OpenClaw Multi-Agent System

1. Course Content

Course Overview

• The OpenClaw multi-Agent feature allows running multiple isolated agents in the same Gateway. Each agent has its own workspace, state directory, and session history
• The multi-Agent architecture is suitable for multi-role division scenarios. For example, one agent is responsible for programming development while another handles social media operations, with data isolated from each other and no mutual interference

Learning Objectives

1. Understand the concept and applicable scenarios of the OpenClaw multi-Agent system
2. Master creating new agents through CLI commands
3. Master the method of editing agent prompt files (AGENTS.md / SOUL.md / IDENTITY.md, etc.)
4. Master the operation of deleting agents

2. Case Demo

OpenClaw runs one agent by default (agentId is main). Additional isolated agents can be created in the following ways.

2.1 Creating with the CLI Agent Wizard

For example, create a work agent named work:

xxxxxxxxxx
openclaw agents add work

After execution, OpenClaw automatically completes the following operations:

1. Create the agent directory: ~/.openclaw/agents/<agentId>/
2. Create workspace bootstrap files: including AGENTS.md, SOUL.md, IDENTITY.md, TOOLS.md, USER.md, etc.

2.2 Agent Directory Structure

After each agent is created, a complete file structure is generated in the ~/.openclaw/agents/<agentId>/ directory:

xxxxxxxxxx
~/.openclaw/agents/<agentId>/
├── agent/
│   ├── auth-profiles.json    # Authentication profile file
│   ├── config.json           # Agent configuration
│   └── ...
└── sessions/                 # Session history storage
    └── <SessionId>.jsonl

The agent's workspace files are located in their respective workspace directories, where you can edit the bootstrap files to define the agent's behavior and personality.

2.3 Verifying Agent Creation

After creation, you can view all agents and their binding status with the following command:

xxxxxxxxxx
openclaw agents list --bindings

2.4 Restarting the Gateway for the Configuration to Take Effect

After creating an agent or modifying the binding configuration, you need to restart the Gateway:

xxxxxxxxxx
openclaw gateway restart

After restarting, verify the agent and channel status:

xxxxxxxxxx
openclaw agents list --bindings

2.5 Agent Conversation

xxxxxxxxxx
openclaw tui

• In the dialog box, enter /agent and press Enter to select the newly created agent

• Enter text in the input box to converse with the agent

4. Deleting Agents

4.1 Deleting with CLI Commands

For example, delete an agent named work:

xxxxxxxxxx
openclaw agents delete work

4.2 Manually Cleaning Up the Agent Directory

After running the CLI command, it is recommended to also clean up the following residual data:

​x
# Delete the agent data directory
rm -rf ~/.openclaw/agents/<agentId>
​
# Delete the agent workspace
rm -rf ~/.openclaw/workspace-<agentId>

4.3 Cleaning Up Binding Configuration

After deleting an agent, you also need to remove the corresponding binding records and the agent list entries from the openclaw.json configuration file:

1. Open the configuration file: ~/.openclaw/openclaw.json
2. Remove the corresponding agent entry from agents.list
3. Remove the binding rules related to this agent from bindings
4. Save and restart the gateway:

openclaw gateway restart