What is LangSmith and what is it used for?
LangSmith is LangChain's platform for observing, evaluating and improving applications and agents built on language models. It is used to see what an agent did at each step, measure answer quality and test prompt changes before they reach production.
The starting point is the trace: a record of a full run, with every model call, tool invoked, input, output, latency and cost. With traces, a team can find where an agent went wrong or got slow without reproducing the problem by hand.
LangSmith does not depend on the LangChain framework. It traces applications built with the OpenAI SDK, the Anthropic SDK, the Vercel AI SDK, LlamaIndex or custom code, and it accepts traces over OpenTelemetry.
Evaluations sit on top of traces. You can run evaluations on datasets before deploying, score live production traffic, use a model as a judge, write evaluators in code and send runs to human review in annotation queues. The Playground tests a prompt against many examples without writing code.
The audience is engineering teams that put agents and AI applications into production and need evidence of quality, not just impressions.






