Track AI model executions with LangFuse observability for better performance insights

About this template This template is to demonstrate how to trace the observations per execution ID in Langfuse via ingestion API.

Good to know Endpoint: https://cloud.langfuse.com/api/public/ingestion Auth is a Generic Credential Type with a Basic Auth: username = youpublickey, password = yoursecretkey.

How it works Trigger: the workflow is executed by another workflow after an AI run finishes (input parameter executionid).

Remove duplicates Ensures we only process each executionid once (optional but recommended).

Wait to get execution data Delay (60-80 secs) so totals and per-step metrics are available.

Get execution Fetches workflow metadata and token totals.

Code: structure execution data Normalizes your run into an array of perModelRuns with model, tokens, latency, and text previews.

Split Out → Loop Over Items Iterates each run step.