Langfuse

Langfuse accepts OpenTelemetry traces. Authenticate with HTTP Basic auth built from your public and secret keys, passed as a ready-made header value.

workspace:
  telemetry:
    exporter: otlp
    endpoint: https://cloud.langfuse.com/api/public/otel/v1/traces
    headers: {Authorization: ${LANGFUSE_BASIC_AUTH}}

Export LANGFUSE_BASIC_AUTH as Basic <base64(public_key:secret_key)> (US region: https://us.cloud.langfuse.com/...; self-hosted: your own host).

Pecca resolves ${VAR} through its secrets connector; the value never appears in .pecca/config.resolved.yaml.

Span attributes

Every @pecca.replace invocation emits one span named pecca.call:

Attribute	Meaning
pecca.path	workspace/project/call
pecca.mode	record, shadow or live
pecca.version	model version used
pecca.served_by	llm, model or fallback
pecca.confidence	calibrated confidence
pecca.agreement	model agreed with the LLM (shadow only)
pecca.latency_ms	wall time of the call

Finding where the LLM is still used

In the Traces view, filter on span metadata:

metadata.pecca.served_by = fallback

Your wrapped LLM spans (with gen_ai.* attributes) nest under the pecca.call span, so Langfuse shows which calls the model absorbed and which still reached the LLM.

Screenshot

Placeholder: docs/assets/monitoring/langfuse.png (the maintainer adds a screenshot of the query above).

Pecca has no vendor-specific code: it speaks OTLP over HTTP. Endpoint paths and header names are the vendor's and can change, so check their current documentation. To read traces back as training data, see the otel-traces datasource.