Skip to content
Clacky raynanwuyep.com ↗

Clicky is Mac-only. So I built it for Windows. Open source.

Meet Clacky

An AI buddy that lives on your Windows PC. She follows your cursor, sees what you see, and actually does things.

Demo video dropping soon

On a Mac? Get the original at heyclicky.com ↗

Built on Claude (Computer Use) · Deepgram streaming voice · Python + PyQt6

What she does

She teaches

Say "Walk me through Premiere Pro" and she tours the app, pointing at each panel as she explains it.

She learns your routines

Say "Save this as my morning routine" once, and run it forever. (She's also learnt I suck at Elden Ring.)

She acts

Opens apps, clicks, types, searches. A real computer-use agent, on your machine.

She's reversible

"Clean up my desktop" sweeps everything into folders. Don't like it? "Undo." Every file, back where it was.

Try saying

  • hey Clacky, what am I looking at?
  • walk me through how I'd edit my first video
  • Clacky, save this as game time: open Steam and Elden Ring tutorials
  • Clacky, go: open my calendar
  • my desktop's a mess, clean it up
  • actually, undo.

No coding required

Just want the app?

Right now, getting Clacky running takes a few technical steps. I'm building a version that just works: download it, open it, and start talking. No code, no setup. Leave your email and I'll send it to you the moment it's ready, and you'll be first in line.

No spam. Just one email when it's ready.

How it works

Under the hood: a Qt overlay renders the buddy; Deepgram streams speech while you talk; a two-tier router (local fast-paths plus a Haiku classifier) picks the lane; Claude's Computer Use API sees the screen and drives the mouse; every file operation is journaled for one-word undo.

Voice-to-action in under ~1s perceived

Streaming STT, streamed model responses, sentence-pipelined TTS, warm connections, and instant acks.

read routing.py

Pointer and voice, always in sync

The tour returns inline [POINT:x,y] tags inside the spoken text, so cursor timing derives from the words themselves. One channel, so they can't drift.

read tour.py

Trust by construction

Move-only file ops, journaled undo, protected system files, and your keys stay on your machine.

read actions.py

Get it

  1. 1Download the .exe (or pip install from source).
  2. 2Paste your Anthropic and Deepgram keys into the setup wizard.
  3. 3Hold Ctrl+Alt+M and say "what's on my screen?"

Credits

Clacky stands on great shoulders: Clicky by farza (the original), clicky-windows by Bitshank, OpenClicky by Jason Kneen, and Cua by trycua. Built on Anthropic's Claude. MIT licensed, like everything it learned from.

I build things like this.

If you are working on agents, voice, or things that need hands, I would love to talk.