S

Agent and Model Evaluation

Application

by Splunk

Production scoring for agent traffic: every interaction graded for correctness, safety and tool use rather than a sample of them, with the evaluations built for agents instead of adapted from model benchmarks.

Find alternatives to Agent and Model Evaluation

What it does

• Evaluation and observability
• Guardrails and safety

Features 3

• Full-traffic scoring with LunaLuna, a small model, scores correctness, safety and tool use on 100% of production traffic.
• Autotune evaluatorsEvaluators adapt to a team's own criteria instead of generic out-of-the-box metrics.
• Quality next to latency and costEvaluation scores sit on the same trace as latency, errors and token cost.

Sources

Pricing

Not found · 19 Sep 2026

Read 2026-09-19: the page states no rate, plan or allowance. Splunk's own pricing page sells workload and ingest packages for the platform and names no line item for this product.

Description

Official documentation · 19 Sep 2026

Splunk's own page description, read 2026-09-19: "Score the correctness, safety, and tool use of every agent interaction in production, not a sample, with evaluations built for AI agents."

What it does

Official documentation · 19 Sep 2026

Scoring production agent traffic is evaluation-observability; scoring specifically for safety is guardrails-safety.

Sold within

Not found · 19 Sep 2026

This product's own page title files it under Observability Cloud ("Evaluate Agents and Models Observability Cloud | Splunk"), so the parent is Splunk Observability Cloud -- an existing record. Splunk states no entitlement sentence for it: no page says whether it is included in an Observability Cloud subscription or metered on top, so the access key is left absent rather than defaulted.

Name

Official documentation · 19 Sep 2026

Page <h1> "Agent and model evaluation" under the <title> "Evaluate Agents and Models Observability Cloud | Splunk"; Splunk's A-Z product index lists it as "Agent and Model Evaluation".

URL

Official documentation · 19 Sep 2026

The product's own page in Splunk's A-Z product index, read on 2026-09-19.

Deployment

Official documentation · 19 Sep 2026

Delivered through Splunk's hosted platform; the page names no self-hosted edition of its own.

Also from Splunk

• Splunk AI Assistant Platform

  Assistant inside Splunk Cloud Platform that turns plain-English questions into SPL, the query language Splunk searches machine data with, and explains existing searches back to an analyst.

• Splunk AI Toolkit Developer tool

  Platform for building, deploying and running agentic AI systems on Splunk data.

• Splunk Enterprise Security Application

  Integrated threat detection, investigation and response platform with agentic AI, SOAR, UEBA and SIEM unified into one SecOps experience.

• Splunk IT Service Intelligence Application

  AIOps and service-intelligence solution that uses AI-driven field discovery and AI-generated event correlation, summaries and root-cause guidance to connect IT events to business service health.

• Splunk Observability Cloud Application

  Observability platform where AI agents correlate signals across domains, propose remediation actions such as rollbacks or capacity changes, and provide dedicated agent/AI-system observability alongside infrastructure monitoring.

• AI SRE Application

  Splunk's site-reliability agent: an AI teammate for IT and SRE teams that troubleshoots an incident, finds the root cause and proposes the fix, working from the telemetry already in Splunk rather than from a human's search.

• Splunk Agent Launchpad Agent platform

  No-code builder and launcher for AI agents inside Splunk: security, IT and network teams create an agent and start it from the workflow where the signal appeared, rather than from a separate console.

• Splunk Agent Observability Platform

  Observability for somebody else's AI stack: Splunk traces a customer's own agents, the models they call, the vector databases behind them and the infrastructure underneath, and reports where the stack is failing.

• Agent Visibility Application

  The tracing half of Splunk's agent observability: one agent followed end to end from its own reasoning steps down to the GPU the run executed on, so a failure can be attributed to a layer rather than guessed at.

• Runtime AI Guardrails Application

  Splunk's enforcement layer for agents: a guardrail built from a customer's own evaluations blocks an unsafe agent action before the tool call executes, and logs every allow-or-block decision for audit.

• Detection Studio Developer tool

  The authoring environment for Splunk Enterprise Security's detections: a detection engineer plans, develops, tests, deploys and then monitors a detection in one place, instead of moving it between a notebook, a ticket and production.

Open Splunk in the directory

Something wrong here? Send a correction — quote this product id: splunk-agent-and-model-evaluation.