Skip to main content

Intro

Run an evaluator over a single input and get its score back, or run it as a guardrail and gate on one boolean. List the built-in evaluators to see which ids you can address and what each one needs.

Addressing an evaluator

The {evaluator} segment takes any of three things:
  • a built-in id, such as ragas/faithfulness. Two-segment ids use the two-segment form of the path.
  • the slug of a monitor configured in your project, which brings that monitor’s saved settings with it
  • evaluators/{slug|id} for an evaluator you saved, including workflow and code evaluators
GET /api/evaluations/list returns the built-in ids along with the data fields each one requires. For what an individual evaluator scores and which settings it accepts, see its page under Built-in Evaluators.

Guardrails

POST /api/guardrails/{evaluator}/evaluate is the same call in gating mode. Every outcome carries passed, including a skip and an evaluator error, so a request you are guarding is never blocked by the guard itself failing. Check passed and let the request through when it is true.
Last modified on August 7, 2026