Variable Extractor Node

Definition

This Variable Extractor node utilizes a "large language model" to analyze and interpret natural language input, extracting the user-defined variable parameters in a structured format.

How to configure

1. Model

From the model dropdown, select the large language model you wish to use. You can adjust the "temperature" of the language model, which dictate the degree of divergence (or randomness) of the model. Higher values lead to a more creative and varied response, while lower value generates a more focused output.

Detailed explanation is as follows:

2. Input Variable

This refers to the source of the natural language data input, which can be either:

• The default {x}Query from the start node in the workflow
• The output variable from the previous node

3. Extraction Parameters

Click "+" to add and configure "extraction parameters." Each parameter requires you to fill in 3 variables: "extraction prompt," "name," and "type."

• Extraction Prompt: Provide a clear description of the content to be extracted, with a specific name (e.g., Extract the phone number).
• Name: Define a custom name or identifier for the extracted data (e.g., customer-phone).
• Type: Mention the data type of the output variable (e.g., string, Array[String]).

An example of Input text is: "My phone number is 123456, email is xxxxxx@airdroid.com, and order number is 987654." Here:

4. Example

Example Content is optional extraction guidance that will be added to the prompt of the Variable Extractor. It can be used to explain how variable values should be selected, ignored, normalized, or disambiguated. It can include examples, but it is not limited to examples only.

It is usually used in the following scenarios:

• The input text does not follow a fixed expression format.
• Multiple fields need to be extracted at once.
• Fields can be easily confused with each other.
• Rules need to be specified for missing fields, multiple values, date formats, enum values, and similar cases.

Example Content can be written as natural language rules, or in the format of “sample input + expected output”. The following format is recommended:

Extract the following fields from the input text:
- Order_Number: The order number, usually starting with DDH
- Email: The user's email address
- Phone: The contact phone number
- Delivery_Date: The expected delivery date

Sample input:
Hello, I would like to check order DDH12345678. My email is xxxxxx@example.com, my phone number is 12345678, and I hope it can be delivered on May 1.

Expected output:
{
  "Order_Number": "DDH12345678",
  "Email": "xxxxxx@example.com",
  "Phone": "12345678",
  "Delivery_Date": "2026-05-01"
}

Rules:
If a field does not appear in the input, return an empty value.
If the same field appears multiple times, return a string or an array based on the field type.

If only one field needs to be extracted, Example can contain just one simple rule or sample. If multiple fields need to be extracted, it is recommended to write Example as a complete sample input and provide the expected output for all fields.

Use Case

Sample input:

Hello, I would like to check order DDH12345678. My email is xxxxxx@example.com, and my phone number is 12345678.

Extraction parameters:
- Extraction prompt: Order number; Name: Order_Number; Type: String
- Extraction prompt: Email; Name: Email; Type: String
- Extraction prompt: Phone number; Name: Phone; Type: String

Example:
Please extract the order number, email address, and phone number from the user input.
The order number usually starts with DDH.
If a field is not mentioned, return an empty value for that field.

The following result is output, successfully converting the user's natural language into structured parameters that can be used by subsequent workflow nodes.

{
  "Outputs": [
    {
      "Name": "$AllFound",
      "Type": "bool",
      "Value": true
    },
    {
      "Name": "Order_Number",
      "Type": "string",
      "Value": "DDH20250423"
    },
    {
      "Name": "Email",
      "Type": "string",
      "Value": "xxxxxx@example.com"
    },
    {
      "Name": "Phone",
      "Type": "string",
      "Value": "12345678"
    }
  ]
}