qareef drives your product in a live browser, reads the screen the way a tester does, records every action as a flow, and turns it into real Playwright tests that run on every change — with screenshots and traces to prove it.
Free while in early access · no card · nothing to install on your site.
Waterfront rooms above the old harbour
Quiet courtyard stay near Alfama
Garden rooms with reef-to-table dining
await page.getByRole('button',
{ name: 'Book now' }).click();
await expect(page
.getByTestId('confirmation'))
.toContainText('TIDE-');
Every flow goes through the same five stops. Nothing is a black box — open the generated test, read the trace, and see exactly which model made which click.
Drive your app in a live, recorded browser — or describe the goal in plain English and let the operator drive: OCR reads the screen, a cheap model picks the next visible label.
Steps land in a timeline you can drag to reorder or delete before saving. A flow is a named, replayable recording of what just happened.
An Automate job turns the flow into real Playwright test code, with assertions proposed by a model (stamped decided_by) that you review in the built-in IDE.
Runs execute against an environment and keep a screenshot and trace per run. Verdicts are PASS, FAIL, or UNMEASURED — never a forced false fail.
A failing run auto-files a bug with the evidence attached. Flows whose last run failed surface in Maintenance so nothing quietly rots.
Point qareef at a URL. It navigates your app like a new teammate, discovers the workflows, and drafts a test outline you approve before a single test is written.
The agent is exploring Tidepool Hotels (a fictional demo app) in a real browser, discovering ~57 workflows and grouping them into a test outline. You pick the flows; qareef turns each one into deterministic Playwright code.
If a person can do it in a browser, a flow can do it on request. Two things people ask us for that a test tool wasn't built for — and this one was.
Record the flow once — search, log in, pull the record — and parameterise the inputs. It becomes plain Playwright that runs on request or on a schedule and hands back what it read, plus the screenshot and trace that prove where it came from.
{account} or {date} become request parametersCounty portals, ERPs behind a login, vendor back-offices that were last redesigned a decade ago. There's no integration to buy — but there is a screen, and qareef reads screens. Runs on any browser surface: Chromium, Chrome, or Camoufox with a persistent, realistic profile.
decided_by with its cost — nothing anonymousNo CAPTCHA solving. Ever.
If a site challenges the agent, the run stops, records the screen, and hands it to a person. qareef automates the work you're already allowed to do in a browser — it doesn't defeat access controls.
Canvases, maps, third-party payment frames, 2FA prompts, file pickers — anywhere the DOM churns but a human eye still knows exactly what to click.
Drawing zones on a canvas, verified by what's on screen
“We run the smoke suite on every pull request now. The agent draws, drags, and pays like a real user — and hands us plain Playwright when it's done.”
— What we're building toward. Sample quote; replaced when a design partner is quotable.
See the productDeterministic Playwright handles the hardest paths — canvases, iframes, drag-and-drop — so runs are reproducible, not improvised.
Generated tests execute with no model in the loop. Your suite runs in CI at Playwright speed, at Playwright cost: zero tokens.
Everything qareef writes is plain Playwright you can read, edit, and run anywhere — your repo, your CI, your rules.
Parallel by default: generated tests are plain Playwright specs, so they shard across workers and CI runners like any other suite. These are product facts, not customer metrics — we don't have those yet and won't invent them.
qareef plugs into the way you already ship. Every pull request gets a smoke pass; every merge to staging gets the full suite. When something breaks, the agent files the bug with a trace and screenshot — and when a test is flaky rather than your app, it surfaces in Maintenance with the generated code one click away.
You review the evidence, not the noise.
Failures arrive with a trace, screenshots, and the generated test open in the IDE — flaky tests get repaired, real bugs get filed.
“The first mapping session found a checkout path nobody on the team had a test for.”
“UNMEASURED saved us from chasing a ‘failure’ that was just the staging box being asleep.”
“We pointed it at a county portal with no API and had a scheduled pull running the same afternoon.”
No. A mapping session records real clicks, types, drags and navigations as you drive your app, or you describe a goal in plain English and the operator reads the screen with OCR and picks the next visible label. Either way, the flow becomes real Playwright test code you can read and edit — selectors included — so you're never stuck depending on OCR at test time.
When the agent decides the next click instead of you, that decision is logged with the exact model that made it (for example decided_by: deepseek/deepseek-chat), plus its token cost. Nothing the agent decides is anonymous, and run cost is visible, not billed as an opaque "AI credit."
Most runners only have pass and fail, so anything a check can't actually observe — a page that never loaded, a drifted viewport, a flaky network hop — gets forced into one of those two boxes, usually FAIL. qareef's runs carry a third state, UNMEASURED, for exactly that case. More in the blog post.
Record the flow once — search, log in, pull a record — and parameterise the inputs. The flow becomes plain Playwright that runs on request or on a schedule; each run returns what it read as structured output together with the screenshot and trace that prove where it came from. No API on the target site is required. Details on the website → API page.
No, never. If a site challenges the agent, the run stops, records the screen as evidence, and hands it to a person. qareef automates the work a person is already allowed to do in a browser; it does not defeat access controls. More on the legacy system automation page.
The core — flow runtime, engine adapter, OCR-driven operator, OCR/pixel measurement, the LLM adapter with cost logging, and evidence-first document storage — is being built in the open under packages/core, MIT-licensed. The hosted app (Flows, Runs, Bugs, Requests, Maintenance, the IDE, Automate jobs) is not part of that core.
Node 22, Playwright, and either a Mac (for on-device Apple Vision OCR) or a 27B-class vision model reachable over the network — plus an OpenRouter key for the cheap text model that decides what to click. Full list on the docs page.
Most closed "coverage as a service" platforms pair a managed cloud runner with dedicated human QA engineers. qareef looks similar at the surface — flows, runs, bugs, maintenance — but its automation core is open source, it uses an OCR-reading operator instead of requiring stable DOM selectors up front, and every model decision is cost-logged and attributed. It does not (yet) offer a managed human-verification service. Full, sourced comparison on the compare page.
Leave an email and we'll reach out when there's a seat — or skip the form and talk to the people building it.