AI-powered web accessibility scanner that finds what automated tools miss.
Over 96% of the top 1 million websites have detectable WCAG failures. Automated tools like Lighthouse and axe-core catch only ~30–40% of real accessibility issues — the rest require human eyes or, now, AI vision.
Aria Access closes that gap.
| Feature | Traditional Tools | Aria Access |
|---|---|---|
| Automated rule-based scan | ✅ | ✅ (axe-core) |
| AI Vision analysis | ❌ | ✅ (GPT-4o) |
| Contextual fix suggestions | ❌ | ✅ |
| WCAG 2.2 criteria mapping | Partial | ✅ Full |
| Severity triage | Basic | ✅ Critical → Minor |
| Scan-to-report in seconds | Varies | ✅ ~10s |
| No install required | ❌ | ✅ Pure SaaS |
- Dual-engine scanning — axe-core catches rule violations; GPT-4o Vision spots visual/contextual issues a rule engine never could (low contrast text in images, misleading link text, decorative elements mislabelled)
- Accessibility score 0–100 with a breakdown by severity: Critical, Serious, Moderate, Minor
- WCAG 2.2 criteria tags on every issue so developers know exactly which success criterion is violated
- Actionable fix suggestions — not just "this is broken", but here's the code to fix it
- Instant, no-install — paste a URL and get results in ~10 seconds
- Auth-gated deep scans via Clerk — free demo on curated sites, full scans for signed-in users
- Persistent scan history with Neon Serverless Postgres + Drizzle ORM
- Fully accessible UI — Aria Access itself passes its own scanner
┌─────────────────────────────────────────────────────────────┐
│ Next.js 16 App Router │
│ │
│ Landing Page ──► /api/scan ──► Puppeteer (Chromium) │
│ │ │
│ ┌─────┴──────┐ │
│ │ │ │
│ axe-core Screenshot │
│ (DOM audit) (via page.screenshot) │
│ │ │ │
│ └──────┬─────┘ │
│ │ │
│ GPT-4o Vision │
│ (visual analysis) │
│ │ │
│ Merged ScanResult │
│ │ │
│ Neon Postgres (scan history) │
└─────────────────────────────────────────────────────────────┘
Stack:
- Framework: Next.js 16 (App Router, Server Actions, Edge-ready)
- Language: TypeScript 5 (strict)
- Styling: Tailwind CSS 4 + shadcn/ui
- Auth: Clerk (JWT, social login, magic link)
- Database: Neon Serverless Postgres + Drizzle ORM
- AI: OpenAI GPT-4o Vision API
- Scanning: Puppeteer Core + @sparticuz/chromium (serverless-compatible)
- Accessibility engine: axe-core 4.12
- Animation: Motion (Framer Motion v12)
- Deployment: Vercel
- Node.js 20+
- A Clerk account (free tier works)
- A Neon database (free tier works)
- An OpenAI API key with GPT-4o access
git clone https://github.com/supratimsarkar/aria-access.git
cd aria-access
npm installcp .env.example .env.localOpen .env.local and fill in:
# Clerk — https://dashboard.clerk.com
NEXT_PUBLIC_CLERK_PUBLISHABLE_KEY=pk_test_...
CLERK_SECRET_KEY=sk_test_...
# Neon — https://console.neon.tech
DATABASE_URL=postgresql://...
# OpenAI — https://platform.openai.com/api-keys
OPENAI_API_KEY=sk-...npx drizzle-kit pushnpm run devOpen http://localhost:3000 — paste any URL and hit Scan.
Scan any of the curated demo sites (github.com, stripe.com, amazon.com) without signing in to see Aria in action.
Scan any URL on the internet. Your scan history is saved to your account.
| Severity | Meaning |
|---|---|
| 🔴 Critical | Blocks access entirely for some users |
| 🟠 Serious | Severely degrades experience |
| 🟡 Moderate | Causes friction, workarounds exist |
| ⚪ Minor | Best-practice violation, low impact |
Each issue card shows:
- The failing element (DOM selector or screenshot region)
- The WCAG 2.2 criterion violated
- A plain-English description
- A concrete code fix
aria-access/
├── app/
│ ├── api/
│ │ └── scan/ # POST /api/scan — the core scanning endpoint
│ ├── layout.tsx
│ └── page.tsx # Landing page
├── components/
│ ├── landing/ # Hero, HowItWorks, ScanDemo, etc.
│ ├── scan-form.tsx # URL input + scan trigger
│ ├── scan-results.tsx # Score ring + issues list
│ ├── issue-card.tsx # Individual accessibility issue
│ └── score-ring.tsx # SVG score visualisation
├── db/
│ ├── index.ts # Drizzle client (Neon serverless)
│ └── schema.ts # Database schema
├── lib/
│ ├── types.ts # ScanResult, Issue, Severity types
│ └── utils.ts # cn() and helpers
└── public/
└── demo.svg # Animated product demo
- PDF / HTML report export — downloadable audit reports for clients
- CI/CD integration — GitHub Action that fails the build on critical regressions
- Scheduled monitoring — weekly scans with email diff reports
- Multi-page crawl — scan entire site, not just one URL
- Team workspaces — share reports, assign issues, track remediation
- Browser extension — scan the page you're on with one click
- WCAG 2.2 AA/AAA toggle — choose your compliance target
- Remediation tracking — re-scan and see your score improve over time
Contributions are very welcome! See CONTRIBUTING.md for guidelines.
Found a bug? Open an issue. Have an idea? Start a discussion.
If you discover a security vulnerability, please see SECURITY.md for responsible disclosure instructions. Do not open a public issue.
MIT — free to use, modify, and distribute. Attribution appreciated.
"The power of the Web is in its universality. Access by everyone regardless of disability is an essential aspect." — Tim Berners-Lee, W3C Director
An estimated 1.3 billion people live with some form of disability. Web accessibility is not a niche compliance checkbox — it is the difference between a product that works for everyone and one that silently excludes hundreds of millions of people.
Aria Access makes it effortless to know where you stand and exactly what to fix.
Built with care by Supratim Sarkar