Splunk Agent Observability Instrumentation for LangChain Apps
Wrap-up
You took the workshop 18 multi-agent travel planner and instrumented it with Splunk Agent Observability by adding just
two things: galileo_context.init(...) and a single GalileoCallback on the LangGraph run config.
With that, every agent node’s LLM call now appears as a nested span in a single Splunk Agent Observability trace per
request — no per-node changes required and very low maintenance.
You now have the same workload traced in two observability tools (Splunk Observability Cloud from workshop 18, and Splunk Agent Observability here), which is a useful basis for comparison. What does Agent Observability show you that Observability Cloud doesn’t, and vice versa?
Next, we’re going to expand on this by:
- Adding Splunk Agent Observability metrics (for example,
Context Adherence) to the captured traces. - Working through the features that help Splunk Agent Observability support better observability of agents.
- Leveraging powerful features like Signals in AI Assistant.
- Using a dedicated
GalileoLogger(project=..., log_stream=...)to route specific runs to different log streams. - Adding more complexity to our agents.