Chapter

Splunk Agent Observability

See inside your agentic AI applications. Instrument them, trace and evaluate agent behavior, surface emerging issues, and apply runtime guardrails with Splunk Agent Observability (powered by Galileo).

2 hours Authors Denis Abrantes, Derek Mitchell, Sam Goldfield, & Tim Hard

Agentic AI applications think, plan, and act on their own. That autonomy is exactly what makes them powerful, and exactly what makes them hard to trust. When an agent reasons its way to the wrong answer, calls the wrong tool, hallucinates, or leaks sensitive data, where do you even look?

Splunk Agent Observability, powered by Galileo, closes that gap. It gives you observability across the entire agent lifecycle, from development through production, so you can see what your agents actually did, measure how well they did it, catch failures before your users do, and enforce safe behavior at runtime.

The scenario.

Careful Health Provider is launching an agentic AI healthcare assistant that answers patient questions about medications and looks up patient records. They already have strong observability for infrastructure and application performance, but agentic systems introduce a new class of risk: unpredictable reasoning, hallucinations, cost spikes, and exposure of sensitive data. One bad answer, telling a patient to take double their prescribed dose, is all it takes to make headlines. In this workshop, you help Careful Health Provider get ahead of that risk.

In this hands-on workshop you’ll take a working healthcare assistant, built with Streamlit, LangGraph, and PostgreSQL/pgvector, and progressively make it observable, measurable, and safe.

Splunk Agent Observability, powered by Galileo

Splunk Agent Observability brings observability to the agent lifecycle, from development through production:

  • Accurate evaluations: evaluate agent, output, and RAG quality with high-accuracy evals and both out-of-the-box and custom metrics.
  • Instant visibility: see the root cause of errors across complex, multi-step agent workflows.
  • Runtime guardrails: block hallucinations, prompt injections, and safety violations before they reach your users.

Primary references