Instrument your agent How spans are identified How evaluation fields are extracted Example spans with event records Example spans without event records

Google ADK

This page explains how to instrument a Google Agent Development Kit (ADK) agent, how spans are identified, and how evaluation fields are extracted.

Topics

Instrument your agent
How spans are identified
How evaluation fields are extracted
- From event records
- From span attributes
Example spans with event records
Example spans without event records

Instrument your agent

You can instrument a Google ADK agent with the OpenInference instrumentation library (openinference-instrumentation-google-adk). This library emits telemetry under the scope name openinference.instrumentation.google_adk, which Amazon Bedrock AgentCore Evaluations reads.

When your agent runs with the AWS Distro for OpenTelemetry (ADOT), such as on Amazon Bedrock AgentCore Runtime, you do not need to add explicit instrumentation code. Adding the instrumentation library to your project’s dependencies is enough. ADOT discovers it at startup and activates it automatically.

Add the instrumentation library to your dependencies.

Note

Use version 0.1.13 or later. This is the earliest version tested with the evaluation service.

requirements.txt:


openinference-instrumentation-google-adk>=0.1.13

pyproject.toml:


[project]
dependencies = [
    "openinference-instrumentation-google-adk>=0.1.13",
]

Note

Instrumentation is one step in setting up observability. To export telemetry for evaluation, complete the full setup in Set up observability.

How spans are identified

Google ADK is instrumented with the OpenInference convention, so AgentCore Evaluations classifies spans using the openinference.span.kind attribute.

Span type	Identifying attribute
Invoke agent	`openinference.span.kind` = `CHAIN` or `AGENT`
Execute tool	`openinference.span.kind` = `TOOL`
Inference	`openinference.span.kind` = `LLM`

Google ADK emits a nested span tree: an outer invocation span (CHAIN) wraps an agent_run span (AGENT), which in turn wraps the call_llm (LLM) and execute_tool (TOOL) spans. The outer CHAIN span carries the user prompt; AgentCore Evaluations uses it as the invoke agent span.

How evaluation fields are extracted

Google ADK wraps its conversation content in the Gemini content format. The user prompt is nested under a new_message object as {"new_message": {"parts": [{"text": "…"}], "role": "user"}}, and the agent response is nested under a content object as {"content": {"parts": [{"text": "…"}], "role": "model"}}. AgentCore Evaluations unwraps these structures and joins the parts text with newlines. Tool definitions arrive as a serialized Gemini request; AgentCore Evaluations reads the available tools from config.tools[].function_declarations[].

The location of this content depends on how telemetry was collected. The identifying attribute (openinference.span.kind) is on the span in both cases. For more information, see Spans, event records, and telemetry signals.

From event records

When telemetry is split, AgentCore Evaluations reads content from the event record correlated to each span:

User prompt: from the invoke agent span’s event record, in body.input. AgentCore Evaluations unwraps the new_message.parts text.
Agent response: from the invoke agent span’s event record, in body.output. AgentCore Evaluations unwraps the content.parts text.
Tool call: the tool name from the tool.name attribute on the execute tool span. The tool arguments and result come from that span’s event record, in body.input and body.output.

For examples, see Example spans with event records.

From span attributes

When telemetry is not split, the same content stays on the span as attributes:

User prompt and agent response: from input.value and output.value on the invoke agent span. AgentCore Evaluations unwraps the new_message.parts and content.parts text.
Tool call: the tool name from tool.name, and the arguments and result from input.value and output.value, on the execute tool span.

For examples, see Example spans without event records.

Example spans with event records

When telemetry is split, the span carries the identifying attributes and the content lives in a correlated event record. The following examples are from a Google ADK travel-planning agent deployed on Amazon Bedrock AgentCore Runtime.

Note

These examples are not complete spans. They show representative data from a real agent interaction, with some fields omitted and long values truncated for readability.

Example

Invoke agent span

The openinference.span.kind attribute (CHAIN) on the outer invocation span identifies this as an invoke agent span. The span carries no conversation content; it lives in the correlated event record.


{
  "traceId": "6a387ee61078243c1cc455ed45c6c313",
  "spanId": "70f2e87a30c34420",
  "name": "invocation",
  "kind": "INTERNAL",
  "scope": {
    "name": "openinference.instrumentation.google_adk",
    "version": "0.1.14"
  },
  "attributes": {
    "openinference.span.kind": "CHAIN",
    "input.mime_type": "application/json",
    "output.mime_type": "application/json",
    "user.id": "default_user",
    "session.id": "sea-nyc-trip-2-turns-google-adk-adot"
  },
  "status": {
    "code": "OK"
  }
}

The correlated event record carries the conversation. The user prompt is nested under new_message.parts, and the agent response is nested under content.parts.


{
  "spanId": "70f2e87a30c34420",
  "traceId": "6a387ee61078243c1cc455ed45c6c313",
  "scope": {
    "name": "openinference.instrumentation.google_adk"
  },
  "body": {
    "input": {
      "messages": [
        {
          "role": "user",
          "content": "{\"new_message\": {\"parts\": [{\"text\": \"Hey, how can you help me\"}], \"role\": \"user\"}, \"state_delta\": null, \"run_config\": null}"
        }
      ]
    },
    "output": {
      "messages": [
        {
          "role": "assistant",
          "content": "{\"model_version\": \"gemini-2.5-flash\", \"content\": {\"parts\": [{\"text\": \"I can help you with your travel plans! I can:\\n- Search and book flights\\n- Find and book hotels\\n- Suggest and book activities\"}], \"role\": \"model\"}, \"finish_reason\": \"STOP\"}"
        }
      ]
    }
  }
}

Execute tool span

The openinference.span.kind attribute (TOOL) identifies this as an execute tool span; tool.name holds the tool name. The tool arguments and result live in the correlated event record.


{
  "traceId": "6a387ef07b8f4f3732fab45d3c0b51ff",
  "spanId": "9028a8dd94943456",
  "name": "execute_tool search_flights",
  "kind": "INTERNAL",
  "scope": {
    "name": "openinference.instrumentation.google_adk",
    "version": "0.1.14"
  },
  "attributes": {
    "openinference.span.kind": "TOOL",
    "gen_ai.operation.name": "execute_tool",
    "gen_ai.tool.name": "search_flights",
    "gen_ai.tool.type": "FunctionTool",
    "gen_ai.tool.call.id": "adk-12345678-1234-1234-1234-123456789012",
    "tool.name": "search_flights",
    "tool.parameters": "{\"origin\": \"SEA\", \"destination\": \"NYC\", \"date\": \"2025-03-15\"}",
    "session.id": "sea-nyc-trip-2-turns-google-adk-adot"
  },
  "status": {
    "code": "OK"
  }
}


{
  "spanId": "9028a8dd94943456",
  "traceId": "6a387ef07b8f4f3732fab45d3c0b51ff",
  "scope": {
    "name": "openinference.instrumentation.google_adk"
  },
  "body": {
    "input": {
      "messages": [
        { "role": "user", "content": "{\"origin\": \"SEA\", \"destination\": \"NYC\", \"date\": \"2025-03-15\"}" }
      ]
    },
    "output": {
      "messages": [
        {
          "role": "assistant",
          "content": "{\"id\": \"adk-12345678-...\", \"name\": \"search_flights\", \"response\": {\"origin\": \"SEA\", \"destination\": \"NYC\", \"flights\": [ ... ]}}"
        }
      ]
    }
  }
}

Inference span

The openinference.span.kind attribute (LLM) on the call_llm span identifies this as an inference span. It carries the model metadata and, in the indexed llm.input_messages. and llm.output_messages. attributes, the messages for the model call.


{
  "traceId": "6a387ee61078243c1cc455ed45c6c313",
  "spanId": "1c4e5f8a2b9d0e73",
  "name": "call_llm",
  "kind": "INTERNAL",
  "scope": {
    "name": "openinference.instrumentation.google_adk",
    "version": "0.1.14"
  },
  "attributes": {
    "openinference.span.kind": "LLM",
    "gen_ai.operation.name": "generate_content",
    "gen_ai.request.model": "gemini-2.5-flash",
    "llm.model_name": "gemini-2.5-flash",
    "llm.input_messages.0.message.role": "system",
    "llm.input_messages.1.message.role": "user",
    "llm.input_messages.1.message.contents.0.message_content.text": "Hey, how can you help me",
    "llm.output_messages.0.message.role": "model",
    "llm.output_messages.0.message.contents.0.message_content.text": "I can help you plan your trip ...",
    "session.id": "sea-nyc-trip-2-turns-google-adk-adot"
  },
  "status": {
    "code": "OK"
  }
}

Example spans without event records

When telemetry is not split, the same content stays on the span attributes and no separate event record is produced. The following examples are from a Google ADK travel-planning agent.

Note

These examples are not complete spans. They show representative data from a real agent interaction, with some fields omitted and long values truncated for readability.

Example

Invoke agent span

The input.value attribute holds the user prompt (nested under new_message.parts), and the output.value attribute holds the agent response (nested under content.parts).


{
  "traceId": "6a4de7b85e61747e6b568a1f4768e89d",
  "spanId": "31ea3d5882dac680",
  "name": "invocation",
  "kind": "INTERNAL",
  "scope": {
    "name": "openinference.instrumentation.google_adk",
    "version": "0.1.13"
  },
  "attributes": {
    "openinference.span.kind": "CHAIN",
    "input.value": "{\"user_id\": \"test_user\", \"session_id\": \"sea-nyc-trip-2-turns-google-adk-unified\", \"new_message\": {\"parts\": [{\"text\": \"Hey, how can you help me\"}], \"role\": \"user\"}}",
    "input.mime_type": "application/json",
    "output.value": "{\"model_version\": \"gemini-2.5-flash\", \"content\": {\"parts\": [{\"text\": \"I can help you plan your trip! I can:\\n- Search and book flights\\n- Find and book hotels\\n- Suggest and book activities\"}], \"role\": \"model\"}, \"finish_reason\": \"STOP\"}",
    "output.mime_type": "application/json",
    "session.id": "sea-nyc-trip-2-turns-google-adk-unified"
  },
  "status": {
    "code": "OK"
  }
}

Execute tool span

The input.value attribute holds the tool arguments, and the output.value attribute holds the tool result.


{
  "traceId": "6a4de7c376913db82e6f0f336a16731d",
  "spanId": "b64c37adefae74f0",
  "name": "execute_tool search_flights",
  "kind": "INTERNAL",
  "scope": {
    "name": "openinference.instrumentation.google_adk",
    "version": "0.1.13"
  },
  "attributes": {
    "openinference.span.kind": "TOOL",
    "gen_ai.operation.name": "execute_tool",
    "gen_ai.tool.name": "search_flights",
    "gen_ai.tool.call.id": "adk-12345678-1234-1234-1234-123456789012",
    "tool.name": "search_flights",
    "tool.parameters": "{\"origin\": \"SEA\", \"destination\": \"NYC\", \"date\": \"2025-03-15\"}",
    "input.value": "{\"origin\": \"SEA\", \"destination\": \"NYC\", \"date\": \"2025-03-15\"}",
    "output.value": "{\"id\": \"adk-12345678-...\", \"name\": \"search_flights\", \"response\": {\"origin\": \"SEA\", \"destination\": \"NYC\", \"flights\": [ ... ]}}",
    "session.id": "sea-nyc-trip-2-turns-google-adk-unified"
  },
  "status": {
    "code": "OK"
  }
}

Inference span

The openinference.span.kind attribute (LLM) on the call_llm span identifies this as an inference span. The messages for the model call are inline on the indexed llm.input_messages. and llm.output_messages. attributes.


{
  "traceId": "6a4de7b85e61747e6b568a1f4768e89d",
  "spanId": "2d5f6a9b3c0e1f84",
  "name": "call_llm",
  "kind": "INTERNAL",
  "scope": {
    "name": "openinference.instrumentation.google_adk",
    "version": "0.1.13"
  },
  "attributes": {
    "openinference.span.kind": "LLM",
    "gen_ai.operation.name": "generate_content",
    "gen_ai.request.model": "gemini-2.5-flash",
    "llm.model_name": "gemini-2.5-flash",
    "llm.input_messages.0.message.role": "system",
    "llm.input_messages.1.message.role": "user",
    "llm.input_messages.1.message.contents.0.message_content.text": "Hey, how can you help me",
    "llm.output_messages.0.message.role": "model",
    "llm.output_messages.0.message.contents.0.message_content.text": "I can help you plan your trip ...",
    "session.id": "sea-nyc-trip-2-turns-google-adk-unified"
  },
  "status": {
    "code": "OK"
  }
}

Warning Javascript is disabled or is unavailable in your browser.

To use the Amazon Web Services Documentation, Javascript must be enabled. Please refer to your browser's Help pages for instructions.

Document Conventions

LlamaIndex

Claude Agent SDK