Skip to main content
As your team reviews completions and traces, the insights agent surfaces patterns and groups them into actionable findings.
Review Insights panel showing identified patterns across reviewed completions

​
How they work

Review Insights workflow showing how human labels feed into AI analysis
When humans label data in Freeplay the Insights agent analyzes each reviewed item in the background. It identifies common patterns, groups related items into themes, and suggests actions based on what it finds. The inputs to Review Insights include:
  • Human labels — annotations, notes, and scores from human reviewers
  • LLM-as-a-judge evaluations — scores and reasoning from your auto-evaluators applied during review
  • Logs — the completions or traces that were evaluated

​
When they run

Review Insights run anytime a human label is added to data. Every annotation — notes, human evals, or LLM-as-a-judge evals — triggers the agent to analyze and update insights. When combined with Review Queues these review insights can point to key issues in your system.
Review Insights can be disabled in Project Settings > AI Features.
Review Insights themes are generated automatically and may occasionally be too broad or too narrow. Regularly review themes and use merge/prune actions to keep them useful.