Semantic Kernel

Semantic Kernel emits GenAI spans of its own once one setting is on. Python.

Turn tracing on

pip install semantic-kernel opentelemetry-sdk opentelemetry-exporter-otlp-proto-http

Set the exporter variables from OpenTelemetry, and this one:

SEMANTICKERNEL_EXPERIMENTAL_GENAI_ENABLE_OTEL_DIAGNOSTICS=true

Then once at startup:

from opentelemetry import trace
from opentelemetry.sdk.trace import TracerProvider
from opentelemetry.sdk.trace.export import BatchSpanProcessor
from opentelemetry.exporter.otlp.proto.http.trace_exporter import OTLPSpanExporter

provider = TracerProvider()
provider.add_span_processor(BatchSpanProcessor(OTLPSpanExporter()))
trace.set_tracer_provider(provider)

What is recorded

• Each model call, with its model, provider and input and output tokens.
• Each tool call, named <plugin>-<function>.
• A model call that fails, as failed.

Semantic Kernel does not report cached tokens, so a call that read from the cache is priced as if it had not.

Checked with Semantic Kernel 1.36.

Budgets

Spans arrive after the call, so spend counts against your budgets and raises alerts, but a budget cannot stop a call recorded this way.