Multi-Agent System overview

A Multi-Agent System (MAS) puts several AI Agents behind a single conversation. A router reads each incoming message, decides which agent is best suited to answer, and hands the conversation to it. To the person chatting, it looks like one assistant that happens to be good at everything.

You build a MAS in Agentic Workspace → MAS Systems.

The Conversation Router

The Conversation Router is the orchestration pattern behind every system you build. It works turn by turn:

A message arrives.
The router classifies it against what each member agent can do.
One agent answers.
The next message is classified again — so the conversation can move between specialists.

Each turn is answered by exactly one agent. The router never fans a question out to several agents, and it never runs a plan-then-verify loop. That focus is what makes it predictable.

Skills are the routing signal

This is the most important idea on this page. The router does not read descriptions you write in the MAS builder. It routes on the skills defined on each member agent — their names, descriptions, tags, and example queries.


"What was total revenue last week?"
        │
        ▼
   ┌─────────┐   classify against every member's skills
   │ Router  │──────────────────────────────────┐
   └─────────┘                                  │
        │  best match                            │
        ▼                                        ▼
  Revenue Analyst                        Inventory Specialist
  skill: Sales Lookup                    skill: Stock Check

Well-written skills give you accurate routing with no extra configuration. Vague or overlapping skills are the single biggest cause of messages landing on the wrong agent. Skills are edited on each agent, not in the MAS builder — see Skills.

💡

Before building a system, open each candidate agent and check that its skills have real descriptions and example queries. That work pays off directly as routing accuracy.

When to use a MAS

Use a MAS when	Use a single AI Agent when
Distinct domains need genuinely different instructions, tools, or data	One set of instructions covers everything
Different teams own different agents and ship on their own schedules	One team owns the whole assistant
You want each specialist to stay small and testable on its own	The skills are few and closely related
You already have working agents and want one front door for them	You are starting from scratch

A single AI Agent already picks between its own skills. Reach for a MAS when the specialists deserve to be separate agents — separate prompts, separate tools, separate owners — not merely separate skills.

What a system is made of

Piece	What it is
The router	Configuration, not an agent. It classifies and delegates; it never answers by itself.
Member agents	The specialists. Each must be an AI Agent with status Active.
Default action	What happens when nothing matches confidently.
Handoff behaviour	What the user sees when the conversation moves to another agent.

Task Agents cannot be members. If a system needs a deterministic pipeline, expose it as a tool on one of the member agents instead — see Tools.

Lifecycle

Create the system — Create a MAS.
Connect agents — Connect agents.
Shape routing — Routing.
Handle the edges — Fallback & handoff.
Test in the builder — Test the MAS.
Publish a version — Publish.
Deploy it — Deploy.

Availability

Starter MAS Systems is available on Starter and above. On the Free plan the section is hidden from the project sidebar.
Member agents must be published before the system can be deployed.

Create a MAS — start building.
AI Agent overview — build the specialists first.