Turn any old website into an API
Record the flow once, parameterise the inputs, call it like an endpoint.
- Inputs like
{account}become request parameters - Structured output — what the operator read, not a DOM scrape
- Every run keeps its evidence
Agentic QA for web apps · open-source core
QA that operates your app like a person
It drives your product in a live browser, reads the screen the way a tester does, and turns what it did into real Playwright tests — with screenshots and traces to prove it.
Free while in early access · no card · nothing to install on your site.
Built on · works with — no customer logos yet, we're pre-launch
PlaywrightChromiumChromeCamoufoxApple Vision OCROpenRouterDeepSeekGitHub
Demos
Watch the agent drive
Five things QA Reef does, playing out on a mock app. Not a video — the page is driving itself.
You drive once. It records every action and writes the Playwright.
Recorded steps
Waiting for the agent…
City
Check-in
Check-out
Guests
Search
Tidepool Marina Suites
Waterfront rooms above the old harbour
PoolWi-FiBreakfast
€182/night
Book nowCoral Court Hotel
Quiet courtyard stay near Alfama
SpaParking
€139/night
Booking confirmed — TIDE-40192
Flow outline · Book a room end-to-end
The outline builds as the agent works.
reef agent
Reading the screen…
Screenshot per step
Trace recording on
decided_by deepseek/deepseek-chat · $0.0007
The app changed and the locator broke. It finds the button by what it says on screen.
Replay
Waiting for the agent…
CityLisbon, Portugal
DatesMar 14 – 18
Search
Tidepool Marina Suites
Waterfront rooms above the old harbour
€182/night
Reserve nowLocator
#book-btn not foundTrying captured alternates…
healed:
#book-btn → role=button name="Reserve now"reef agent
Replaying the recorded flow…
Alternates captured at record time
Patch proposed, not auto-merged
decided_by deepseek/deepseek-chat · $0.0004
A 2004 portal with no usable selectors. It reads the screen and clicks what it sees.
What it saw
Waiting for the agent…
PROPERTY SEARCH
Submit Query
DOCUMENT
PERMIT APPROVED
reef agent
No stable ids on this page. Reading it instead.
Apple Vision OCR · word boxes
Screenshot kept as evidence
decided_by deepseek/deepseek-chat · $0.0011
The same flow, parameterised, called like an endpoint.
Run
Waiting for a request…
REQUEST
GET /county/lookup?parcel=Send
RESPONSE 200
reef agent
Same recorded flow, inputs swapped for request parameters.
Structured output, not a DOM scrape
Screenshot + trace attached to the reply
Plain Playwright at run time · 0 tokens
A deployment appears; the suite crawls it before anyone clicks Promote.
Checks
Waiting for a deployment…
dpl_9fQ2mKfeat/promo-code
BUILDING
//search/checkout/account/help/legal
1 finding · /checkout — promo field rejects a valid code
PromoteView evidence
reef agent
Watching for a new deployment…
Verdicts: PASS · FAIL · UNMEASURED
Bug filed with the trace attached
Illustrative mock — sample deployment
How it works
One loop: map it, save it, generate it, run it, fix it.
Five stops, nothing hidden: open the generated test, read the trace, see which model made which click.
Map
Drive your app in a recorded browser, or describe the goal and let the operator drive by reading the screen.
Save as a flow
Steps land in a timeline you can reorder or delete. A flow is a named, replayable recording.
Automate → test code
The flow becomes real Playwright, with model-proposed assertions stamped decided_by for you to review.
Run with evidence
Every run keeps a screenshot and a trace. Verdicts are PASS, FAIL or UNMEASURED — never a forced false fail.
Bugs & maintenance
A failing run files a bug with the evidence attached, and the flow surfaces in Maintenance.
Beyond QA
The same agent that tests your app can run anyone's
If a person can do it in a browser, a flow can do it on request.
Automate legacy systems with no API
County portals, ERPs behind a login, vendor back-offices. No integration to buy — but there is a screen, and QA Reef reads screens.
- Clicks by what is visible, so churning markup does not break the job
- Scheduled or on-demand, a screenshot per step for the audit trail
- Every decision stamped
decided_bywith its cost
No CAPTCHA solving. Ever. If a site challenges the agent, the run stops and hands the screen to a person.
Capabilities
Built for the UIs other tools give up on
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
Canvas apps
Whiteboards, diagram editors and floor-plan tools render into one <canvas> — nothing for a selector to grab. It finds the tool, the handle and the shape by what is on screen, draws, and checks the result visually.
Example flow
See the product
- Open the lobby plan
- Pick the Zone tool
- Draw a five-point polygon
- Name it “Quiet zone”
- Assert the zone appears in the legend and the saved plan
Panning a map, hitting the pin it can see, checking the card that opens
Maps & GIS
Slippy maps and GIS viewers redraw tiles on every pan and hide pins inside one canvas. It works from the rendered map: pans, zooms, clicks the pin it can see, and checks the popup appeared.
Example flow
See the product
- Search “Lisbon”
- Drag the map east and click “Search this area”
- Open the pin nearest the harbour
- Assert the card shows the same hotel and price as the list
Typing into a hosted card field, then approving the 3-D Secure challenge
Payments
Hosted card fields live in a third-party iframe and 3-D Secure opens another, often cross-domain. It types into the frame it can see and follows the challenge, using your provider's test cards.
Example flow
See the product
- Add a room to the cart and open checkout
- Enter the provider's test card in the hosted field
- Approve the 3-D Secure prompt
- Assert “Order confirmed” and read back the order number
Signing in, then reading the one-time code off the test inbox
Auth & 2FA
SSO, magic links and one-time codes break brittle scripts. It drives the sign-in like a person and reads the code from the test inbox or a TOTP secret you supply. It does not solve CAPTCHAs — ever.
Example flow
See the product
- Open the sign-in page and enter the test account
- Wait for the code prompt
- Read the six-digit code from the test inbox
- Type it and assert the dashboard greets the right user
Attaching a fixture file and waiting for the preview to show up
File uploads
Native pickers, drag-to-upload zones and preview thumbnails are invisible to DOM-only tools. It attaches the fixture, watches the upload finish on screen, and checks what should appear afterwards.
Example flow
See the product
- Open online check-in
- Click “browse files” and pick passport.jpg
- Wait for the thumbnail and the progress bar to finish
- Assert the “Verified” badge before continuing
The long tail: every surface where a person can still see what to click
And more
The same screen-first approach covers the long tail. If a person can do it in a browser, it can be recorded as a flow and turned into Playwright.
- PDF viewers
- Nested iframes
- WAF-guarded portals
- Legacy framesets
- Canvas charts
- Drag-and-drop boards
Example flow
See the product
- Log in to the 2004 portal through its frameset
- Export the monthly statement as a PDF
- Read the total off the rendered PDF
- Assert it matches the figure on the new dashboard
Coverage and runs, by the numbers we can stand behind
3verdict states: PASS · FAIL · UNMEASURED
2evidence files per run: screenshot + trace
0tokens at run time — no model in the loop
0CAPTCHAs solved. Ever.
100%of agent decisions stamped with model id + cost
MITlicence for the core, extraction in progress
Product facts, not customer metrics — we don't have those yet and won't invent them.
What we want to be able to quote here
“The first mapping session found a checkout path nobody on the team had a test for.”
Sample — engineering lead
“UNMEASURED saved us from chasing a ‘failure’ that was just the staging box being asleep.”
Sample — QA engineer
“We pointed it at a county portal with no API and had a scheduled pull running the same afternoon.”
Sample — operations
FAQ
Questions we'd ask too
Do I have to write Playwright selectors to get started?
No. Drive your app and it records the clicks, or describe the goal and the operator reads the screen. Either way you get real Playwright code, selectors included — you are never stuck depending on OCR at test time.
What is the PASS / FAIL / UNMEASURED verdict?
Anything a check could not actually observe — a page that never loaded, a flaky network hop — gets forced into FAIL by most runners. Runs here carry a third state instead. More in the blog post.
How does "turn any old website into an API" work?
Record the flow once and parameterise the inputs. It runs on request or on a schedule and returns structured output with the evidence. No API on the target site required — details on the website → API page.
Does QA Reef solve CAPTCHAs?
No, never. If a site challenges the agent, the run stops, records the screen as evidence, and hands it to a person. QA Reef 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.
Is QA Reef open source?
The core — flow runtime, engine adapter, OCR-driven operator, the cost-logging LLM adapter and evidence storage — is MIT-licensed under packages/core. The hosted app is not part of that core.
How is this different from other AI-driven QA platforms?
Closed "coverage as a service" platforms pair a managed runner with human QA engineers. This one looks similar at the surface, but the automation core is open source, the operator reads the screen instead of needing stable selectors, and every model decision is cost-logged. There is no managed human-verification service. Sourced comparison on the compare page.
Ready to let an agent QA your app?
Leave an email and we'll reach out when there's a seat — or skip the form and talk to the people building it.