Google ADK

Google's Agent Development Kit (ADK) supports custom OpenAI-compatible endpoints via its LiteLLM integration.

Install

pip install google-adk litellm

Configure

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

agent = Agent(
    model=LiteLlm(
        model="openai/{{selectedModel.id}}",
        api_base="https://api.doubleword.ai/v1",
        api_key="{{apiKey}}",
    ),
    name="my_agent",
    instruction="You are a helpful assistant.",
)

The openai/ prefix tells LiteLLM to use the OpenAI-compatible code path. The part after the slash should match a model name configured in your Control Layer.

Prompt caching

Doubleword caches when cache_control sits on a message content block. ADK sends its instruction as a plain string, so litellm attaches the marker to the message itself and the endpoint does not act on it. Neither ContextCacheConfig nor cache_control_injection_points produced a cache write when tested against ADK 2.9.1 with litellm 1.101.0, so caching cannot be driven from this integration today.

See the prompt caching guide.