Skip to content

Promptfoo

Promptfoo runs repeatable tests on prompts and compares the responses against checks you choose.

Visit Promptfoo

What can you check before changing a prompt?

Write two versions of your instructions and list a few inputs both versions should handle. Promptfoo can load the instructions from files and run the same cases against both. A check can look for a required word or compare an answer to expected text. Read the answers too. Passing one check only confirms that one requirement.

Does running the test locally keep the examples on your computer?

Promptfoo runs the evaluation on your computer in a local setup. A hosted model you select can still receive the prompts and test inputs and charge for use. Check the model and any automated scoring or sharing settings before testing. Use made-up examples and leave credentials out of the test files.

What can you keep after testing?

The open-source Community version covers local evaluation. Keep your prompt files, sample inputs and checks. Export results to JSON or CSV if you need them elsewhere. Inspect those files before sharing because they can contain the prompt, inputs and model answers. Managed team sharing and access controls are listed under Enterprise.

Can you tell why one answer passed and another failed?

On paper, use a fictional event with a confirmed date and no venue. Write two possible answers. Keep models, sharing and provider accounts disconnected.

  1. Choose what to check
    Both answers must keep the supplied date. The one that invents a venue fails even if it includes the correct date.
  2. Keep a repeatable example
    Write the event input, both answers and the pass rule in one sheet. Use the same example to judge a later wording change.