TutorialFeatured

Setting Up Multi-Agent Routing in OpenClaw

February 1, 20268 min readReviewed March 8, 2026

Multi-agent routing allows you to route different channels and accounts to isolated OpenClaw agents with separate sessions. This enables powerful use cases like separating work and personal accounts, or dedicating agents to specific tasks.[1]

What is Multi-Agent Routing?

By default, OpenClaw runs a single agent instance that handles all your conversations. Multi-agent routing lets you:

Isolate contexts: Separate work conversations from personal ones
Dedicate agents: Assign specific agents to specific platforms or tasks
Independent sessions: Each agent maintains its own conversation history and context
Separate configurations: Different API keys, settings, or skills per agent

Use Cases

Work vs Personal: Route work Slack to Agent A, personal WhatsApp to Agent B
Platform-specific: Dedicated agents for Discord moderation vs Telegram assistance
Task-specific: One agent for coding tasks, another for research
Team isolation: Multiple team members with separate OpenClaw instances

Configuration File Structure

Multi-agent routing is configured in your OpenClaw configuration file:

~/.config/openclaw/routing.json

Basic Setup Example

Here's a simple configuration that routes two different channels to separate agents:

            {
  "routes": {
    "telegram:personal": {
      "agent": "personal-agent",
      "config": {
        "model": "claude-opus-4.5",
        "contextLimit": "200k"
      }
    },
    "slack:work": {
      "agent": "work-agent",
      "config": {
        "model": "gpt-4",
        "contextLimit": "100k",
        "skills": ["jira", "confluence"]
      }
    }
  }
}
        

Platform-Specific Routing

Configure routing for specific platforms using the platform identifier:

WhatsApp Routing

            "whatsapp:+1234567890": {
  "agent": "primary-agent"
}

"whatsapp:+0987654321": {
  "agent": "secondary-agent"
}
        

Telegram Routing

            "telegram:personal_bot": {
  "agent": "personal-agent",
  "whitelist": ["@user1", "@user2"]
}
        

Discord Routing

            "discord:guild_123": {
  "agent": "discord-agent",
  "channels": {
    "general": "read-write",
    "admin": "admin-only"
  }
}
        

Agent Configuration Options

Each routed agent can have its own configuration:

model — Which LLM to use (claude-opus-4.5, gpt-4, etc.)
contextLimit — Maximum tokens for context window
temperature — AI randomness/creativity setting
skills — Array of skills to load for this agent
systemPrompt — Custom system prompt override

Advanced: Conditional Routing

You can route based on message content using patterns:

            "routes": {
  "telegram:personal": {
    "agent": "personal-agent",
    "conditions": {
      "code_keywords": ["bug", "fix", "deploy"],
      "route_to": "coding-agent"
    }
  }
}
        

Testing Your Configuration

After setting up routing, verify your configuration:

            # Validate routing configuration
openclaw routing validate

# Test routing for a specific channel
openclaw routing test --channel="telegram:personal"

# View active routes
openclaw routing list
        

Reloading Configuration

After making changes, reload the routing configuration:

openclaw reload

Or restart the daemon:

openclaw daemon restart

Troubleshooting

Messages Not Routing

If messages aren't reaching the expected agent:

Check the route pattern matches your channel identifier
Verify the agent name is spelled correctly
Check logs: openclaw logs --tail
Ensure no conflicting routes exist

Agent Not Starting

Verify API keys are configured for each agent
Check if the agent name exists in your config
Ensure sufficient system resources for multiple agents

Best Practices

Start simple: Begin with one route, test, then expand
Name descriptively: Use clear agent names like "work-coding" not "agent2"
Document routes: Comment your routing.json for future reference
Monitor costs: Each agent has separate API costs
Backup config: Keep copies of working configurations

What's Next?

With multi-agent routing configured, explore these advanced topics:

Complete OpenClaw Guide — Master all features
Security Guide — Secure your multi-agent setup
Official Routing Documentation — Deep dive reference

References

OpenClaw Documentation - Multi-Agent Routing - Accessed February 2026
OpenClaw GitHub - Routing Configuration Examples - Accessed February 2026