Notes · AI

Why I built my own iOS app to draft WhatsApp replies

· Sebastian Schuon · Originally on LinkedIn

A man lies on a child's bed with star wallpaper and stuffed animals, typing on his phone

Gmail has drafts. WhatsApp doesn’t. That’s why I now have my own iOS app.

Until today my interface to automation was Claude Code (I prefer the UI version). For mail that was fine: Claude proposes a reply, puts it in Gmail as a draft, I edit and send it (and Claude then does learn on the differences from what I send to what it proposed.

I wanted the same for WhatsApp. But WhatsApp has no cross-device drafts. There is no way to put a prepared reply into the app and finish it later on the phone.

So I needed a place to see the proposed reply, edit it and send it. One evening on holiday, at dinner, I asked Claude to build that place as an iOS app. Crazily it worked one-shot (just the Mac developer configuration with Testflight took a few more prompts)

This is the setup:

  • The app is only the screen. The CLIs and skills keep running on the Mac at home.
  • Phone and Mac talk over Tailscale, a private VPN. Nothing is exposed to the internet.
  • It build a node server that runs on the Mac and uses the CLI for deterministic actions (like sending) and to interface Claude (draft messages via claude /p)
  • The CLI is wacli. As read status not really synced well, I decided to use “archive” for state management. And that works quite well (there had been a few issues, Claude wrote its own patch for the CLI 😜)

The AI never sends. It proposes, I send. And recently I implemented a neat functionality: “snooze”, just was we have it in Superhuman. Really handy.

Has anyone found a way to pre-draft WhatsApp replies without building their own app?

More from AI