Auto-fixing malformed output in LangChain

Introduction

Sometimes, the output from a language model can be messy or broken. Auto-fixing helps clean and correct this output automatically so it is easier to use.

When a language model returns incomplete or broken JSON data.

If you want to avoid manual fixing of output errors in your app.

When you need reliable, clean data from AI responses for further processing.

To improve user experience by showing well-formed results.

When integrating AI output into systems that require strict formats.

Syntax

LangChain

from langchain.output_parsers import AutoFixParser

parser = AutoFixParser.from_llm(llm)
fixed_output = parser.parse(malformed_output)

Use AutoFixParser.from_llm() to create a parser linked to your language model.

Call parse() with the raw output to get a cleaned, fixed version.

Examples

This example shows fixing a JSON string missing a closing brace.

LangChain

from langchain.output_parsers import AutoFixParser

parser = AutoFixParser.from_llm(llm)
raw_output = '{"name": "John", "age": 30'  # missing closing brace
fixed = parser.parse(raw_output)
print(fixed)

Here, the parser tries to convert a plain text into a proper JSON format.

LangChain

parser = AutoFixParser.from_llm(llm)
malformed = 'Name: John Age: 30'  # not JSON
fixed = parser.parse(malformed)
print(fixed)

Sample Program

This program uses LangChain's AutoFixParser to fix a JSON string missing a closing brace. It prints the corrected JSON object.

LangChain

from langchain.chat_models import ChatOpenAI
from langchain.output_parsers import AutoFixParser

llm = ChatOpenAI(temperature=0)
parser = AutoFixParser.from_llm(llm)

malformed_output = '{"city": "Paris", "temperature": 20'  # missing closing brace
fixed_output = parser.parse(malformed_output)
print(fixed_output)

Important Notes

AutoFixParser relies on the language model to guess the fix, so results depend on model quality.

It works best with common formats like JSON or simple key-value pairs.

Always validate the fixed output before using it in critical systems.

Summary

Auto-fixing cleans broken AI outputs automatically.

It saves time by avoiding manual corrections.

Use it when you expect messy or incomplete AI responses.