splunk-otel-instrumentation-openai-agents 0.1.5

Splunk OpenTelemetry OpenAI Agents instrumentation

OpenTelemetry OpenAI Agents Instrumentation (Alpha)

This package provides OpenTelemetry instrumentation for OpenAI Agents SDK, a framework for building agentic AI applications. It leverages Splunk distribution of opentelemetry-util-genai for producing telemetry in semantic convention. Core concepts, high-level usage and configuration

Status: Alpha (APIs and produced telemetry are subject to change).

Installation

pip install splunk-otel-instrumentation-openai-agents

Quick Start

Manual Instrumentation (development/debugging)

from opentelemetry.instrumentation.openai_agents import OpenAIAgentsInstrumentor
from agents import Agent, Runner, function_tool

# manual instrumentation, easy to debug in your IDE
OpenAIAgentsInstrumentor().instrument()

@function_tool
def get_weather(city: str) -> str:
    return f"The forecast for {city} is sunny."

agent = Agent(
    name="Travel Concierge",
    instructions="You are a concise travel concierge.",
    tools=[get_weather],
)

result = Runner.run_sync(agent, "How is the weather in Paris?")
print(result.final_output)

Zero-code instrumentation

In zero-code instrumentation mode, ensure you install opentelemetry-distribution and run your app with the OpenTelemetry OpenAI Agents Instrumentor enabled:

opentelemetry-instrument python your_agents_app.py

from agents import Agent, Runner

agent = Agent(
    name="Travel Concierge",
    instructions="You are a concise travel concierge.",
)

result = Runner.run_sync(agent, "What are top attractions in Tokyo?")
print(result.final_output)

What Gets Instrumented

This instrumentation captures:

• Workflows -> Mapped to Workflow spans
• Agents -> Mapped to AgentInvocation spans
• Tool Calls -> Mapped to ToolCall spans
• Generations -> Mapped to LLMInvocation spans

All spans are properly nested with correct parent-child relationships and include rich attributes about the operation.

Configuration

Enabling message content

Message content such as prompts, completions, tool arguments, and return values are not captured by default. To enable content capturing, set the environment variable OTEL_INSTRUMENTATION_GENAI_CAPTURE_MESSAGE_CONTENT to one of the following values:

• true — Legacy. Treated as span_and_event for backward compatibility.
• span_only — Capture content on span attributes only.
• event_only — Capture content on event attributes only.
• span_and_event — Capture content on both span and event attributes.

export OTEL_INSTRUMENTATION_GENAI_CAPTURE_MESSAGE_CONTENT=span_and_event

Other environment variables

# Capture tool/function definitions on LLM spans (disabled by default)
export OTEL_INSTRUMENTATION_GENAI_CAPTURE_TOOL_DEFINITIONS=true

# Disable metrics
export OTEL_INSTRUMENTATION_OPENAI_AGENTS_CAPTURE_METRICS=false

Backward compatibility: The deprecated OTEL_INSTRUMENTATION_GENAI_CAPTURE_MESSAGE_CONTENT_MODE variable is still honored when the primary variable is set to a truthy value (true/1/yes/on).

Instrumentation Options

from opentelemetry.instrumentation.openai_agents import OpenAIAgentsInstrumentor

# Basic instrumentation
OpenAIAgentsInstrumentor().instrument()

# With custom tracer provider
OpenAIAgentsInstrumentor().instrument(tracer_provider=my_tracer_provider)

# Uninstrumentation
OpenAIAgentsInstrumentor().uninstrument()

Requirements

• Python >= 3.10
• openai-agents >= 0.3.3
• OpenTelemetry API >= 1.37
• splunk-otel-util-genai >= 0.1.9

Trace Hierarchy Example

Workflow: Travel Planning (Workflow)
└── Agent: Travel Concierge (AgentInvocation)
    ├── LLM: gpt-4o-mini (Generation)
    │   └── gen_ai.choice
    ├── Tool: get_weather (ToolCall)
    └── LLM: gpt-4o-mini (Generation)
        └── gen_ai.choice

Each span includes rich attributes:

• gen_ai.system = "openai"
• gen_ai.framework = "openai_agents"
• gen_ai.operation.name = "invoke_workflow" | "invoke_agent" | "execute_tool" | "chat"
• Framework-specific attributes (agent name, tool names, etc.)

Testing

Run the package tests (from repository root or this directory):

pytest instrumentation-genai/opentelemetry-instrumentation-openai-agents-v2/tests

Contributing

Issues / PRs welcome in the main splunk-otel-python-contrib repository. This module is alpha: feedback on attribute coverage, performance, and OpenAI Agents surface expansion is especially helpful.

Links

• OpenAI Agents SDK
• OpenTelemetry Python
• Splunk GenAI Utilities

License

Apache-2.0