Documentation

Creating Agents

Visual Builder

Build complex agent workflows visually with drag-and-drop nodes, conditional logic, and real-time testing. No coding required.

Intermediate15-30 minutes

Overview

The Visual Builder at Create → Visual Builder provides a canvas-based workflow editor. You build agent logic by:

  1. Adding nodes to the canvas
  2. Connecting nodes to define the flow
  3. Configuring each node's settings
  4. Testing the workflow
  5. Deploying the agent

Node Types

Five types of nodes are available, each with a distinct color and purpose:

Trigger Nodes

Triggers define when your agent activates. Every workflow starts with at least one trigger:

  • User Message — Activates when a user sends a message (most common)
  • Scheduled — Activates on a time-based schedule
  • Webhook — Activates when an external service calls the agent's webhook URL

Action Nodes

Actions perform work using the agent's tools:

  • Web Search — Searches the web for information → web_search
  • Send Email — Sends an email → email
  • Save Data — Renders and stores output → template_render
  • Run Code — Executes JavaScript → code_execute

Additional tools are automatically mapped based on node names. For example, adding a "Scrape" node maps to web_scrape, a "Chart" node maps to generate_chart, and so on. See the Tools Reference for all available tools.

Condition Nodes

Conditions add branching logic. A condition node has two outputs:

  • "Yes" port — Follows this path when the condition is true
  • "No" port — Follows this path when the condition is false

Example: Check if the user's question is about finance. If yes, route to a financial analysis action. If no, route to a general response.

LLM Nodes

LLM nodes represent an AI response step. Configure:

  • The prompt/system instructions for this specific step
  • Temperature and response length

Output Nodes

Output nodes define what the user receives:

  • Respond — Sends a message back to the user
  • Return Data — Returns structured data as output

Building a Workflow

Adding Nodes

Open the Node Library panel on the left side of the canvas. Click or drag a node template to add it to the canvas. Nodes appear at the cursor position.

Connecting Nodes

Click a node's output port and drag to another node's input to create a connection. Connections define the order of execution:

  • Standard connections have a "default" port
  • Condition nodes have "yes" and "no" ports

Configuring Nodes

Click a node to select it and open the properties panel. Each node type has specific configuration options:

  • Triggers — Schedule, webhook URL, etc.
  • Actions — Tool-specific parameters
  • Conditions — The condition expression to evaluate
  • LLM — Prompt text, temperature
  • Outputs — Response format

Deleting Nodes

Select a node and press Delete or use the context menu. Associated connections are automatically removed.

Default Workflow

New sessions start with a basic three-node workflow:

User Message→AI Response→Respond

Add action and condition nodes between the LLM and output nodes to create more complex flows.

Testing

Click Test Workflow to simulate execution. The test runner reports:

  • Number of nodes executed
  • Number of connections traversed
  • Any errors encountered

Deploying

Name your agent, then click Deploy. The system:

  1. Generates a system prompt from the workflow structure
  2. Maps node types to SDK tools automatically
  3. Creates the agent with type "visual"

See the full Publishing guide for post-deployment steps.

Example: Research Agent Workflow

Here's how you might build a research agent:

1.User Message (trigger)
2.Web Search (action)
3.Web Scrape (action)
4.Has data? (condition)Yes →
5.AI Analysis (LLM)
6.Respond (output)

Next Steps