Reference · Glossary
LangSmith
Last updated
A hosted tracing and evaluation platform (from the LangChain team) that logs every LLM call in your app — prompts, responses, latency, cost, tool calls — as a searchable trace tree, plus dataset-based evaluation runs.
#When to use
Debugging a multi-step chain or agent in production where you need to see exactly which step failed or produced a bad output, and tracking quality/cost over time.
#When not to
A single-prompt script with no chain of calls — plain logging may be enough until your pipeline has several dependent steps worth tracing.
#Example
Wrapping your app with LangSmith tracing turns "the agent gave a wrong answer" into "the retrieval step returned the wrong chunk, which is why step 3 hallucinated" — visible in one trace.