Where should ClearRun go in an AI pipeline?

Right after the model writes the answer and before anyone acts on it. Score the final answer, with its prompt, then decide in your own code what to do with a low score.

Last updated

A simple flow

  1. Your app sends the prompt to a model.
  2. The model returns an answer.
  3. Your server calls clearrun.evaluate({ prompt, responseText }).
  4. Your code decides: show it, show it with a note, retry, or send it to a person.
  5. Store the verdictId with the answer.

Score once

Score the answer people will read. Skip middle steps of a chain. Each call uses credits.

In CI

Score saved sample answers in CI to catch prompt or model changes before release. See How to use ClearRun in CI without blocking every pull request.