Fix ENTITY_TYPES Environment Variable Handling

[danielaskdd]

Fix ENTITY_TYPES Environment Variable Handling

🐛 Problem Description

The ENTITY_TYPES environment variable was incorrectly processed when provided as a JSON array string, causing entity extraction to fail completely. Instead of parsing the JSON string into a proper Python list, the system was converting it character-by-character, resulting in unusable single-character tokens being passed to the LLM.

Before Fix:

ENTITY_TYPES='["Product","Manufacturer","Brand"]'
# Result: ['[', '"', 'P', 'r', 'o', 'd', 'u', 'c', 't', '"', ',', ...]

Impact:

• Entity extraction completely broken
• LLM receives meaningless character arrays instead of entity type names
• Custom entity types cannot be used effectively

🔧 Solution

Modified the get_env_value function in lightrag/utils.py to properly handle JSON parsing for list-type environment variables.

Key Changes:

• Added JSON parsing logic specifically for list type environment variables
• Implemented type validation to ensure parsed values are actually lists
• Added comprehensive error handling with graceful fallback to defaults
• Maintained full backward compatibility for all other environment variable types
• Added informative logging for debugging invalid configurations

✅ After Fix

Now Works Correctly:

ENTITY_TYPES='["Product","Manufacturer","Brand","Store","Category","Service","Technology","Feature","Location","Price","Model"]'
# Result: ["Product","Manufacturer","Brand","Store","Category","Service","Technology","Feature","Location","Price","Model"]

🧪 Testing

• ✅ Valid JSON arrays are correctly parsed into Python lists
• ✅ Invalid JSON gracefully falls back to default entity types
• ✅ Non-list JSON objects fall back to defaults with appropriate warnings
• ✅ Empty/missing environment variables use default values
• ✅ Full backward compatibility maintained for string, int, and bool types
• ✅ Entity extraction prompts now receive proper entity type lists
• ✅ LLM can correctly classify entities using custom types

📋 Usage

Users can now set custom entity types using standard JSON array syntax:

export ENTITY_TYPES='["Product","Manufacturer","Brand","Store","Category","Service","Technology","Feature","Location","Price","Model"]'

The system will properly parse this into a Python list and use it in entity extraction prompts, allowing the LLM to correctly classify entities according to the specified custom types.

🔄 Backward Compatibility

This fix maintains 100% backward compatibility:

• All existing environment variable handling continues to work unchanged
• Default entity types are used when ENTITY_TYPES is not set
• Other list-type environment variables (if any) will benefit from the same JSON parsing logic