Add http client property in serialized config

[zhangshan123]

What happened?

Describe the bug When using company's customized model, if we do not pass the http_client parameter, the endpoint will reject our access

To Reproduce Using the simple code: def load_config(file_path: str = os.path.join(os.path.dirname(file), "model_config.yaml")): with open(file_path, "r") as file: if file_path.endswith('.json'): config_data = json.load(file) elif file_path.endswith('.yaml'): config_data = yaml.safe_load(file) if len(config_data['config'].get("api_key", "")) == 0: config_data['config']['api_key'] = os.getenv("AZURE_API_KEY") if config_data['config']['model'].startswith("gemini-"): config_data['config']["http_client"] = httpx.AsyncClient(verify=False) return config_data

model_config = load_config() model_client = OpenAIChatCompletionClient(**model_config['config'])

Expected behavior We can pass http_client property to AsyncOpenAI(http_client=xxx)

Screenshots N/A

Additional context Add http_client to BaseOpenAIClientConfiguration,such as: class BaseOpenAIClientConfiguration(CreateArguments, total=False): model: str api_key: str timeout: Union[float, None] ...... http_client: AsyncClient | None

Which packages was the bug in?

Python Extensions (autogen-ext)

AutoGen library version.

Python 0.7.5

Other library version.

No response

Model used

gemini-2.5-pro

Model provider

Azure OpenAI

Other model provider

No response

Python version

3.12

.NET version

None

Operating system

None

[priteesshhh]

I analyzed this issue but couldn't identify the affected files. Could you please provide more details about which files need to be modified?

[priteesshhh]

I analyzed this issue but couldn't identify the affected files. Could you please provide more details about which files need to be modified?

[zhangshan123]

I analyzed this issue but couldn't identify the affected files. Could you please provide more details about which files need to be modified? @priteesshhh

file path: https://github.com/microsoft/autogen/blob/main/python/packages/autogen-ext/src/autogen_ext/models/openai/config/init.py

#Add the http_client property to BaseOpenAIClientConfiguration, such as: class BaseOpenAIClientConfiguration(CreateArguments, total=False): model: str api_key: str timeout: Union[float, None] max_retries: int model_capabilities: ModelCapabilities # type: ignore model_info: ModelInfo add_name_prefixes: bool """What functionality the model supports, determined by default from model name but is overriden if value passed.""" include_name_in_message: bool """Whether to include the 'name' field in user message parameters. Defaults to True. Set to False for providers that don't support the 'name' field.""" default_headers: Dict[str, str] | None http_client: AsyncClient | None

[Shizoqua]

Hi @zhangshan123 i'd like to contribute to this issue. I understand this involves adding an HTTP client property to the serialized configuration. I've also found that model clients use dump_component() for serialization (e.g. in _openai_client.py).

My plan is to:

1. Investigate how HTTP clients are currently configured in model clients (OpenAI, Anthropic, etc.)
2. Identify where the HTTP client property should be added to the serialized config
3. Ensure the HTTP client is properly serialized/deserialized when using dump_component() and load_component()
4. Add appropriate tests to verify serialization works correctly

Could you provide a bit more context on:

1. Which specific HTTP client property needs to be serialized? (e.g., timeout, connection pool settings, custom headers?) 2.Are there any specific model clients this should apply to or should it be a general feature?

I'm ready to start once I have clarification. Thanks