Running Multiple OpenClaw Agents

Why Multiple Agents?

You might want separate OpenClaw agents for different contexts:

• Work agent — Connected to Slack, focused on professional tasks
• Personal agent — Connected to WhatsApp, casual and personal
• Project agent — Dedicated to a specific project with specialized skills

Each agent has its own memory, skills, and configuration.

Setting Up Multiple Workspaces

Each OpenClaw agent lives in its own workspace directory:

# Work agent
mkdir -p ~/openclaw-work
cd ~/openclaw-work
openclaw onboard

# Personal agent
mkdir -p ~/openclaw-personal
cd ~/openclaw-personal
openclaw onboard

Workspace Structure

~/openclaw-work/
  openclaw.json       # Work configuration
  MEMORY.md           # Work-related memories
  USER.md             # Professional profile
  skills/             # Work-specific skills

~/openclaw-personal/
  openclaw.json       # Personal configuration
  MEMORY.md           # Personal memories
  USER.md             # Personal profile
  skills/             # Personal skills

Running Multiple Agents Simultaneously

Start each agent in its workspace:

# Terminal 1
cd ~/openclaw-work && openclaw start

# Terminal 2
cd ~/openclaw-personal && openclaw start

Port Configuration

If both agents run on the same machine, configure different ports:

{
  "gateway": {
    "port": 3001
  }
}

Shared vs Separate

Shared Across Agents

• Global skills in ~/.openclaw/skills/ (available to all)
• System-level configuration

Separate Per Agent

• MEMORY.md and USER.md
• Channel connections
• Model selection
• Workspace skills
• Conversation history

Use Cases

Work/Life Separation

Keep work context out of personal conversations and vice versa.

Project Specialization

Create an agent loaded with project-specific documentation, skills, and memory for deep context.

Testing

Run a test agent to experiment with new skills or models without affecting your main setup.