Observe system execution
Collect signals that help explain what the running system did and where it failed.
Desired outcome
Traceable operational behavior and failure signals.
Explore from the outcome
Where this function can lead
These documented connections are starting points for selection, not a ranked list.
Ways to do it
Methods that can help achieve this outcome.
- LLM application tracing
Record request-level model, retrieval, and tool steps as linked observations.
Link evidence: Langfuse observability overview
- Telemetry instrumentation
Emit traces, metrics, and logs from a running system for later analysis.
Link evidence: OpenTelemetry instrumentation
Tools and services
Software or services that can put a method to work.
- Langfuse
Application tracing platform for LLM calls, retrieval, and tool steps.
Link evidence: Langfuse observability overview
- OpenTelemetry
Toolkit for instrumenting systems and exporting traces, metrics, and logs.
Link evidence: What is OpenTelemetry?
Specifications and guidance
Standards and frameworks relevant to the work.
- OpenTelemetry GenAI Semantic Conventions
Developing conventions for recording generative-AI spans, metrics, and events.
Link evidence: OpenTelemetry GenAI semantic conventions
Typical inputs
- Runtime events and instrumentation
Typical outputs
- Traces, metrics and logs
Sources
Last reviewed 2026-09-30