// tia engine

Only run the tests that matter

Test Impact Analysis maps your whole suite to the code each test depends on, so a change to one file re-runs only the tests it can break. A 15-second suite comes back in under a second, fast enough for an agent to run it after every edit.

FAIL  Tests\MathTest

× it adds numbers

---

FAILED  Tests\MathTest > adds numbers

Failed asserting that 2 is identical to 3.

at tests/MathTest.php:4

 1<?php

 2

 3it('adds numbers', function () {

→4    expect(1 + 1)->toBe(3);

 5});

Tests:    1 failed, 773 passed (2658 assertions)

Duration: 4.18s

// readable failures

Failures that explain themselves

A failed test prints the assertion, the diff and the exact line of code in your terminal. Most of the time you know the fix before you open the editor.

// all-in-one framework

Everything you need to test your app

Parallel execution, coverage, mutation and snapshots ship in the box. You install Pest once and stop assembling a toolchain.

it('may login', function () {

visit('/login')

->type('email', 'hi@example.com')

->type('password', 'password')

->press('Continue')

->assertSee('Dashboard')

->assertNoJavascriptErrors();

});

// modern browser testing

Run browser tests with Pest

Browser tests are plain Pest tests. Playwright drives a real browser while you keep the syntax and the suite you already have.

$ pest --agent='visit("/checkout")->assertNoJavascriptErrors();'

✓ verify — page loaded · 0 js errors · 0 console warnings

agent → checking the fix on mobile…

$ pest --agent='visit("/checkout")->on()->mobile()->screenshot(filename: "checkout");'

✓ verify — layout stable · screenshot saved to .pest/screenshots

agent → verifying the backend too…

$ pest --agent='$order = \App\Models\Order::factory()->create(); expect($order->total)->toBeGreaterThan(0);'

✓ verify — 1 passed (1 assertion)

// agent browser

Your agents verify their own work

A coding agent can check its own change against your live app with one short CLI command: visit the page, assert on what rendered, save a screenshot as proof. Each probe runs as an isolated, throwaway test, so nothing extra lands in your suite.

it('answers support questions', function () {

expect(SupportAgent::class)

->prompt('How do I reset my password?')

->repeat(50)

->toBeCorrect()

->toBeSafe(0.9)

->toSatisfy('Gives the password reset steps');

});

// pest --evals — 50 samples 96%

// ai evals

Eval your LLM features like code

Evals are ordinary Pest tests. Assert exact strings where the answer is fixed, and let AI scorers grade relevance or safety where it is not. When a prompt regresses, CI fails.