bb
Try the demo

Playwright grid · flake analytics

Tame the flaky test.

Self-hosted Playwright grid with flake analytics built in.

Try the live demo View on GitHub

MIT licensed · One Postgres, one container · Runs in your VPC

rollup key (testId, branch, env, target)
flake rule same commit, both outcomes
deploy one container, one Postgres

How it works

One reporter. One Postgres. Your tests stay yours.

  1. 01

    Drop in the reporter

    Add `@brittlehq/playwright-reporter` to your config. Run tests like you do today.

  2. 02

    Run tests anywhere

    CI, laptop, container. Anywhere Playwright runs. No agent, no daemon.

  3. 03

    Triage from the dashboard

    Open the run, click the failing env. Flakes and regressions are tagged inline.

~/app · pnpm test

$ pnpm test

Running 24 tests using 3 workers

  • ✓ search box accepts query input (1.2s)
  • ✓ article infobox renders metadata (0.9s)
  • ✗ search results echo query [webkit] (3.1s)
  • ✓ history page renders revisions (1.4s)

[@brittlehq/playwright-reporter]

→ uploaded 24 sessions · 1 failed

→ brittle.dev/p/web/runs/rh-9j2

1 failed · 23 passed (12.4s)

What's different

Triage that matches how flaky tests actually fail.

brittle.dev/p/web/runs/rh-9j2
Failure analysis 3 failing · main

tests/search

  • search results echoes query webkit REGRESSED
  • autocomplete dropdown appears chromium KNOWN

tests/visual

  • article first-paint chrome firefox NEW

per-env aggregation

See the failing browser, not just the test.

TestCase rows are keyed per env, so a WebKit-only failure gets its own row.

brittle.dev/p/web/sessions/9j2-rh

search results echoes query

chromium-130 · tests/search/search.spec.ts:36

Failed

Video

ready

Trace

ready

Screenshot

ready

Timeline

  • 00:00.0 page.goto · /search
  • 00:01.2 expect.toHaveText · "results"
  • 00:03.4 TimeoutError · locator not visible

grid runtime

Every session, replayable.

Video, HAR, and trace come back with every session.

brittle.dev/p/web/sessions/9j2-rh/test-results

search results echoes query

Failed 3.1s
00:01.4
03:08
  • page.goto · /wiki/Albert_Einstein 0.8s
  • expect.toBeVisible · h1 0.3s
  • expect.toHaveText · "Einstein" 2.0s

test replay

Watch the failure happen.

Scrub the recording, jump to the failing step, open the trace.

brittle.dev/p/web/tests/wiki-search
article infobox renders metadata 3 envs · main

chromium-x/desktop

STABLE

webkit-x/desktop

FLAKY

firefox-x/desktop

STABLE

per-env signals

One test, every browser, every classification.

Small multiples per env. Stable on chromium and flaky on webkit reads as exactly that.

Integration

Four lines.
The rest is your test code.

Sits alongside your existing reporters. Forwards results, traces, and screenshots.

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

export default defineConfig({
  testDir: './tests',

  // drop in the brittle reporter
  reporter: [
    ['@brittlehq/playwright-reporter', {
      hubUrl: process.env.BRITTLE_HUB_URL,
      token: process.env.BRITTLE_TOKEN,
    }],
  ],

  projects: [
    { name: 'chromium' },
    { name: 'firefox'  },
    { name: 'webkit'   },
  ],
});

Open source · MIT

Run it on brittle.dev, or run it yourself.

Hub, dashboard, reporter, node executors. All MIT-licensed.

brittlehq/brittle on GitHub Self-host guide