Evaluation sets contain cases used to check agent behavior. An evaluation runs those cases again and records metric results. This page indexes the Agent API Server development evaluation endpointsThese /dev/... endpoints require the development UI and evaluation dependencies; see Agent API Server overview. Harness deployments use their corresponding evaluation endpoints
Evaluation can call models and tools, incur charges, or perform external actions. Review user data, tool permissions, and metric settings when preparing cases
Create a set, add cases from existing sessions, and run an evaluation with selected metrics. Read, replace, or delete individual cases as needed. Check per-case status and metric results to determine whether the evaluation passed
These endpoints remain available for existing clients. Check each page for request and response differences before migration; changing only the URL can be insufficient. Listing evaluation cases currently uses only the underscore path in the table above
