Multi-Agent Workflows

Deploy multiple secure OpenClaw instances that communicate with each other for specialized tasks.

What are Multi-Agent Workflows?

Instead of one agent doing everything, you can deploy multiple specialized agents that work together. Each agent is a separate secure OpenClaw container on a shared network, allowing them to call each other's APIs.

Plan limits: Starter (1 agent), Team (5 agents), Business (25 agents), Enterprise (100 agents)

How Agents Communicate

All your agents share a private internal network. They can call each other's APIs directly — no public endpoints, no extra configuration. Find each agent's internal address in your Clawctl dashboard under Agents.

Use Cases

Code Review Pipeline

A PR agent reviews code, calls security-scanner for vulnerabilities, then calls docs-updater to regenerate documentation.

PR → security-scanner → docs-updater → merge

Research Team

A researcher agent finds papers and data, then hands off to a summarizer agent for synthesis and reporting.

researcher → summarizer → report

DevOps Automation

A monitoring agent detects issues and triggers a remediation agent to apply fixes automatically.

monitor → alert → remediation

Domain Specialists

Separate agents for finance, security, and legal review, each with specialized skills and audit trails.

request → finance | security | legal → approved

Setting Up Agents

Via Dashboard

Go to Dashboard → Agents to manage your agents visually. You can add, view, and remove agents with a few clicks.

Inter-Agent API Calls

From within an agent, you can call another agent's API. Here's an example of how agent "default" might call agent "code-reviewer":

// Call another agent's API via its internal address
const response = await fetch('<agent-internal-url>/api/review', {
  method: 'POST',
  headers: { 'Content-Type': 'application/json' },
  body: JSON.stringify({ pr_url: 'https://github.com/...' })
});

Security Note

Each agent has its own audit trail. High-risk actions still require approval, regardless of which agent initiates them. The approval system tracks which agent requested each action.