json-healer-lite

A zero-dependency Python micro-tool to instantly repair and parse malformed, truncated, or markdown-wrapped JSON strings generated by LLMs.

The Problem

When using Large Language Models (LLMs) to generate JSON, you frequently encounter three issues that crash json.loads():

1. Markdown Wrappers: The LLM wraps the output in ```json ... ```.
2. Token Limits / Streaming: The response is cut off mid-generation, leaving unclosed brackets [{ or unclosed strings.
3. Trailing Commas: The LLM leaves a trailing comma like {"a": 1, }.

The Solution

json-healer-lite is a pure standard library tool that acts as a robust middle layer. It cleans the string, balances the stack of missing brackets, and safely parses the JSON—saving your API pipelines from crashing.

Installation

pip install json-healer-lite

Usage

from json_healer_lite import heal_json, JSONHealerError

# Typical LLM output that got cut off due to max_tokens:
broken_llm_output = """
{
  "status": "success",
  "data": [
    {"name": "Alice", "score": 95},
    {"name": "Bob", "score": 82,
"""

try:
    # Heals markdown, trailing commas, and auto-closes the string, dict, and array!
    healed_data = heal_json(broken_llm_output)
    
    print(healed_data)
    # Output: {'status': 'success', 'data': [{'name': 'Alice', 'score': 95}, {'name': 'Bob', 'score': 82}]}

except JSONHealerError as e:
    print(f"JSON was too corrupted to save: {e}")

Why zero-dependency

Frameworks like LangChain provide output parsers, but pulling in a massive dependency tree just to fix a few missing brackets adds latency and bloat (especially in serverless environments like AWS Lambda). This micro-tool operates in under 1ms and weighs just a few kilobytes.