The best way to do AI evals

Get the right output,
first time, every time

Stop tweaking system prompts on vibes. eval.dog runs every version of your prompt against your examples and your checks, so you know what actually improved.

Free plan included. No credit card required.

Build AI systems you can trust

Everything here is something you do: write a check, read an output, disagree with the judge, change the prompt.

Versioned prompts

Every edit becomes a version you can run again. Any result traces back to the exact text that produced it.

Checks in your own words

Ask one yes or no question, the way you would ask a colleague. "Does it say when the refund lands?" is a check.

Runs you can watch

Every example goes to your prompt, then every check is answered about the output it produced. Answers land as they happen.

Evidence, not a score

A judge that answers no quotes the line it read and says why it read it that way. You can disagree with it.

Version comparison

How often each check answered yes, version by version, side by side. That is how you know what your edit did.

Regressions first

A check that answered yes last run and no this one goes to the top of the table, marked. It is the first thing you read.

How it works

From vibes to verified in four steps.

  1. 1

    Paste your system prompt

    Create a prompt, pick the model it runs on, and you are versioned from the first save.

  2. 2

    Add examples

    Type the inputs your prompt has to handle, or import a CSV. Bring the outputs you already have with them.

  3. 3

    Write a check

    One yes or no question in your own words. A judge answers it, or code does when the answer is exact.

  4. 4

    Run it, then change something

    Every check is answered about every output, with the evidence quoted. Edit the prompt, run again, compare the versions.

Simple, honest pricing

A call is one output generated, or one check answered about it. Start free, upgrade when your evals do.

Free

Kick the tires. No card required.

$0 / month

  • 2 prompts
  • 200 calls a month
  • Unlimited examples and checks
  • Version comparison
Start free

Pro

Most popular

For builders shipping prompts to production.

$29 / month

  • Unlimited prompts
  • 3,000 calls a month
  • Unlimited examples and checks
  • Version comparison
  • CSV import of examples and outputs
Go Pro

Your prompts deserve better than vibes

One prompt, three examples, one check. That is enough to find out what your prompt gets wrong.

Get started free
eval.dog — The best way to do AI evals