Improve Evaluators with AI
Improve evaluators using AI by providing natural language feedback.
Authorizations
OAuth access token issued by Cekura for connected apps.
Body
Clarify what a batch of evaluator scenarios should cover before generating them.
Improving existing evaluators is not done here — the Cekura agent reads the
evaluators attached to its turn and writes changes back with the ordinary
scenario write endpoints. See test_framework.scenario_chat_context.
What the generated evaluators should cover (e.g., 'cover refund disputes and angry callers'). Do not use for agent prompt/config improvements.
Agent the generated evaluators are for. Required on the first turn; on a follow-up it may be omitted and is read off session_id.
Project ID - optional, automatically inferred from evaluators if not provided
Session ID for tracking - automatically created if not provided
Number of scenarios the clarification should be sized for
x >= 1Knowledge Base file IDs for additional context
Tools that will be enabled on the generated evaluators (validated up front)
Inbound numbers that will be attached to the generated evaluators (validated up front)