How to Build a Custom Output Parser in LangChain for Non‑Standard LLM Formats

LangChain ships with many built‑in output parsers, but some LLMs produce highly specific or legacy formats (e.g., custom XML or a bespoke "Key: Value" text) that the defaults cannot handle. In such cases you create a custom output parser that extends the library’s core parsing capabilities.

Why a custom parser?

Handle non‑standard formats : fine‑tuned models may output data that is not JSON or CSV.

Increase robustness : embed custom error handling, retries, or correction logic directly in the parser.

Encapsulate complex logic : keep data‑transformation and validation code reusable and isolated.

How to create a custom output parser

Creating a parser is straightforward and consists of three main steps:

Inherit BaseOutputParser : your class must subclass langchain_core.output_parsers.BaseOutputParser.

Implement parse() : this method receives the raw LLM string and returns any structured object you need (a dict, a custom class, etc.). All parsing logic lives here.

(Optional) Implement get_format_instructions() : returning a clear instruction string helps the LLM format its output so that parse() can succeed.

Below is a minimal template that follows these steps:

from langchain_core.output_parsers import BaseOutputParser
from typing import TypeVar

T = TypeVar("T")  # for type hinting

class MyCustomParser(BaseOutputParser[T]):
    """Template for a custom parser."""

    def parse(self, text: str) -> T:
        """Parse the LLM output string into a structured object."""
        # Insert your parsing logic here
        # ...
        return your_parsed_object

    def get_format_instructions(self) -> str:
        """Return formatting instructions for the LLM."""
        # Provide clear guidance on the expected output format
        # ...
        return "Please return your answer in the xxx format."

    @property
    def _type(self) -> str:
        """Unique name for the parser type."""
        return "my_custom_parser"

In the accompanying example the parser is designed to read a simple Key: Value string and convert it into a Python dict. This demonstrates the full implementation flow from inheritance to returning a concrete data structure.

References

How to: write a custom output parser class – https://python.langchain.com/docs/how_to/custom_output_parser