent: code-expert
  Display Name: Code Expert
  Model: gpt-4o
  Temperature: 0.3
  Max Tokens: 4096
  System Prompt: You are a senior full-stack engineer...
  Tools: code_exec, web_search
  Channels: discord-dev
  Status: Active
  Created: 2026-03-14
  Messages: 1,520

Modify Agent Configuration

# Change model
openclaw agents update code-expert --model gpt-4o-mini

# Change system prompt
openclaw agents update code-expert --system "New system prompt"

# Change temperature
openclaw agents update code-expert --temperature 0.5

# Add a tool
openclaw agents update code-expert --add-tool image_gen

Assign an Agent to a Channel

# Assign agent to a channel
openclaw agents assign code-expert --channel discord-dev
openclaw agents assign assistant-cn --channel wechat-main

# View an agent's channel assignments
openclaw agents channels code-expert

A channel can only be bound to one agent, but one agent can serve multiple channels.

Inter-Agent Routing

Configure message routing rules to automatically dispatch messages to different agents based on content:

{
  "routing": {
    "enabled": true,
    "rules": [
      {
        "match": "code|programming|bug|function",
        "agent": "code-expert"
      },
      {
        "match": "translate|translation",
        "agent": "translator"
      },
      {
        "match": ".*",
        "agent": "default"
      }
    ]
  }
}

Enable/Disable Agents

# Disable an agent (stops processing messages)
openclaw agents disable customer-svc

# Enable an agent
openclaw agents enable customer-svc

Delete an Agent

openclaw agents delete old-agent

Note: Deleting an agent does not delete its conversation history.

Agent Statistics

openclaw agents stats code-expert

Agent Stats: code-expert (last 7 days)
  Messages processed: 342
  Avg response time: 2.1s
  Token usage: 125,000
  Estimated cost: $1.25
  Active sessions: 5
  Error rate: 0.3%

Bulk Management

# Export all agent configurations
openclaw agents export agents-backup.json

# Import agent configurations
openclaw agents import agents-backup.json

# Bulk update model
openclaw agents update-all --model gpt-4o-mini

Defining Agents in Config Files

{
  "agents": {
    "code-expert": {
      "displayName": "Code Expert",
      "model": "gpt-4o",
      "systemPrompt": "You are a senior engineer...",
      "temperature": 0.3,
      "tools": ["code_exec", "web_search"],
      "maxTokens": 4096
    },
    "translator": {
      "displayName": "Translator",
      "model": "gpt-4o-mini",
      "systemPrompt": "You are a professional translator...",
      "temperature": 0.3,
      "maxTokens": 2048
    }
  }
}

Summary

openclaw agents lets you run multiple AI agents with different capabilities and personas within a single OpenClaw instance. Through thoughtful agent specialization and routing configuration, you can deliver customized AI services to different user groups.