Skip to main content

​
What the canvas is

When you open a bot and go to the Agents tab, you’ll find a visual canvas where each node represents an agent — a conversation unit with its own identity, instructions, and tools. Agents connect to each other to form the call’s complete flow: the bot can move from one agent to another based on what the person says, which conditions are met, or which actions run.
When does it make sense to split the flow across several agents instead of one? See Design principles.
Agent canvas with connected nodes

​
Agent types

Agents are distinguished by their color on the canvas, which indicates their role in the flow:
The Post-Conversation agent is an agent like any other — it has Identity, Tools, and Logic —, it just triggers automatically when the call ends, and can’t be disabled. Besides assigning the label, its tool supports additional parameters (for example, a conversation summary or other custom variables) that, with the help of Python code, become available to users in the platform’s operational call view.

​
Connecting agents

Connections between agents define when and where the conversation can “jump.” Each connection is configured from the Agent Handoff tab within the source agent.
An agent can have multiple handoff rules to different destinations. The LLM evaluates the conditions and picks the right agent based on the conversation’s context.
You can also start a connection directly by dragging on the canvas, without going into the agent’s tab. See Connecting agents by dragging on the Canvas.

​
Show the agents panel

With bots that have many agents, the canvas can get hard to scan at a glance. The ☰ icon in the canvas controls group (next to Zoom In, Zoom Out, and Fit View) opens a panel with the bot’s full agent list, their type, and a search box to find one by name. Agents panel open with the full list and search box Click any agent in the list to open its editor, same as clicking its node on the canvas.

​
Exploring an agent

Click any node on the canvas to open the agent editor. Inside you’ll find four tabs:

Identity

The agent’s initial message and instructions.

Tools

Functions the agent can execute.

Logic

Python code to prepare the context.

Agent Handoff

Handoff rules to other agents.