Skip to main content
POST
Start an eval run

Authorizations

Authorization
string
header
required

Dashboard JWT token from Clerk

Path Parameters

agentId
string
required

Body

application/json
dataset_id
string
required
Example:

"evalDs_2Nk9pQ"

rubric_name
enum<string>
required
Available options:
correctness,
helpfulness,
safety,
groundedness,
custom
Example:

"correctness"

rubric_prompt
string

Required when rubric_name is custom.

Example:

"Score 0-100 on factual accuracy against the expected answer."

judge_model
string

Claude model id for the judge. Defaults to the agent's judge model.

Example:

"claude-haiku-4-5-20251001"

threshold
integer
default:70
Required range: 0 <= x <= 100
Example:

70

Response

202 - application/json

The eval run was accepted and enqueued for scoring.

data
object
meta
object