description: Parse the output of an LLM call into a given (JSON) structure.

Structured Output Parser

The Structured Output Parser is a node used to parse the output of a Language Model (LLM) call into a predefined JSON structure. This node is particularly useful when you need to extract structured information from the LLM's response in a consistent format.

It is ideal for scenarios where the output structure is relatively flat and simple.

Key Features

• Basic Schema Support: Allows users to define expected output fields and their types, usually via JSON schema or example-based schema generation.

• Simple Validation: Checks if the output matches the expected structure, but with less sophistication than advanced parsers.

• Ease of Use: Quick to set up for common use cases where outputs are not deeply nested or highly complex.

• Intended for Final Output: Best used for structuring the last step in a workflow, not for intermediary formatting.

Parameters

Inputs

1. Autofix (Optional)

   • Type: boolean

   • Description: Enables automatic error correction by making additional model calls when parsing fails

   • Default: false

2. JSON Structure (Required)

   • Type: datagrid

   • Description: Defines the expected JSON structure for model outputs

   • Schema:

     • Property: JSON property name

     • Type: Data type (string, number, boolean)

     • Description: Property description

   • Default:

       [
         {
           "property": "answer",
           "type": "string",
           "description": "answer to the user's question"
         },
         {
           "property": "source",
           "type": "string",
           "description": "sources used to answer the question, should be websites"
         }
       ]
   
   

Input

The node takes the following inputs:

• Configuration for autofix (boolean)

• JSON structure definition (datagrid)

Output

The node outputs a StructuredOutputParser instance that can be used to parse LLM responses according to the defined JSON structure.

Flow of the Structured Output Parser Node

1. Schema Definition

   • The user defines the expected output schema, usually as a JSON Schema or using libraries like Zod (TypeScript) or Pydantic (Python).

   • The schema specifies required fields, types, and descriptions for each output property.

2. LLM Call

   • The language model generates a response, typically as unstructured text or loosely structured JSON.

3. Parsing Step

   • The Structured Output Parser node receives the LLM output.

   • It attempts to parse the output according to the defined schema.

   • If the output matches the schema, it is transformed into a structured object.

   • If not, the parser may raise an error or attempt to fix the output (depending on configuration).

4. Output Delivery

   • The parsed, structured data is passed on to the next node in the workflow or consumed by downstream systems.

{% hint style="info" %} This section is a work in progress. We appreciate any help you can provide in completing this section. Please check our Contribution Guide to get started. {% endhint %}