Implementation:Openai Openai python Lazy Proxy

Knowledge Sources	Openai_Openai_python
Domains	SDK_Infrastructure
Last Updated	2026-02-15 00:00 GMT

Overview

Concrete tool for lazy object proxy pattern provided by the openai-python SDK.

Description

The _proxy module defines the LazyProxy[T] abstract base class, a generic proxy that pretends to be an instance of type T by forwarding attribute access, string representation, and other dunder methods to the lazily loaded underlying object. Subclasses must implement the abstract __load__() method, which is called on first access via __get_proxied__(). The proxy overrides __getattr__, __repr__, __str__, __dir__, and the __class__ property to transparently delegate to the proxied object. Special handling ensures that proxies which themselves return proxies (for chained attribute access like proxy.foo.bar) do not cause infinite recursion. The __as_proxied__() helper returns the proxy itself typed as T for static type checking convenience.

Usage

Use this base class when implementing deferred initialization patterns, such as the module-level client proxies in _module_client or the deprecation proxies in _old_api.

Code Reference

Source Location

• Repository: openai-python
• File: src/openai/_utils/_proxy.py
• Lines: 1-65

Signature

T = TypeVar("T")

class LazyProxy(Generic[T], ABC):
    def __getattr__(self, attr: str) -> object: ...
    def __repr__(self) -> str: ...
    def __str__(self) -> str: ...
    def __dir__(self) -> Iterable[str]: ...
    @property
    def __class__(self) -> type: ...
    def __get_proxied__(self) -> T: ...
    def __as_proxied__(self) -> T: ...
    @abstractmethod
    def __load__(self) -> T: ...

Import

from openai._utils._proxy import LazyProxy
# or via the convenience re-export:
from openai._utils import LazyProxy

I/O Contract

Inputs

Name	Type	Required	Description
(none for base class)	N/A	N/A	Subclasses define __load__() to return the target object
attr	str	Yes	Attribute name forwarded via __getattr__

Outputs

Name	Type	Description
__load__() result	T	The lazily loaded target object
__as_proxied__() result	T	The proxy itself, cast to type T for type-checker convenience
__getattr__ result	object	Attribute value from the proxied object

Usage Examples

Basic Usage

from openai._utils import LazyProxy
from typing_extensions import override

class ConfigProxy(LazyProxy[dict]):
    @override
    def __load__(self) -> dict:
        # Expensive initialization deferred until first access
        return {"model": "gpt-4o", "temperature": 0.7}

config = ConfigProxy().__as_proxied__()
# No loading happens yet

print(config["model"])  # Triggers __load__(), prints "gpt-4o"

Related Pages

• Environment:Openai_Openai_python_Python_3_9_Plus