Skip to content

Langfuse

AgentV does not export completed eval transcripts or run bundles to Langfuse. Keep AgentV run artifacts as the source of truth for scoring, transcripts, metrics, and summaries. If you need Langfuse trace inspection, instrument the system under test or provider wrapper to emit OpenTelemetry/OpenInference spans directly to Langfuse during execution.

When the external run produces a Langfuse trace ID or UI URL, attach it to the AgentV result metadata as safe external_trace metadata:

{
"external_trace": {
"provider": "langfuse",
"trace_id": "trace-123",
"ui_url": "https://cloud.langfuse.com/project/example/traces/trace-123"
}
}

Do not store API keys, authorization headers, cookies, or raw private payloads in external_trace. AgentV may use the safe URL as a link-out; Langfuse remains the external trace viewer and AgentV remains the eval artifact owner.