Recipe

LLM observability stack setup

Ship a production-grade tracing, logging, and evaluation pipeline for your LLM-powered features in under an hour.

Tracing

OpenTelemetry auto-instrumentation for every chain, tool call, and retrieval step. Export to your existing collector.

Logging

Structured JSON logs with trace correlation IDs. Ship to your warehouse via vector or direct sink.

Evaluation

Offline eval harness with golden datasets. Score faithfulness, relevance, and latency on every deploy.

Stack components

OpenTelemetry SDK
Langfuse / Braintrust
Pydantic Logfire
Ragas / DeepEval