Last week I was deep in a debugging session — juggling a terminal, browser docs, and VS Code — when I realized I'd spent 15 minutes just context-switching between windows. Every AI coding tool I've used lives inside an app. Cursor is in VS Code. Copilot is in the editor. Claude Code is in the terminal.
Then I found cue — an open-source macOS AI copilot that floats over your entire screen. And it changed how I work.
What Makes cue Different
Most AI coding assistants are trapped inside your editor. You switch to Chrome? They're gone. You're in Terminal? They can't see what you're doing.
cue takes a different approach:
| Feature | cue | Cursor | GitHub Copilot | Claude Code |
|---|---|---|---|---|
| Floats over screen | ✅ | ❌ | ❌ | ❌ |
| Sees your meetings | ✅ | ❌ | ❌ | ❌ |
| Hears audio input | ✅ | ❌ | ❌ | ❌ |
| Open source | ✅ | ❌ | ❌ | ❌ |
| Free | ✅ | $20/mo | $10/mo | $20/mo |
| Works across apps | ✅ | ❌ | ❌ | ❌ |
The "sees your meetings" part is what sold me. During a standup, cue can actually listen to the conversation and help you take notes, extract action items, or even suggest code changes based on what was discussed.
My Setup Experience
# Clone the repo
git clone https://github.com/Blueturboguy07/cue.git
cd cue
# Install dependencies
npm install
# Run in development mode
npm run dev
The whole setup took about 3 minutes. No API keys needed for basic features — it runs locally on your Mac.
How I Use It Daily
During code reviews: I open cue floating above my PR diff. It watches the code I'm reading and offers context without me having to copy-paste anything.
In meetings: It listens to standup calls and generates a summary with action items. This alone saves me 10 minutes per meeting.
While debugging: I can point at error messages on screen and cue understands the context — no need to explain what I'm looking at.
The Open-Source Advantage
This is the part that matters most to me. After spending months with MonkeyCode and other open-source tools, I've become a firm believer that AI coding tools should be transparent.
With cue, I can:
- See exactly what data is sent to the cloud (nothing, by default)
- Modify the prompts to match my coding style
- Add custom integrations with my workflow
- Contribute features back to the community
What's Still Rough
Let's be honest — it's not perfect:
- macOS only. Linux and Windows users are out of luck for now.
- Resource usage. It's a Tauri app, so it's lighter than Electron, but it still uses ~200MB RAM.
- Model quality. The default local model is decent but not Claude-level. You can plug in your own API key for better results.
The Bigger Picture
We're entering an era where AI assistants don't just live in your code editor — they live everywhere on your screen. cue is early, but it's pointing in the right direction.
I've been combining it with MonkeyCode for the actual coding work, and cue for the "ambient intelligence" layer. It's the closest I've felt to having an actual AI pair programmer.
What about you? Would you want an AI that sees and hears everything on your screen, or is that too much? And if you're on Mac, give cue a try — I'd love to hear your experience. 👇
P.S. If you know of similar tools for Linux/Windows, drop them in the comments. I've been looking.
Top comments (0)