📊We’ve rebuilt Dashboards in LangSmith to make them more useful for investigati
- Source
- docs.langchain.com
- Author
- LangChain
- Date

Context
An dashboard can overcount user activity if it counts every internal step as a separate request. LangChain's LangSmith documentation distinguishes a trace, which groups work for one operation, from the individual runs recorded within it.
Its email-assistant example combines a tree filter, selecting traces that reached a particular decision step, with a root-only filter so internal steps do not inflate the count. The resulting chart can track how often the assistant chose to reply, notify the user or take no action. Choosing the unit being counted is what makes those workflow statistics interpretable.
- AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.
“Metadata and tags are not propagated from parent to child runs, or vice versa. If you want to see both trace charts and LLM call charts grouped by a metadata key, both traces (root runs) and LLM runs need that metadata attached .”
LangChain
“Tracing projects : add one or multiple as needed per chart. Metrics are computed by pooling runs across every selected project into a single set, not shown per project.”
LangChain
“Group by and multi-metric are mutually exclusive on a single chart. Only one group-by attribute is allowed. Donut, ranked bar, and table charts do not support multiple data series (extras are dropped or blocked).”
LangChain
“Cloning a prebuilt dashboard converts its charts into fully editable custom charts.”
LangChain
Checking sign-in…
Loading comments…





