Mar 30
Loading video player...
Build a full observability stack for your LangGraph agents using Langfuse for trace-level inspection and OpenTelemetry for infrastructure metrics with real-time Grafana alerts. Never find out your agent failed from a user complaint or a surprise invoice again. What you'll learn: Instrument LangGraph nodes with a single Python decorator that captures both Langfuse traces and OTel metrics Track token spend, latency, and error rates per agent node in production Set up a Grafana dashboard with real-time alerts that fire when latency or errors spike Correlate a Grafana alert to the exact Langfuse trace that caused it using a shared trace ID The video starts with the problem — agents failing silently in production with no visibility into cost or performance. From there we build the observability decorator, wire up the OTel collector and Prometheus pipeline, and deploy a pre-configured Grafana dashboard. By the end, you'll have agent tracing, infrastructure monitoring, and anomaly alerting running on a two-node LangGraph agent. [GitHub Repo] https://github.com/ByteBuilderLabs/AI-Demos/tree/main/agent_observability [Langfuse Docs] https://langfuse.com/docs [OpenTelemetry Python] https://opentelemetry.io/docs/languages/python/ [Grafana] https://grafana.com/docs/ Subscribe to ByteBuilder for weekly AI engineering tutorials — from concept to working code. [GitHub] https://github.com/ByteBuilderLabs #ai #aiagents #langchain #langgraph #python #coding