Create Eval
Create an eval for an agent.
Evals are the criteria used to grade calls after analysis. scoring_type is
either PASS_FAIL for a boolean check or SCORE_1_5 for a rating, and
rule_logic describes what to evaluate.
An agent needs at least one PASS_FAIL eval before it can run simulations.
Authorizations
Use Tuner API key (tr_api_...) or user session token. Find your API key in Workspace Settings > API Keys.
Path Parameters
Tuner's internal numeric ID for the agent.
Workspace ID. Find this in Workspace > General Settings.
Body
Input schema for creating/updating an eval.
1 - 200PASS_FAIL for a yes/no check, SCORE_1_5 for a 1–5 rating. Simulations require at least one PASS_FAIL eval.
^(PASS_FAIL|SCORE_1_5)$Instruction the AI applies to each call. For PASS_FAIL, phrase it so that passing is unambiguous.
1x >= 0Response
Successful Response
Response schema for an eval.
Unique identifier for the eval.
12
Name of the eval.
"Verified caller identity"
PASS_FAIL for a yes/no check, SCORE_1_5 for a 1–5 rating. Simulations require at least one PASS_FAIL eval.
PASS_FAIL, SCORE_1_5 "PASS_FAIL"
Instruction the AI applies to each call. For PASS_FAIL, phrasing is written so that passing is unambiguous.
"The agent confirmed the caller's name and date of birth."
Whether this eval was added automatically as a system default rather than created by a user.
false
Whether this eval can be deleted. System defaults may be marked non-removable; attempting to delete a non-removable eval returns a 403.
true
Display order relative to other evals for the agent.
0
The inputs the AI considers when evaluating this eval, e.g. "system_prompt", "tools", "allowed_capabilities", "workflow", or "metadata." for a specific metadata field.