Docsprefactor.aiGet started, free

Your agent passed every test. Then it met a customer.

Prefactor Evals is the free hub for evaluating AI agents. Learn to build your evaluation framework, copy evals from a library of 297, and see what running them costs.

Start the guideGet five evals for your agent
In testing, all twelve golden cases pass. In the first live week, 31 runs fail in one category, over_eager_cancel.In testingWith customers, week oneships12 golden cases100%of the golden set passes.day 1112434445464731wrong runs. One category: over_eager_cancel

Learn it. Copy it. Cost it.

A guide to the framework, a library of evals, and a calculator for live volumes. All free.

  1. Decide what good means
  2. Build the golden set
  3. Heuristics, judges and people
  4. Put the checks into one framework
  5. Evaluate every live run
  6. Every change is a version
  7. What goes wrong most often

Learn the framework

Seven short modules. Leave with an evaluation framework for your own agent and the evals to hand to engineering.

Start the guide

Copy the evals

297 evals from Prefactor, the community and open source. Search by what you want to check, copy into Prefactor.

Browse the library

Know what it costs to run

Your run volume in, a monthly number out, and what rule checks save.

Open the calculator

Get five evals for your agent

Two questions. Five evals matched from the library, ready to copy. No account.

Run them in production.

Prefactor runs your evals on every live run and shows every failure by category.

Set up Prefactor