langsmith-trace

LangSmith tracing enables observing execution trees and diagnosing agent behavior by recording nested LLM, tool, and workflow activity. You can instrument end-to-end capture using LangChain environment variables, OpenTelemetry integration, or framework-agnostic wrappers.

You can export LangSmith trace data using the langsmith CLI to list, fetch full trace hierarchies, and export run data. Execute trace and run subcommands with appropriate filters and project scoping for downstream analysis or dataset generation.

LangSmith tracing works across Python and JavaScript/LangChain applications and non-LangChain frameworks. You enable trace capture using OpenTelemetry instrumentation or framework-agnostic traceable wrappers for these environments.

A trace represents the complete execution tree, while a run is an individual node within that hierarchy. Disambiguating traces versus runs helps you choose the correct CLI operations and filters when inspecting and exporting data.

Querying LangSmith runs requires a LANGSMITH_API_KEY for authentication. Optional project scoping allows you to filter trace inspection and export tasks to specific projects when using the CLI.

The best way to list recent LangSmith traces is using the langsmith CLI trace list command. Provide your correct project name and LANGSMITH_API_KEY to retrieve and inspect the most recent hierarchical execution data.