ConeReplay and LangSmith.
This is not a scored bake-off. ConeReplay has not published a head-to-head quality, latency, or token number against LangSmith. LangSmith is a separate product. Use its documentation for tracing and evals in the LangChain stack.
Where ConeReplay fits a LangGraph app
The SDK extra conereplay[langgraph] records a LangGraph app with attach_to_langgraph_app. Replay then checks a proposed change against that corpus. Observability of the live run and a pre-deploy regression check are different jobs. ConeReplay is the regression check.
Read how cone-bounded replay works