Structured Output

Structured Output is a feature that returns AI output in a format conforming to a specified JSON schema. It is useful when you want to use data in subsequent processing or iterate over data with a Loop node.

Supported Nodes

Structured output can be used with the following nodes.

Node	Notes
LLM Node	Available with all models
Agent Node	Available with all agents except Deep Research

Enabling Structured Output

Turn on the switch in the “Structured Output” section of the node settings panel to enable structured output.

After turning on the switch, click the “Edit” icon to open the schema editor dialog.

How to Configure a Schema

The schema editor dialog offers three ways to create and edit schemas.

Visual Editor

A GUI-based editor for adding and editing properties. Build your schema by entering property names, types, and descriptions.

• Click the ”+” button to add a property
• Select a type (string / number / boolean / array / object)
• For object and array types, you can define nested properties

JSON Editor

An editor for directly editing JSON Schema. It includes syntax highlighting and validation, so you can quickly catch any errors.

Generate with AI

Not sure where to start?

AI generation is recommended. Simply describe the output you want in natural language, and the schema will be automatically generated.

Just tell the AI what kind of output you want in a chat format, and it will automatically generate a schema. You can use the generated schema as-is, or fine-tune it with the Visual Editor or JSON Editor.

Caution

Generating schemas with AI consumes credits.

Note

All three editors edit the same schema. Changes made in any editor are reflected in the others.

Supported Types

The following types can be specified for schema properties.

Type	Description	Example
string	Text string	"Tokyo"
number	Numeric value	42, 3.14
boolean	Boolean value	true, false
array	Array (list)	["A", "B", "C"]
object	Object (nested structure)	{"name": "Taro", "age": 30}

Using Structured Output in Subsequent Nodes

Each field of a node configured with structured output can be referenced as a variable in subsequent nodes.

Type / in the prompt input field of a subsequent node to open the variable picker, which lists the fields defined under the previous node’s structured_output.

Examples of using structured output

Structured output variables can be used as input for various nodes.

• Loop Node — Array-type fields can be used directly as iteration data. See Loop Node for details.
• If-Else Node — Build conditional expressions based on field values to branch workflow execution. See If-Else Node for details.
• Intent Recognition Node — Specify field values as the target for evaluation, and let AI automatically select the branch based on the content. See Intent Recognition Node for details.

Best Practices

For better output

• Write prompts with JSON output in mind: Structured output returns the AI’s final output in JSON format. Therefore, your prompt must be written to produce output that matches the schema. For example, if your schema has a candidate_list property, include an instruction like “Output a list of candidates” in your prompt.
• Use descriptions: Setting a description for each property helps the model understand the intent of each property, improving output accuracy.