Data Extraction: The JsonExtractor Node

Data Extraction: The JsonExtractor Node

The JsonExtractor node is a utility for extracting a specific field from a JSON string. It takes a JSON object (or a string containing JSON, optionally wrapped in markdown code blocks), resolves any workflow variables, and returns the value of the specified field as a string. This is useful for parsing structured LLM outputs or extracting data from API responses within a workflow.

How It Works

1. Configuration Loading: The processor loads the jsonToExtractFrom and fieldToExtract fields from the node's configuration.
2. Variable Substitution: Both fields are processed with the WorkflowVariableManager, allowing you to use variables like {agent1Output} or {agent1Input} to reference JSON strings from other nodes.
3. Markdown Stripping: If the JSON string is wrapped in markdown code blocks (```json or ```), the node automatically strips this formatting before parsing.
4. JSON Parsing: The cleaned string is parsed as JSON. If parsing fails or the result is not a JSON object, the node returns an empty string.
5. Field Extraction: The specified field is extracted from the parsed JSON object.
6. Type Conversion: The extracted value is converted to a string:
   • Strings are returned as-is
   • Numbers and booleans are converted to their string representation
   • Nested objects and arrays are returned as JSON strings
   • null values return an empty string

Properties

Variable Usage

Both jsonToExtractFrom and fieldToExtract support all available workflow variables. This allows you to:

• Reference JSON output from a previous node (e.g., {agent1Output})
• Pass JSON strings from a parent workflow (e.g., {agent1Input})
• Dynamically specify which field to extract based on workflow state

Handling Markdown Code Blocks

LLMs often wrap JSON output in markdown code blocks for formatting. This node automatically handles the following formats:

```json
{"name": "value"}

{"name": "value"}

The node will strip these code block markers before parsing the JSON.

-----

#### **Full Syntax Example**

This example extracts the "name" field from a JSON object returned by a previous LLM node.

```json
{
  "title": "Extract Character Name from LLM Response",
  "type": "JsonExtractor",
  "jsonToExtractFrom": "{agent1Output}",
  "fieldToExtract": "name"
}

Example Input (from agent1Output):

{
  "name": "Socg",
  "file": "Socg.txt",
  "description": "A mysterious character"
}

Example Output: Socg

Example with Nested Objects

When extracting a nested object or array, the node returns it as a JSON string.

{
  "title": "Extract Character Details",
  "type": "JsonExtractor",
  "jsonToExtractFrom": "{agent1Output}",
  "fieldToExtract": "metadata"
}

Example Input:

{
  "name": "Socg",
  "metadata": {"age": 25, "traits": ["brave", "clever"]}
}

Example Output: {"age": 25, "traits": ["brave", "clever"]}

Example with Dynamic Field Name

You can use variables to dynamically determine which field to extract.

{
  "title": "Extract Dynamic Field",
  "type": "JsonExtractor",
  "jsonToExtractFrom": "{agent1Output}",
  "fieldToExtract": "{agent2Output}"
}

If agent2Output contains "description", this will extract the "description" field from the JSON.

Behavior and Edge Cases

• Missing Field: If the specified field does not exist in the JSON object, the node returns an empty string.
• Invalid JSON: If the input cannot be parsed as JSON, the node returns an empty string and logs a warning.
• Non-Object JSON: If the JSON is valid but is an array or primitive (not an object), the node returns an empty string. Only JSON objects with key-value pairs are supported.
• Null Values: If the field exists but its value is null, the node returns an empty string.
• Empty String Values: If the field exists but its value is an empty string, the node returns an empty string.
• Output Variable: The extracted string is assigned to a variable based on the node's position in the workflow (e.g., {agent1Output}, {agent2Output}, etc.).