Instrument the Application

10 minutes

You can’t observe what you don’t capture. In this chapter you’ll add Splunk Agent Observability (Galileo) tracing to the assistant so that every user turn becomes a trace, with a nested span for each LLM and tool call, and you’ll do it without rewriting the agent.

Persona

As Careful Health Provider’s AI engineer, you want end-to-end visibility into the agent’s decisions with minimal code change and minimal maintenance. Rather than hand-instrument every step, you’ll attach a single LangChain callback at the graph level and let Splunk Agent Observability capture the whole tree automatically.

Instrumentation is remarkably lightweight: a Splunk Agent Observability callback is a standard

LangChain callback handler. Attach it to a LangGraph run and it captures prompts, responses, model names, token usage, timing, and span nesting for you.

Where to work

This chapter works in the ~/workshop/healthcare-assistant/2-app-with-instrumentation folder.

Continue to the subsections to add the SDK, attach the callback, and generate traffic.