Trace Google ADK agents

This guide shows you how to trace Google Agent Development Kit (ADK) agents in LangSmith. You’ll configure automatic tracing for your ADK agents to capture agent invocations, tool calls, and LLM interactions.

This guide covers ADK’s standard execution paths (text agents, tools, and multi-agent workflows). To trace a Gemini Live voice agent, which uses ADK’s Runner.run_live streaming loop, see Trace Gemini Live agents.

In an agent-based workspace, create an agent once before sending traces. Use agent addressing to select its identifier and environment. If an integration requires project_name, use the selected environment’s tracer_session_name as a compatibility value.

Installation

Install the required packages using your preferred package manager:

uv add "langsmith[google-adk]"

pip install langsmith[google-adk]

Setup

Sending traces to an agent requires langsmith>=0.14.9. Set your API keys, agent identifier, and environment:

export LANGSMITH_TRACING=true
export LANGSMITH_ENDPOINT=https://api.smith.langchain.com
export LANGSMITH_API_KEY=<your_langsmith_api_key>
export LANGSMITH_AGENT_ID="<your-agent-id>"
export LANGSMITH_AGENT_ENVIRONMENT="local"

export GOOGLE_API_KEY=<your_google_api_key>

LANGSMITH_TRACING=true
LANGSMITH_ENDPOINT=https://api.smith.langchain.com
LANGSMITH_API_KEY=<your_langsmith_api_key>
LANGSMITH_AGENT_ID="<your-agent-id>"
LANGSMITH_AGENT_ENVIRONMENT="local"

GOOGLE_API_KEY=<your_google_api_key>

To create a Google API key, refer to Google AI Studio.

Configure tracing

To trace ADK agents, use configure_google_adk() from the LangSmith SDK. Call this function once before creating any ADK agents:

from langsmith.integrations.google_adk import configure_google_adk

configure_google_adk()

The function accepts the following optional parameters:

address: An agent address to send traces to instead of the one LANGSMITH_AGENT_ID and LANGSMITH_AGENT_ENVIRONMENT name, for example ls.agent_address("my-agent", "production"). Cannot be combined with project_name. Requires langsmith>=0.14.14.
project_name: Compatibility override for project-based tracing. Leave it unset to use the configured agent and environment.
name: Name for the root trace. Defaults to "google_adk.session".
metadata: Dictionary of key-value pairs for additional context.
tags: List of strings to categorize traces.

Example

This example creates a weather agent with a tool, then runs it with tracing enabled:

import asyncio

from dotenv import load_dotenv  # Optional
from google.adk.agents import Agent
from google.adk.runners import Runner
from google.adk.sessions import InMemorySessionService
from google.genai import types
from langsmith.integrations.google_adk import configure_google_adk

load_dotenv()  # Optional


async def main():
    # Configure LangSmith tracing
    configure_google_adk()

    # Define a tool
    def get_weather(city: str) -> dict:
        """Get weather for a city."""
        return {"city": city, "temperature": "72°F", "conditions": "Sunny"}

    # Create the agent
    agent = Agent(
        name="weather_agent",
        model="gemini-3.8-flash",
        description="Provides weather information.",
        instruction="Use the get_weather tool to answer weather questions.",
        tools=[get_weather],
    )

    # Set up session and runner
    session_service = InMemorySessionService()
    session = await session_service.create_session(
        app_name="weather_app",
        user_id="user_123",
        session_id="session_456",
    )

    runner = Runner(
        agent=agent,
        app_name="weather_app",
        session_service=session_service,
    )

    # Run the agent
    async for event in runner.run_async(
        user_id="user_123",
        session_id=session.id,
        new_message=types.Content(
            role="user",
            parts=[types.Part(text="What's the weather in San Francisco?")],
        ),
    ):
        if event.is_final_response():
            print(event.content.parts[0].text)


if __name__ == "__main__":
    asyncio.run(main())

View traces in LangSmith

After running your agent, you can view traces in the LangSmith UI that include:

Agent invocations: Complete flows through your ADK agents
Tool calls: Individual function calls made by agents
LLM interactions: Requests and responses from Gemini models
Multi-agent workflows: Traces from sequential and parallel agent compositions

Custom metadata and tags

Add metadata and tags when configuring tracing to categorize and filter traces:

from langsmith.integrations.google_adk import configure_google_adk

configure_google_adk(
    metadata={
        "environment": "production",
        "team": "ml-platform",
    },
    tags=["adk", "weather", "v2"],
)

Multi-agent workflows

The integration automatically traces multi-agent workflows including sequential and parallel agent compositions:

import asyncio

from dotenv import load_dotenv  # Optional
from google.adk.agents import Agent, SequentialAgent
from google.adk.runners import Runner
from google.adk.sessions import InMemorySessionService
from google.genai import types
from langsmith.integrations.google_adk import configure_google_adk

load_dotenv()  # Optional


async def main():
    # Configure LangSmith tracing
    # Traces go to LANGSMITH_PROJECT env var by default.
    # Pass project_name="my-project" to override.
    configure_google_adk()

    # Create sub-agents
    translator = Agent(
        name="translator",
        model="gemini-3.8-flash",
        description="Translates text to English.",
    )

    summarizer = Agent(
        name="summarizer",
        model="gemini-3.8-flash",
        description="Summarizes text concisely.",
    )

    # Create a sequential agent that runs sub-agents in order
    pipeline = SequentialAgent(
        name="translate_and_summarize",
        sub_agents=[translator, summarizer],
        description="Translates text then summarizes it.",
    )

    # Set up and run
    session_service = InMemorySessionService()
    session = await session_service.create_session(
        app_name="pipeline_app",
        user_id="user_123",
        session_id="session_456",
    )

    runner = Runner(
        agent=pipeline,
        app_name="pipeline_app",
        session_service=session_service,
    )

    events = runner.run_async(
        user_id="user_123",
        session_id=session.id,
        new_message=types.Content(
            role="user",
            parts=[types.Part(text="Quelle est la plus haute tour de Paris?")],
        ),
    )

    async for event in events:
        if event.is_final_response():
            print(event.content.parts[0].text)


if __name__ == "__main__":
    asyncio.run(main())

Connect these docs to your agent of choice via MCP for real-time answers.

Edit this page on GitHub or file an issue.

Tracing setup

Configuration & troubleshooting

Installation

Setup

Configure tracing

Example

View traces in LangSmith

Custom metadata and tags

Multi-agent workflows

​Installation

​Setup

​Configure tracing

​Example

​View traces in LangSmith

​Custom metadata and tags

​Multi-agent workflows

Installation

Setup

Configure tracing

Example

View traces in LangSmith

Custom metadata and tags

Multi-agent workflows