DEV Community

Cover image for I Built Smart Reply AI: Quick Replies That Work Even Offline
Mahmud Rahman
Mahmud Rahman

Posted on

I Built Smart Reply AI: Quick Replies That Work Even Offline

We all do it: someone sends a message, and we stare at the screen thinking, "What do I even reply?"

So I built Smart Reply AI, an open-source assistant that helps you reply, write better, translate, and summarize, right where you already chat.

What can it do?

It has four modes:

  • Smart Reply: gives you up to 4 reply ideas for any message
  • Smart Enhance: fixes grammar and makes your text clearer
  • Smart Translate: translates into 11 languages
  • Smart Summarize: turns long emails and notes into short takeaways

You can also pick a tone: professional, friendly, casual, concise, formal, or flirty.

Works where you work

You don't need to copy-paste all the time:

  • Windows: press Ctrl + Shift + R and a small pill pops up next to your cursor. Click it and the reply is pasted for you. The whole app is under 2 MB.
  • Android and smartwatches: replies show up right under your notifications and on your Wear OS watch.
  • Chrome/Edge extension: suggestion chips appear inside WhatsApp Web, Gmail, and more.
  • Also available: Flutter app (Android, iOS, macOS, Linux), a React web app, and a native HarmonyOS app.

Fast, and private if you want

The best part is the hybrid engine. It runs a local rules engine in under 1 ms, and at the same time asks a cloud LLM (Groq, OpenRouter, Ollama, or any OpenAI-compatible API). If the cloud answers in time, you get better suggestions. If not, you still get instant ones.

Want full privacy? Use offline-only mode. No network calls at all.

Bring your own API key, and it stays on your device.

A bit of backend love

The Node.js backend uses a SHA-256 cache and singleflight deduplication. If 500 people ask for the same thing at once, only 1 request goes to the LLM, and the other 499 share the result. Nice for saving rate limits.

Try it in 30 seconds

cd backend
npm install
npm run dev
Enter fullscreen mode Exit fullscreen mode

It runs on localhost:5006, and it works even without an API key thanks to the built-in offline engine.

Or skip the setup and grab a ready-made build from the releases page.

Links

It's still growing, so stars, feedback, and PRs are very welcome. Thanks for reading! 🙌

Top comments (0)