Skip to main content

Multi-Agent System

Build and deploy multi-agent systems with ARC Protocol.

Overview

Multi-agent systems distribute functionality across specialized agents that communicate via ARC Protocol.

Architecture

┌─────────────────────────────────────────┐
│ Client Application │
└────────────────┬────────────────────────┘
│ ARC Protocol
┌────────────┼────────────┐
▼ ▼ ▼
┌─────────┐ ┌─────────┐ ┌─────────┐
│ Agent A │ │ Agent B │ │ Agent C │
│Finance │ │ HR │ │ Support │
└─────────┘ └─────────┘ └─────────┘

Deployment Options

Single Server

Deploy multiple agents on one server for simplicity and low latency.

Best for:

  • Development and testing
  • Related business domains
  • Low to medium traffic
  • Shared dependencies

Guide: Single Server Deployment

Distributed

Deploy agents across multiple servers for scaling and isolation.

Best for:

  • Production systems
  • Independent scaling requirements
  • Fault isolation
  • High availability

Guide: Distributed Deployment

Agent Communication

Agents communicate via ARC Protocol with workflow tracing.

Guide: Agent Communication

Key Concepts

Agent Independence

Each agent is self-contained with its own:

  • Business logic
  • Data storage
  • Dependencies
  • Scaling characteristics

Protocol-Based Communication

Agents communicate through ARC Protocol, not direct calls:

  • Loose coupling
  • Technology independence
  • Easy testing and mocking
  • Clear boundaries

Workflow Tracing

traceId enables end-to-end tracking across agent interactions.

When to Use Multi-Agent

Use multi-agent systems when:

  1. Domain Separation - Clear business domain boundaries
  2. Independent Scaling - Different agents have different load
  3. Team Organization - Different teams own different agents
  4. Technology Diversity - Agents use different tech stacks

Quick Start

1. Define Agents

Finance Agent - Handles invoices, payments
HR Agent - Manages employee queries, benefits
Support Agent - General customer support

2. Choose Deployment

  • Development: Single server
  • Production: Distributed

3. Implement

Follow the appropriate guide:

4. Connect

Implement agent-to-agent communication: