Google ADK (gateway)

Route Google ADK’s underlying LLM calls through the Respan gateway to use 250+ models from different providers. The examples use ADK’s LiteLlm adapter and load keys from the repo root .env.

Setup

Install packages

$ pip install respan-ai respan-instrumentation-google-adk "google-adk[extensions]"

Set environment variables

$ export RESPAN_API_KEY="YOUR_RESPAN_API_KEY"
$ export RESPAN_MODEL="gpt-4o-mini"

Set RESPAN_GATEWAY_API_KEY or RESPAN_GATEWAY_BASE_URL only if they differ from your tracing key or base URL.

Point Google ADK to the Respan gateway

Use LiteLlm with the OpenAI-compatible Respan endpoint to route ADK’s LLM calls.

1 import os
2 from google.adk.agents import Agent
3 from google.adk.models.lite_llm import LiteLlm
4 
5 model_name = os.getenv("RESPAN_MODEL", "gpt-4o-mini")
6 if "/" not in model_name:
7     model_name = f"openai/{model_name}"
8 
9 agent = Agent(
10     name="assistant",
11     model=LiteLlm(
12         model=model_name,
13         api_key=os.getenv("RESPAN_GATEWAY_API_KEY") or os.environ["RESPAN_API_KEY"],
14         api_base=os.getenv("RESPAN_GATEWAY_BASE_URL", "https://api.respan.ai/api"),
15     ),
16     instruction="You are a concise assistant.",
17 )

Switch models

Change the model parameter on LiteLlm to use different gateway-backed providers through the same endpoint.

1 LiteLlm(model="openai/gpt-4o-mini", api_key=..., api_base="https://api.respan.ai/api")
2 LiteLlm(model="openai/gemini-2.5-flash", api_key=..., api_base="https://api.respan.ai/api")

See the full model list.

$	export RESPAN_API_KEY="YOUR_RESPAN_API_KEY"
$	export RESPAN_MODEL="gpt-4o-mini"

1	import os
2	from google.adk.agents import Agent
3	from google.adk.models.lite_llm import LiteLlm
4
5	model_name = os.getenv("RESPAN_MODEL", "gpt-4o-mini")
6	if "/" not in model_name:
7	model_name = f"openai/{model_name}"
8
9	agent = Agent(
10	name="assistant",
11	model=LiteLlm(
12	model=model_name,
13	api_key=os.getenv("RESPAN_GATEWAY_API_KEY") or os.environ["RESPAN_API_KEY"],
14	api_base=os.getenv("RESPAN_GATEWAY_BASE_URL", "https://api.respan.ai/api"),
15	),
16	instruction="You are a concise assistant.",
17	)

1	LiteLlm(model="openai/gpt-4o-mini", api_key=..., api_base="https://api.respan.ai/api")
2	LiteLlm(model="openai/gemini-2.5-flash", api_key=..., api_base="https://api.respan.ai/api")