SpeakAI — Local English Practice

AI-powered English speaking practice that runs 100% on your device.
Powered by RunAnywhere Web SDK (llama.cpp WASM) — no API keys, no server.

---

Requirements

• Node.js 18+ — download from https://nodejs.org
• npm (comes with Node.js)
• Chrome or Edge browser (Web Speech API for microphone)

Check your version:

node --version   # should be v18 or higher
npm --version

---

Setup & Run

# 1. Go into the project folder
cd speakai-local

# 2. Install dependencies (only needed once — takes 1-2 minutes)
npm install

# 3. Start the dev server
npm run dev

Then open http://localhost:5001 in Chrome or Edge.

---

How it works

1. Open the app → choose a personality and practice mode
2. On the practice page, you can:
   • Skip → use browser speech only (instant, no download)
   • Download a model → downloads ~300MB AI model to your browser once, then works offline
3. Click the mic button or press Space to speak
4. The AI responds in text + voice

---

Deployment to Render (Static Site)

Since the AI runs entirely in the browser using WebAssembly, this app can be deployed as a Static Site (often free on Render).

1. Dashboard: Create a new Static Site on Render.
2. Repo: Connect this GitHub repository.
3. Settings:
   • Build Command: npm install && npm run build
   • Publish Directory: dist/public
4. Rewrite Rules:
   • Go to the Redirects/Rewrites tab for your service.
   • Add a Rewrite rule:
     • Source: /*
     • Destination: /index.html
     • Action: Rewrite
   • This ensures that refreshing the page on sub-paths (like /practice/...) works correctly.

The app serves static files from dist/public.

Folder structure

speakai-local/
├── client/
│   ├── public/wasm/        ← WASM binaries (RunAnywhere SDK)
│   └── src/
│       ├── pages/          ← HomePage, PracticePage
│       ├── lib/runanywhere.ts  ← SDK integration
│       └── index.css       ← RunAnywhere orange theme
├── server/                 ← Express dev server
├── package.json
└── vite.config.ts

---

Notes

• Mic permission: Chrome will ask for microphone access when you first click the mic button — click Allow
• Model download: the AI model (~300MB) downloads once and is cached in your browser. You don't need to re-download it each time.
• No internet needed for AI inference — only for the initial model download from HuggingFace