Skip to main content
Choose a multi-prompt agent when a call has strict steps or branches, such as a keypad menu, capture or an API lookup before the next question. For most calls, a single-prompt agent with one sectioned prompt is simpler to build and maintain. In the dashboard: Sidebar → Assistants → Create agent → Multi Prompt. The dashboard calls these agents Multi Prompt and lists them under the Multi-node filter on the Agents page (multi-node is the internal name). Multi-prompt agents support voice and web calls, not WhatsApp: the Multi Prompt card is not shown on the WhatsApp tab, and Convert to multi-prompt is not offered for WhatsApp agents.

​
Before you start

A knowledge base

Optional: for Agent nodes that answer from your documents.

Tools

Optional: on-call tools for Agent nodes, pre-call and post-call tools for the agent.

​
Create a multi-prompt agent

1

Open the create dialog

On the Agents page (listed under Assistants in the sidebar), click Create agent. The Create Assistant dialog opens.
2

Pick Multi Prompt

On the Outbound, Inbound or Webcall tab, under Create from scratch, click Multi Prompt (“Create assistant from scratch for complex workflows.”).
3

Start building

There is no form. The dashboard creates an agent named Untitled Workflow with a Start node and the custom variables callee_name and mobile_number, then opens the canvas. Whichever tab you start from, the new agent is an outbound agent.
You see “Workflow created successfully” and the canvas with a single Start node.
You can also turn an existing single-prompt agent into a multi-prompt one. See Convert a single-prompt agent.

​
The canvas

The editor has two parts. The canvas on the left holds the nodes and their connections. The panel on the right shows the settings of the selected node, or Global Settings when nothing is selected; drag the divider to resize it. The video adds a node, opens a node’s settings, then opens Global Settings.

Working on the multi-prompt canvas

  • Add a node: click the dot on a node’s output. A picker lists the node types you can add there (connection rules).
  • Drag to add: drag from an output dot to empty canvas. A preview of the next node follows the cursor; press Tab to switch its type, then release to place it. Drag onto an existing node to connect the two.
  • Reuse an Agent node: once the flow has an Agent node, Agent opens a submenu with Blank agent and From existing (copies another Agent node’s setup).
  • Name a node: click the name or the description at the top of its panel and type. Enter saves the name (Cmd/Ctrl+Enter for the description), Esc cancels. The name is what you see in the test summary, in the call’s path in Logs and from the MCP server, so name nodes by what they do: “Greet and confirm identity”, “Eligible?”.
  • Canvas controls: Zoom in, Zoom out, Fit to screen, Organize nodes (auto-layout left to right) and Hand tool (H) for panning.
  • Delete: each node has a delete action with a confirmation. To delete several, drag a box around them and press Delete or Backspace; Delete selection asks you to confirm. The Start node cannot be deleted.
Node picker listing Agent, Speak, Router, Action, Keypad Menu, Collect, Widget (greyed out) and End, each with a one-line description, with the Router submenu offering LLM Router and Logic Router

The node picker, opened from a router branch

Most node panels have their own Save button, which reads Saved once the node has no pending edits. Saved node changes go into the agent’s draft; they reach callers only when you publish.
Settings panel for the Agent node Greet and confirm identity, with Objective, Response Guidelines, Intro Message, Conversation flow and FAQs fields filled in and a Save button

An Agent node's settings with the Prompt section open

​
Example flow

A rescheduling call, built from the node types below. Edge labels are the router’s conditions and the nodes’ outputs.

​
Troubleshooting

The last publish found problems in the flow. Click Review issues and open each issue to jump to the node or connection that needs fixing, then publish again.

​
Next steps

Publish changes and run A/B tests

Drafts, publishing, versions and traffic split.

Add tools to your agent

Pre-call, on-call and post-call tools.