Semantic Kernel (Python)

Microsoft Semantic Kernel for Python emits native OpenTelemetry spans, metrics, and logs to the global OpenTelemetry providers. Because logfire.configure() sets those global providers, SK’s telemetry flows to Logfire automatically once you enable SK’s experimental GenAI diagnostics with an environment variable.

Note

Unlike the .NET version, Semantic Kernel for Python does not require a source allowlist — it uses the global tracer/meter directly. So logfire.configure() plus the diagnostics env var is the whole integration.

Installation

Terminal

pip install logfire semantic-kernel

Usage

Set the diagnostics flag in your terminal before starting the script. The SENSITIVE variant records prompts and completions; use SEMANTICKERNEL_EXPERIMENTAL_GENAI_ENABLE_OTEL_DIAGNOSTICS=true instead for metadata only.

Terminal

export SEMANTICKERNEL_EXPERIMENTAL_GENAI_ENABLE_OTEL_DIAGNOSTICS_SENSITIVE=true

Then call logfire.configure() before creating the agent:

import asyncio

from semantic_kernel.agents import ChatCompletionAgent
from semantic_kernel.connectors.ai import FunctionChoiceBehavior
from semantic_kernel.connectors.ai.open_ai import OpenAIChatCompletion
from semantic_kernel.functions import kernel_function

import logfire

# Sets the global OTel tracer + meter provider exporting to Logfire.
logfire.configure(service_name='semantic-kernel-agent')


class WeatherPlugin:
    @kernel_function(description='Get the weather for a city')
    def get_weather(self, city: str) -> str:
        return f'The weather in {city} is sunny, 21C.'


async def main() -> None:
    agent = ChatCompletionAgent(
        service=OpenAIChatCompletion(ai_model_id='gpt-4o-mini'),  # uses OPENAI_API_KEY
        name='weather_agent',
        instructions='Use the weather plugin before answering.',
        function_choice_behavior=FunctionChoiceBehavior.Required(),
        plugins=[WeatherPlugin()],
    )
    response = await agent.get_response(messages="What's the weather in Paris?")
    print(response.message.content)


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

You’ll see an invoke_agent span in Live and Agents, with child chat.completions and function-invocation spans. With the SENSITIVE flag enabled, the conversation is also available in the agent-run detail. Semantic Kernel runs also appear in the specialized Agents view; the support matrix shows which columns each view populates.

Common pitfalls

• No diagnostics, no gen_ai spans. Without the SEMANTICKERNEL_EXPERIMENTAL_GENAI_* env var, you get function/kernel spans but no AI-connector spans. Set it before the kernel runs. Use the _SENSITIVE variant only when you want prompts/completions recorded.
• Don’t replace the global provider. Don’t call set_tracer_provider() / set_meter_provider() yourself after logfire.configure() — that would override Logfire’s exporter.

Managed prompts

Keep your prompts in Prompt Management and fetch them at runtime:

Terminal

pip install 'logfire[variables]'

import asyncio

from pydantic import BaseModel
from semantic_kernel.agents import ChatCompletionAgent
from semantic_kernel.connectors.ai import FunctionChoiceBehavior
from semantic_kernel.connectors.ai.open_ai import OpenAIChatCompletion
from semantic_kernel.functions import kernel_function

import logfire

logfire.configure()


class WeatherPlugin:
    @kernel_function(description='Get the weather for a city')
    def get_weather(self, city: str) -> str:
        return f'The weather in {city} is sunny, 21C.'


class WeatherInputs(BaseModel):
    city: str


prompt_var = logfire.template_var(
    name='prompt__weather_instructions',
    type=str,
    default='Use the weather plugin to answer questions about {{city}}.',
    inputs_type=WeatherInputs,
)

with prompt_var.get(WeatherInputs(city='Paris'), label='production') as resolved:
    prompt = resolved.value

async def main():
    agent = ChatCompletionAgent(
        service=OpenAIChatCompletion(ai_model_id='gpt-4o-mini'),
        name='weather_agent',
        instructions=prompt,
        function_choice_behavior=FunctionChoiceBehavior.Required(),
        plugins=[WeatherPlugin()],
    )
    response = await agent.get_response(messages="What's the weather in Paris?")
    print(response.message.content)


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

See Use Prompts in Your Application for the full workflow.