Skip to content
OpenObserve
Get Demo Star
OpenObserve
Get Demo Star
Open Menu Close Menu

Google ADK → OpenObserve

Automatically capture agent runs, LLM calls, tool executions, and token usage for every Google ADK agent in your Python application.

Prerequisites

  • Python 3.10+
  • An OpenObserve account (cloud or self-hosted)
  • Your OpenObserve organisation ID and Base64-encoded auth token
  • A Google AI Studio API key (GOOGLE_API_KEY)

Installation

pip install openobserve-telemetry-sdk openinference-instrumentation-google-adk google-adk python-dotenv

Configuration

Create a .env file in your project root:

# OpenObserve instance URL
# Default for self-hosted: http://localhost:5080
OPENOBSERVE_URL=https://api.openobserve.ai/

# Your OpenObserve organisation slug or ID
OPENOBSERVE_ORG=your_org_id

# Basic auth token — Base64-encoded "email:password"
OPENOBSERVE_AUTH_TOKEN=Basic <your_base64_token>

# Google AI Studio API key
GOOGLE_API_KEY=your-google-api-key

Instrumentation

Call GoogleADKInstrumentor().instrument() before importing any ADK modules.

from dotenv import load_dotenv
load_dotenv()

from openinference.instrumentation.google_adk import GoogleADKInstrumentor
from openobserve import openobserve_init

GoogleADKInstrumentor().instrument()
openobserve_init()

import asyncio
from google.adk import Agent, Runner
from google.adk.sessions import InMemorySessionService
from google.genai import types


def get_weather(city: str) -> str:
    """Get current weather for a city."""
    return f"Sunny, 22°C in {city}"


agent = Agent(
    name="weather_agent",
    model="gemini-2.5-flash",
    instruction="You are a helpful assistant. Use get_weather for weather questions.",
    tools=[get_weather],
)

session_service = InMemorySessionService()
runner = Runner(agent=agent, session_service=session_service, app_name="my_app")


async def main():
    await session_service.create_session(
        app_name="my_app", user_id="user1", session_id="session1"
    )
    content = types.Content(
        role="user",
        parts=[types.Part.from_text(text="What's the weather in London?")],
    )
    async for event in runner.run_async(
        user_id="user1", session_id="session1", new_message=content
    ):
        if event.content and event.content.parts:
            for part in event.content.parts:
                if getattr(part, "text", None):
                    print(part.text)


asyncio.run(main())

What Gets Captured

Each agent invocation produces a trace with a hierarchy of spans: invocation [<app_name>] at the root, agent_run [<agent_name>] for the agent execution, call_llm for each LLM API call, and execute_tool <tool_name> for each tool invocation. Multi-turn agents produce multiple call_llm spans within a single agent_run.

Attribute Description
gcp_vertex_agent_name Agent name (e.g. weather_agent)
gcp_vertex_agent_invocation_id Unique ID for the agent invocation
gen_ai_request_model Model used (e.g. gemini-2.5-flash)
gen_ai_system gcp.vertex.agent
gen_ai_conversation_id Session identifier passed to the runner
llm_usage_input_tokens Input tokens for the LLM call
llm_usage_output_tokens Output tokens for the LLM call
llm_usage_tokens_total Total tokens consumed
llm_usage_cost_input Estimated input cost in USD
llm_usage_cost_output Estimated output cost in USD
llm_token_count_completion_details_reasoning Reasoning tokens (Gemini 2.5 thinking budget)
user_id User identifier passed to the runner
duration Span latency
span_status OK or ERROR

Viewing Traces

  1. Log in to OpenObserve and navigate to Traces in the left sidebar
  2. Click any invocation [<app_name>] root span to open the waterfall view
  3. Expand the tree to see agent_run, call_llm, and execute_tool child spans
  4. Click a call_llm span to inspect token counts, model, and cost for that specific LLM call

Google ADK trace in OpenObserve

Next Steps

With Google ADK instrumented, every agent invocation is recorded in OpenObserve with a full span hierarchy. From here you can track token usage per agent run, compare latency across tools and LLM calls, and set alerts on error spans from failed tool executions.

Read More