Works with your existing Playwright tests

AI-Powered Playwright.
Zero Rewrites.

Sentinel adds natural language actions, self-healing locators, and autonomous agents to Playwright — without replacing it. One import, one extra method, your tests stay intact.

npm install @isoldex/sentinel Read the docs →

The problem with selectors

Every Playwright test is coupled to the DOM. When the UI changes — and it always does — tests break. Sentinel decouples your intent from the implementation.

Before — Plain Playwright

test.spec.ts

// Plain Playwright — fragile selectors
await page.click('#email-input');
await page.fill('#email-input', 'user@example.com');
await page.click('button[type="submit"]');

// Every UI change breaks your tests.
// Find the new selector, update, repeat.

After — Sentinel

test.spec.ts

// Sentinel — natural language, self-healing
await ai.act('fill the email field with user@example.com');
await ai.act('click the sign in button');

// Sentinel finds the element itself.
// When the UI changes, it re-discovers it automatically.

Three ways to adopt

Extend any page

sentinel.extend(page) adds AI methods to an existing Playwright Page. Zero test restructuring.

ai test fixture

import { test } from '@isoldex/sentinel/test' — the ai fixture works alongside page and browser.

Autonomous agent

sentinel.run(goal) takes over and completes multi-step flows without step-by-step scripting.

Path 1 — extend(page)

script.ts

import { chromium } from 'playwright';
import { Sentinel } from '@isoldex/sentinel';

const browser = await chromium.launch();
const page    = await browser.newPage();

// One line to add AI to an existing page
const ai = await Sentinel.extend(page, {
  apiKey: process.env.GEMINI_API_KEY,
});

await page.goto('https://myapp.com');

// Mix AI actions with standard Playwright
await ai.act('click the login button');
await page.waitForURL('/dashboard');          // standard Playwright
await expect(page.locator('h1')).toBeVisible(); // standard Playwright
const data = await ai.extract('get user name', { name: 'string' });

Path 2 — Playwright Test fixture

playwright.config.ts

// playwright.config.ts
import { defineConfig } from '@playwright/test';

export default defineConfig({
  use: {
    sentinelOptions: {
      apiKey: process.env.GEMINI_API_KEY,
      enableCaching: true,   // near-zero cost on repeated runs
      headless: true,
    },
  },
});

search.spec.ts

// search.spec.ts
import { test, expect } from '@isoldex/sentinel/test';

test('search and extract results', async ({ page, ai }) => {
  await page.goto('https://news.ycombinator.com');

  // AI action
  const stories = await ai.extract(
    'Get the top 5 story titles and their point counts',
    { stories: [{ title: 'string', points: 'number' }] }
  );

  expect(stories.stories).toHaveLength(5);
  expect(stories.stories[0].title).toBeTruthy();
});

Path 3 — Autonomous agent

agent.ts

// Let the agent figure out the whole flow
const result = await sentinel.run(
  'Log in with credentials from env, navigate to settings, ' +
  'enable 2FA, and extract the backup codes',
  { maxSteps: 30 }
);

console.log(result.goalAchieved); // true
console.log(result.data);         // { backupCodes: [...] }

Plain Playwright vs. Sentinel

Sentinel is additive — it doesn't replace Playwright, it enhances it.

Capability	Plain Playwright	Playwright + Sentinel
Element targeting	CSS selectors / roles	Natural language
When UI changes	Tests break, manual fix	Self-healing — auto re-discovers
Dynamic pages	Requires knowing selectors	Observes + acts on what's visible
Shadow DOM / iframes	partial	✓
Autonomous multi-step	✗	yes (run() agent loop)
Data extraction	Requires selectors + parsing	extract() returns typed objects
Existing test migration	—	extend(page) — one line
Works with expect()	✓	yes — fully compatible
Cost per run	—	~$0.002 (Gemini Flash)
OpenTelemetry tracing	✗	✓

What you gain

🔁

Self-healing locators

Successful selectors are cached and reused. When the DOM changes, Sentinel re-discovers the element automatically — no manual fix needed.

💬

Natural language actions

Write what you want, not how to find it. act('click the checkout button') works regardless of the underlying selector.

🤖

Autonomous agent mode

run(goal) plans and executes multi-step flows without scripting each step. Handles unknown page states, dynamic content, and unexpected redirects.

📊

Typed data extraction

extract('get product name and price', schema) returns fully typed data — no selector hunting, no innerHTML parsing.

📡

OpenTelemetry tracing

Every AI action emits a span. Drop into Datadog, Grafana, or Jaeger to see exactly what Sentinel did, for how long, and at what cost.

💰

~$0.002 per run

Gemini Flash is 40× cheaper than GPT-4o. With self-healing locators and prompt caching, repeated runs cost effectively zero.

Add AI to Playwright in 5 minutes

One install. One import. Your existing tests stay exactly as they are.

Get started →Migration guide →

AI-Powered Playwright.Zero Rewrites.