Example for OpenAI or any other LLM through the yaml config way

[shivsant]

Hi All,
I am trying out the yaml config approach - simplifies a lot of things. How can I create the agent for any other LLM - openai or others through litellm in this?

If someone has some example, can you please point me to that.

I have looked at https://github.com/google/adk-python/tree/main/contributing/samples/tool_agent_tool_config and other examples, but all are based on gemini models.

[adk-bot]

Response from ADK Answering Agent (experimental, answer may be inaccurate)

TLDR: Yes, you can use other LLMs like OpenAI with ADK by using the LiteLLM library. You define the agent in a Python file, specifying the OpenAI model via LiteLlm, and then reference this Python object in your YAML configuration using the code key.

---

Hello! That's a great question. The Agent Development Kit (ADK) is designed to be flexible and supports a wide range of Language Models beyond Gemini, including those from OpenAI, Anthropic, Cohere, and many others through its integration with the LiteLLM library. [1, 3]

While the examples you've seen might be centered on YAML for configuration, the standard way to incorporate other LLMs is to define the agent in a Python file and then reference that agent from your YAML configuration.

Here’s a step-by-step guide on how you can achieve this:

Step 1: Install LiteLLM

First, you need to install the litellm library alongside google-adk:

pip install litellm

Step 2: Set Your API Key

ADK, via LiteLLM, uses environment variables to authenticate with LLM providers. For OpenAI, you would set the OPENAI_API_KEY. [1]

export OPENAI_API_KEY="YOUR_OPENAI_API_KEY"

Step 3: Define Your Agent in Python

Create a Python file (e.g., my_agents/openai_agent.py) and define your LlmAgent there. Instead of a Gemini model string, you'll instantiate the LiteLlm class, passing the LiteLLM model string (e.g., "openai/gpt-4o").

Here is an example of what that Python file would look like. [1]

# File: my_agents/openai_agent.py

from google.adk.agents import LlmAgent
from google.adk.models.lite_llm import LiteLlm

# Define an agent that uses OpenAI's GPT-4o model
my_openai_agent = LlmAgent(
    model=LiteLlm(model="openai/gpt-4o"), # Using LiteLlm to specify the model
    name="openai_agent",
    instruction="You are a helpful assistant powered by OpenAI's GPT-4o.",
    # You can add tools and other parameters here as usual
    # tools=[...]
)

Step 4: Reference the Agent in Your YAML Config

Now, in your main YAML configuration file, you can reference the agent you defined in Python using the code key. This key points to the Python object.

For example, if you have a root agent in YAML that you want to use your new OpenAI agent as a sub-agent, it would look like this: [14]

# File: your_main_config.yaml

name: "root_agent"
model: "gemini-2.0-flash"
instruction: "You are a coordinator agent."
sub_agents:
  - code: my_agents.openai_agent.my_openai_agent

This setup allows you to leverage the full power of Python for complex configurations (like using different model libraries) while still using YAML for the overall structure of your agent team.

For more detailed information, you can refer to the ADK documentation on using different models and the tutorials on building agent teams. [1, 3]

I hope this helps you get started! Let us know if you have any more questions.

[1] https://google.github.io/adk-docs/agents/models/
[3] https://google.github.io/adk-docs/tutorials/agent-team/
[14] https://github.com/google/adk-python/blob/main/src/google/adk/agents/common_configs.py

[shivsant]

@adk-bot I need some improvement on this.

You said the yaml should look like

File: your_main_config.yaml

name: "root_agent"
model: "gemini-2.0-flash"
instruction: "You are a coordinator agent."
sub_agents:

• code: my_agents.openai_agent.my_openai_agent

here you have used gemini models but if it's going to use openai models, it should be using litellm class. Can you show the example how to yse the litellm class in this model key in yaml - that's my main confusion.

[xuanyang15]

@wuliang229 Could you please help with this?

[wuliang229]

Hi @shivsant, there isn't any support for other non-Gemini models in the yaml config design yet. The feature is experimental, so we plan to gradually add more features to it. Right now non-Gemini models are planned.