See every kernel invocation and plugin call.
Semantic Kernel orchestrates plugins, functions, and planners around your models; Obsivara tells you what each kernel run costs and how it behaves. Every invocation traced through plugin and function calls down to the model request — each carrying tokens, latency, and dollars — so a slow or expensive kernel run points to the exact function. Semantic Kernel emits OpenTelemetry GenAI spans; point them at Obsivara, out of the request path.
Semantic Kernel, fully observable.
Kernel invocation tracing
Every kernel run traced as a span tree — plugin calls, native and prompt functions, and the model requests behind them, with inputs, outputs, and timing.
Planner visibility
Planner-generated steps traced in order, so you see how the kernel decided to chain functions and where a plan went off course.
Per-function cost attribution
Tokens and dollars rolled up per function, plugin, and kernel run, so you see which functions drive spend and which models they call.
Failure attribution and health
Function errors, model failures, and timeouts classified and traced to the step that broke, with health scoring and predictive alerts.
Three steps. No code changes.
Point Semantic Kernel's OpenTelemetry at Obsivara
Semantic Kernel emits OpenTelemetry traces and metrics following the GenAI conventions — configure its OTLP exporter to Obsivara, or wrap your app with the Obsivara SDK. Your kernel keeps calling models directly; Obsivara records each run out-of-band, with no proxy in the path and no added latency.
Run your kernels
Traces flow as functions, plugins, and planners execute — no changes to your kernel setup or plugin code.
See it in Obsivara
Functions and plugins appear in the inventory and knowledge map with per-run cost, latency, and health within minutes.
Signals tracked out of the box.
- →Kernel run and function traces
- →Plugin and planner step sequences
- →Tokens and cost per function and run
- →Latency per step and model
- →Error rates and failing functions
- →Health score and predictive alerts
Common questions.
Which Semantic Kernel languages does Obsivara support?
Any that emit OpenTelemetry — the .NET, Python, and Java kernels all produce OTel GenAI spans that Obsivara ingests over OTLP.
Are planner steps traced?
Yes. Planner-generated function chains appear in order in the span tree, so you can see how the kernel assembled and executed a plan.
Do I need to change my kernel code?
No. You enable Semantic Kernel's OpenTelemetry export (or the Obsivara SDK); your plugins, functions, and planners stay exactly as they are.