Radixlink

AI Accelerator for QA · Free for testers

Build Playwright tests with AI.

Connect your GitHub repo. The Radixlink AI Accelerator scans your code, finds the user flows and drafts Playwright tests. You approve each one, and it goes to GitHub with a CI/CD pipeline.

Sign up See how it works
checkout/apply-discount.spec.tsWaiting for your approval
// Drafted by the Accelerator
test('discount code reduces the order total', async ({ page }) => {
  await page.goto('/cart');
  await page.getByLabel('Discount code').fill(code);
  await page.getByRole('button', { name: 'Apply' }).click();
  await expect(page.getByTestId('total'))
    .toHaveText('$90.00');
});
Confidence92%
Tied to a known app rule: the 10% discount on orders over $100. Checks the total, not just that the page loaded.

Your callDoes this check prove the rule it claims to?

RegenerateRejectApprove

A green test is not proof.

AI can write a hundred tests before lunch. Some of them check nothing at all.

We show which tests really check something, so you know which ones to trust. You make the final call.

  • ✓ PASSEDCheckout page loaded18%
  • ✓ PASSEDApply button is visible35%
  • ✓ PASSEDDiscount takes the total to $90.0092%

How it works

From code to approved tests.

  1. Connect GitHubrepo + app URL

    Point it at your repo, or a local folder, and the app URL.

  2. Scan the codemodules & flows

    It reads the code and finds every module and user flow.

  3. App profileyou approve it

    The AI writes what your app does, its records and risks.

  4. Pick flowsonly what you need

    Choose which flows to cover. Nothing is written for the rest.

  5. Generate specsPlaywright

    Playwright specs, each listing what it proves.

  6. Approval gateyou decide

    Read the code. Approve, regenerate or reject.

  7. Commitversioned in Git

    Approved specs go to GitHub with the CI/CD pipeline.

  8. Run & reportwith proof

    Run locally or in CI/CD. HTML and PDF reports.

Inside the Accelerator

See what it does at each step.

It finds the flows for you.

The scan reads your code and groups it into modules. Each module gets a list of real user flows. Tick the ones you want tested.

  • Every module and flow found from the code
  • Find new flows any time with one click
  • Tests are written only for what you select
Detected modules and flows IT asset app
  1. DASHBOARD10 flows
  2. ASSETS20 flows
  3. LICENSES16 flows
  4. ACCESSORIES12 flows
  5. CONSUMABLES16 flows
  6. MAINTENANCE10 flows
View the dashboard summary counts
Adjust a low-inventory item's quantity
Follow a "Needs attention" item to its list
Open the full activity report
Open the calendar from "Today & Up Next"
Generate test specs →

It learns your app before it writes a test.

The AI writes an app profile from the menu, the pages, the code and your documents. You approve it. Then every test is planned from it.

  • Main records, user roles and status flows
  • Risky areas, and the risks each test must cover
  • Upload business rules as documents
  • App memory: tips the AI uses once you approve them
App profile Approved
Kind of app
IT asset management
What it is for
Track IT assets, who has them, software licenses and stock.
User roles
Superadmin, Admin
Status flow
Pending → Unaccepted → Checked out
Risky areas
Asset assignmentMEDIUM License complianceMEDIUM
Its words
Asset = IT item assigned to a personChecked out = issued
Must cover
A quantity of 0, a negative one, or more than is available is refused.

You approve every spec.

Each spec shows what it does in plain English and what it proves. Read the code, check its history, then approve, regenerate or reject.

  • "What these tests prove" on every spec
  • Lint warnings and fragile selectors flagged
  • Run all approved specs, or just the ones you pick
Review & approval gate 2 specs
Create a new assetcreate-asset.spec.ts
View codeHistoryApproved

Fill a unique asset tag, pick a model and status, save, then reload to confirm the record really persists.

▸ What these tests prove (5)Confidence 91%
View the dashboard summary countsdashboard-overview.spec.ts
RegenerateApproveReject
▸ What these tests prove (4)Fragile selectorsNot run yet

Reports that show real checks.

Every run tells you how many tests passed with real checks, not just how many went green. Each test case has its screenshot, video and trace.

  • Positive, negative, validation, boundary and security cases
  • Visual check warns when a page looks broken
  • HTML and PDF reports, plus full execution history
Test execution & reports signin.spec.ts
Run status✓ Passed
Test cases5 / 54 with real checks
Failed0
Valid sign-in opens the signed-in areaPOSITIVEPassed
Wrong password is rejectedNEGATIVEPassed
Each required field shows its own messageVALIDATIONPassed
Email format, at the rule and one step pastBOUNDARYPassed
Signed-out visitor cannot open a protected pageSECURITYPassed
Visual check: date columns are too narrow and wrap across lines.

Integrations

Your GitHub. Your AI.

GitHub

Approved tests and the pipeline that runs them, in your own repository.

Approve→Commit→CI/CD run
  • Connect with GitHub. Sign in once, or pick an existing connection.
  • Version-controlled tests. Approved specs and the CI/CD workflow are committed to your repository.
  • Test every pull request. When a PR opens, the AI updates the tests it affects and writes tests for new features. You review them. After the merge, they run and the results go on the PR.

Bring your own AI

Model-agnostic by design. Today it runs on Claude.

Claude CLIorClaude API
  • Claude CLI. Sign in with your own Claude account and give the Accelerator access to the claude CLI. This is how the free plan works.
  • Claude API. Prefer an API key? Add your own Claude API key instead.
  • Not tied to one model. The Accelerator is built so more models can be added later.
  • See what it costs. Token use and cost are shown for every app, module and test.

What you get

Speed from AI. Control for you.

Approval gateNothing counts until you say so

You sign off on every spec, and on any repair that changes what a test expects.

Real checksSee what each test proves

Each spec lists what it proves. Each run counts the tests that passed with real checks.

Plain PlaywrightYour code, no lock-in

Standard TypeScript specs in your repo. Read, edit and run them yourself.

App memoryIt learns from your fixes

Tips the AI reads before it writes or fixes a test. You approve each one.

Visual checkFlags broken layouts

After each run the AI looks at the screenshot and warns about broken layouts.

Safe loginsEncrypted, never in logs

Stored encrypted and never written into plans, specs or logs.

Who it is for

Built for the people who test.

Testers moving to automation

Start with working Playwright tests, then learn from the code it drafts.

QA leads using AI already

Add the review step that shows which AI-written tests prove something.

Small teams and solo testers

Cover more of your application without writing every test by hand.

From the blog

Building it in the open.

All posts
Architecture

Our AI wrote the same bug twice

Why we are building an Application Profile: a place for what the system learns to live.

Read →
Lessons from the build

One wrong string broke five tests

"signin" versus "sign-in", and what it taught us about starting every generation from zero.

Read →
Models

Same app, three models, three very different answers

The model that passed every test wrote the least worth passing.

Read →

Get free access.

Send us your name, company, role and the application you test. We will email you access.

Sign up by email

sales@radixlink.ai

Already have access? Log in to the Accelerator

Need a team behind it?

Radixlink also provides QA services, delivered with the Accelerator. Our engineers work onsite or offshore. Ask about a two-week, fixed-scope pilot on one module.

Talk to us

sales@radixlink.ai

FAQ

Frequently asked questions

Is it free?

Yes. The Accelerator is free for testers. It runs on your own Claude account through the Claude CLI, so you control what the AI uses.

How do I get access?

Sign up with your details. We email you access.

Do I need my own AI account?

Yes. To use the Accelerator free, you need your own Claude account and give it access to the Claude CLI. You can use a Claude API key instead. The Accelerator is model-agnostic, so more models can be added later.

Where are my tests stored?

In your GitHub repository. Approved specs are committed with a CI/CD workflow, version-controlled, and they stay yours.

How do I know a test is accurate?

Every spec lists what it proves, and each run reports how many tests passed with real checks. A test that only checks the page loaded does not count. You approve each spec before it runs.

Can the AI fix its own broken tests?

It can propose a repair and show you what changed. A repair that changes what a test expects needs your approval, so a real bug cannot be quietly tested away.

Are my app logins safe?

Logins are stored encrypted. They are used only to sign in during scanning and test runs, and are never written into plans, specs or logs.

What do I need to get started?

A GitHub repository (or a local folder) with your app's code, the URL of the app to test, and a login for it if it needs one. For the AI, your own Claude account through the Claude CLI, or a Claude API key.

Can I edit the generated tests?

Yes. They are plain Playwright TypeScript. Read them, change them and run them yourself, with or without the Accelerator.

Does it test pull requests?

Yes, if you turn it on. When a PR opens, the AI updates the tests it affects and writes tests for new features. You review them, and after the merge the results go on the PR.

Does it replace testers?

No. It drafts tests, and you decide which ones count.