OpenAI

OpenAI's API and models are fully supported in AG2. You will need an OpenAI account and to create an API key. See their website for further details.

Installation#

Warning

From version 0.8: The OpenAI package, openai, is not installed by default, install it with the openai extra as shown below.

pip install ag2[openai]

Tip

If you have been using autogen or ag2, all you need to do is upgrade it using:

pip install -U autogen[openai]

or

pip install -U ag2[openai]

as autogen and ag2 are aliases for the same PyPI package.

LLM Configurations#

See the LLM Configuration for further details on setting an creating an LLM configuration.

The below example of an OAI_CONFIG_LIST shows how the OpenAI client class is used. By specifying the api_type as openai, the client will use the OpenAI ChatCompletion API.

[
    {
        "api_type": "openai",
        "model": "o1",
        "api_key": "your OpenAI Key goes here",
    },
    {
        "api_type": "openai", # use OpenAI ChatCompletion API
        "model": "gpt-5-nano",
        "api_key": "your OpenAI Key goes here",
    }
]

As an alternative to the api_key key and value in the config, you can set the environment variable OPENAI_API_KEY to your OpenAI key.

Linux/Mac:

export OPENAI_API_KEY="your_openai_api_key_here"

Windows:

set OPENAI_API_KEY=your_openai_api_key_here

Tip

Interested in trying out image generation with the new OpenAI Responses API? Check out the OpenAPI Responses client.

Extra Headers#

You can pass custom HTTP headers on each API request using the extra_headers parameter. This is useful when working through API gateways, proxies, or OpenAI-compatible servers that require additional headers for authentication, routing, or observability.

from autogen import LLMConfig

llm_config = LLMConfig(
    api_type="openai",
    model="gpt-4o-mini",
    api_key="your OpenAI Key goes here",
    extra_headers={
        "X-Custom-Header": "my-value",
    },
)

Note

extra_headers is a per-request parameter passed on each call. This is different from default_headers, which is set once on the client constructor and applies to all requests made by that client.

Examples#

See the ConversableAgent example in our Basic Concepts documentation to see OpenAI's LLMs in use.