Reference · Glossary
Span (tracing)
Last updated
One **timed unit of work** in a trace — e.g. embed query, retrieve chunks, call chat completion.
When to use
Debugging multi-step AI apps where you need latency and inputs per step.
When not to
Single-call demos with no observability requirement.
Example
Trace shows `retrieve_docs` span returned 0 chunks in 12ms → root cause before blaming the LLM.