Multi agents

A multi agent runs the exact steps you design, in the same order every time - so nothing is missed and every run stays consistent and auditable. It's the deterministic counterpart to a single agent: instead of working out each reply on its own from your knowledge, a multi agent follows a flow you lay out on a canvas.

Reach for one whenever a task has a fixed procedure that must run the same way every time - order returns, loan applications, KYC checks, booking an appointment. These are the cases where an LLM-driven agent, however well-prompted, is hard to keep on the rails: it can skip a step, ask things out of order, or finish early. A multi agent removes that risk by making the path explicit.

Multi agents live on the same Agents page as single agents, under their own tab. You create one from the same Create picker, then build it on the flow canvas.

Nexus Tip

Ask the AI layer to scaffold a multi agent from a one-line description - it lays out the start, the steps, and the exit branches between them, then waits for your confirmation before saving. Refine it on the canvas afterwards.

Ask the AI layerScaffold a multi agent for an order-return flow: collect the order ID, check eligibility, then either confirm the return or transfer to a human

Single vs multi agent - which should I build?

Both are agents you create from the Agents page; the difference is how they decide what to do next.

If the task is…	Build a…	Why
Open-ended Q&A - answering questions from your knowledge	Single agent	It works out each reply on its own; no fixed path to enforce.
Troubleshooting and support - diagnosing errors, fixing login/device issues, guiding someone to a resolution step by step	Single agent	The path depends on what the user says; the agent adapts turn by turn.
A multi-step procedure that must run the same way every time - order returns, loan applications, KYC checks, booking an appointment	Multi agent	The steps and their order are fixed; you want every run consistent and auditable.
A regulated or audited process where skipping or reordering a step is unacceptable	Multi agent	The flow guarantees each step runs, in order, on every run.

A useful test: if you'd describe the work as "it depends on what they ask," build a single agent. If you'd describe it as "first do this, then this, then this - every time," build a multi agent.

You don't have to choose only one. The two compose: a single agent can hand off to a multi agent when a conversation reaches a part that needs a fixed procedure (for example, a support agent that hands off to a kyc-verification multi agent once the user is ready to verify their identity). The single agent stays in charge of the open-ended conversation; the multi agent runs the deterministic part and returns control. See How multi agents become reachable below.

What's in a multi agent

A multi agent is a flow on a canvas - a graph of nodes wired together by exit branches. When you open one, you'll see:

• Start - the entry node. It carries the trigger description (what the multi agent is for / when it runs).
• Nodes - the steps. The main node types are:
  • Agent node - a focused agent that owns one step of the flow (collect details, confirm, answer a sub-question). It has its own instructions, tools, and exits. This is the workhorse node - see The Agent node.
  • Execute Workflow node - runs a workflow (a callable, deterministic skill - an API call, a lookup, a calculation) as a step.
  • Transfer to Agent node - hands the conversation to another agent or flow, then ends this one.
• Exit branches - every node finishes on one of its exits, and each exit is wired by a branch to the next node. This is how the flow's order is enforced: an Agent node that exits on, say, eligible goes to one node; on not_eligible it goes to another; on max_retry or on_error it goes somewhere else. The wiring is the procedure.

Because the path is drawn out explicitly, a multi agent is easy to reason about and audit: you can look at the canvas and see every route the conversation can take.

For a full tour of the canvas, see The flow canvas.

Where multi agents live

Multi agents share the Agents page (AI Agent → Agents) with single agents, but each type has its own tab:

• Single agents - your LLM-driven agents. Each row shows a round avatar, the agent's name, its trigger description, and a Live/Draft status.
• Multi agents - your multi agents. Each row shows a generated flow icon (not an avatar), the agent's name, a Category tag, and a short description.

The active tab is saved in the URL, so a link to the Multi agents tab reopens there.

Two things are deliberately different about multi agents in this list:

• No Live/Draft status. A flow has no live/draft lifecycle, so the Multi agents tab has no Status filter - it filters by Category and a sort only. (The Single agents tab keeps Status, Category, Updated By, and Last Updated.)
• A generated icon instead of an avatar, so you can tell the two types apart at a glance.

To create one, click Create in the page header and pick Multi agent from the "What would you like to create?" picker. (The same picker offers Single agent.) See Create an agent.

How multi agents become reachable

A multi agent is not a tool, and it isn't reached through the tools picker or an @-mention. It becomes reachable the same way a single agent does: as a delegation target.

Open Routing Logic and look at the Routing targets section. Alongside the single agents your bot can hand off to, there's a separate Multi agents group listing every multi agent in the bot. Single agents get All / Selected controls; multi agents additionally get an Off switch, since - unlike single agents - a multi agent can be fully disabled as a delegation target. Leave multi agents on All (the default) to let any single agent hand off to any multi agent, or switch to Selected to scope which ones are valid handoff targets - there's no per-guided-agent toggle elsewhere.

This is what makes the two types compose cleanly. A single agent handles the open-ended conversation, and when it reaches a part that needs a fixed procedure, it hands off to the relevant multi agent - the same delegation mechanism used between two single agents. The multi agent runs its steps to completion and returns control.

Note: A multi agent is not a workflow either. "Workflow" is a reserved term for a callable scheduler skill (an API call, a lookup, a job). A multi agent is a conversational journey across nodes - it can call a workflow (via an Execute Workflow node) and it can be delegated to like an agent, but it is never itself a tool or a workflow.

Tip - name multi agents by the task. order-return, kyc-verification, book-appointment read clearly both in the Guided list and in the Routing targets picker. A vague name like flow2 tells the next builder nothing about when to hand off to it.

Availability

Multi agents are gated. They appear only on Nexus bots that also have the flow-agent capability enabled. When the gate is off, the Agents page behaves exactly as before - no Multi agents tab, no Guided option in the Create picker.

Note: If you don't see the Multi agents tab on the Agents page, or the Create picker shows only Single agent, the capability isn't enabled for your account. Talk to your account team to turn it on.

Read next

• Create an agent - start from the Create picker and open the flow canvas.
• The flow canvas - the Editor / Execute / Logs tabs, nodes, and exit branches.
• The Agent node - instructions, exits, tools, variables, and first response.