Visual canvas

The canvas is where you connect an agent to the resources it uses. The agent sits in the middle, and each tool, knowledge base, MCP server and sub-agent attached to it appears as a node linked to it.

Open the canvas

From the agent list, select the canvas icon (Open canvas) on an agent’s row.
From the agent form, select Open Canvas → in the Visual Canvas Editor banner.
A new agent whose only channel is Web Chatbot opens here as soon as it is created.

Back to form returns to the full agent form.

Attach resources

Find the resource in the sidebar on the left. It is grouped into Custom Tools, Predefined Tools, Knowledge Bases, MCP Servers and Sub-Agents, with a search box at the top.
Select Add to Canvas next to it. Items already attached show Already added.
The resource appears as a node linked to the agent. It is available to the agent from its next message.

You can also select Add resource on the agent node to open the same list as an overlay.

The resources themselves are created elsewhere: custom tools and MCP servers, predefined tools and knowledge bases.

Inspect and remove

Select any node to open its Properties panel.

A resource node shows what it is: a tool’s HTTP method and description, a knowledge base’s document count, an MCP server’s description, a sub-agent’s model.
Remove from agent detaches it. The resource itself is not deleted and stays available to other agents.

Edit the agent

Select the agent node. Its Properties panel has two tabs.

Configuration holds the agent’s Name, Description and System Prompt, its AI Model (Provider, Model, and a Temperature slider from Precise to Creative), Max Output Tokens, and the lists of attached resources, each with a remove button.

Widget holds what you need to put the agent’s web chat on a page: the Chat URL, an Embed snippet (an iframe) to copy, and an Avatar URL shown in the chat widget header.

Select Save Agent to keep your changes.

Sub-agents

A sub-agent is another agent in your workspace that this agent can hand a task to. The parent decides when to delegate from the sub-agent’s name and description, sends it the task, and uses what comes back. Use sub-agents to keep specialist instructions and tools out of a general agent’s prompt.

To attach an existing agent, add it from Sub-Agents in the sidebar. To create one, select New next to Sub-Agents (or Create new sub-agent in the overlay) and fill in:

Field	What it does
Agent ID	Required. Letters, numbers, `_` and `-` only.
Name	Required.
Description	What the sub-agent specializes in. The parent reads it to decide when to delegate.
Personality / System Prompt	Required. At least 10 characters.
AI Model	Provider and Model.

The new sub-agent is created as a regular agent, so it also appears in your agent list, and it is linked to the current agent. Delegation can go at most three levels deep.

Test the agent

Select Test Agent at the top right to open a chat panel beside the canvas. Send messages as a user would, and toggle the execution trace to see each model decision and tool call behind a reply. The reset button clears the chat and reloads the agent with your latest changes.