Skip to main content

Traces

Langflow’s Traces feature records detailed execution traces for your flows and components so that you can debug issues, measure latency, and track token usage without relying on external observability services.

Trace data is stored in the Langflow database in the trace and span tables. Trace data is presented in the Flow Activity and Trace Details pages in the UI, and can be retrieved from the /monitor/traces API endpoint.

Traces are enabled by default. To disable Langflow tracing and use a different tracing provider, set LANGFLOW_NATIVE_TRACING to false.

What traces capture​

The tracer records:

  • Flow-level traces: A trace for each flow run, including total runtime and status.
  • Component spans: Spans for each component in the flow, including inputs, outputs, latency, and errors.
  • LangChain spans: Deeper spans for chains, tools, retrievers, and LLM calls, including model name and token usage where available.

Each span includes:

  • Name and type (for example, chain, LLM, tool, retriever)
  • Start and end time and latency (ms)
  • Inputs and outputs (serialized)
  • Error details, if the span failed
  • Attributes such as token counts and model metadata

View traces in the UI​

To view traces in the Langflow UI, do the following:

  1. Run a flow, such as the Simple Agent starter flow in the Quickstart.
  2. Click Traces. The Flow Activity page opens. Each flow run is displayed as a single trace of all of its spans. Flow runs can be sorted further by session ID, status, or time range. Optionally, click Download to download a JSON file of that flow's trace to your local machine.
  3. Click a flow run to open the Trace Details pane. The Trace Details pane displays spans for your flow run, including a flow-level span for the entire run, and a span for each component. Individual component spans include the component's inputs and outputs, timing, and token usage.

Retrieve traces with the API​

To programmatically query traces, use the /monitor/traces endpoints. For full parameter details and code examples in Python, TypeScript, and curl, see Monitor endpoints: Get traces.

See also​

Search