Fix binding options not loaded from environment variables

[danielaskdd]

Fix binding options not loaded from environment variables

Summary

This PR fixes a critical bug where environment variables for LLM binding options (e.g., OPENAI_LLM_TEMPERATURE, OPENAI_LLM_MAX_TOKENS) were silently ignored, causing the server to always use default values.

Problem

When using OpenAI/Azure OpenAI bindings, environment variables like OPENAI_LLM_TEMPERATURE=0.7 had no effect. The startup log would show:

INFO: OpenAI LLM Options: {}

Root Cause: The _GlobalArgsProxy class did not properly support vars() calls. When binding_options.py called vars(args) on the proxy object, it returned an empty dict instead of the underlying argparse.Namespace attributes, causing all binding-specific options to be lost.

Changes

Commit 1: Refactor argument parsing (a000bdf0)

• Consolidated duplicate --llm-binding and --embedding-binding parsing logic
• Added edge case handling for CLI arguments starting with -
• Improved code maintainability by reducing duplication

Commit 2: Fix GlobalArgsProxy (a555690c)

• Replaced __getattr__ with __getattribute__ to intercept __dict__ access
• When vars(proxy) is called, it now correctly returns the underlying namespace's __dict__
• Enables binding_options.options_dict() to properly extract provider-specific options

Testing

Verified with:

from lightrag.api.config import global_args
from lightrag.llm.binding_options import OpenAILLMOptions

options = OpenAILLMOptions.options_dict(global_args)
print(options)  # {'temperature': 0.7, 'max_tokens': 9000}

Impact

• Before: All OpenAI/Gemini/Ollama LLM binding options were ignored
• After: Environment variables are properly loaded and applied to LLM calls

[danielaskdd]

@codex review

[chatgpt-codex-connector]

Codex Review: Didn't find any major issues. Chef's kiss.

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".