Create Conversation Eval

Create the built-in Conversation Eval that uses the real reply engine. Use this operation when you need an actual scripted dialogue, not the static authored-config check at POST /characters/{slug}/evals. It scores deterministic assertions (forbidden-leakage / memory-recall / per-turn contains) plus LLM-judge rubrics. Async: returns an eval_id; poll get_conversation_eval for the terminal result, transcript, and scored report.

Authentication

AuthorizationBearer

LoreOS runtime API key (ck_...), sent as Authorization: Bearer <key>.

Path parameters

slugstringRequired

Request

This endpoint expects an object.
namestringRequired1-200 characters
turnslist of objectsRequired
seed_contextlist of objects or nullOptional

Context events planted before turn 1 (e.g. a fact to test leak).

assertionsobjectOptional
reply_modeenumOptionalDefaults to deep

Response

Successful Response
schema_version"v0"

LoreOS success-envelope version.

dataobject
next_actionslist of objects

Concrete follow-up API actions; empty when no action is required.

Errors

422
Unprocessable Entity Error