sam-siavoshian

sam-siavoshian / agent-notch

Public

Agent Notch — macOS computer-use agent that lives in the notch. Voice in, screen context in, Claude Sonnet drives the mouse.

agent-notch.vercel.app ai-agent anthropic apple-silicon automation claude
10
0
85% credibility
Found May 17, 2026 at 10 stars -- GitGems finds repos before they trend. Get early access to the next one.
Sign Up Free
AI Analysis
Swift
AI Summary

Agent Notch is a macOS app that lives inside the notch of MacBook Pro laptops. It lets you control your computer by holding down your cursor, speaking a command, and watching an AI assistant carry it out. The notch displays what the AI is thinking and doing in real-time. It learns your apps over time, so repeated tasks get faster. It also shows your Spotify music and lyrics right in the notch.

How It Works

1
💻 You notice a tiny window in your Mac's notch

After installing the app, a small floating companion appears near your cursor. It gently bobs and pulses, waiting for your attention.

2
🔒 You grant a few permissions once

On first launch, the app asks for three permissions: letting it see your screen, control your mouse, and listen through your microphone. You approve them once and never think about it again.

3
👆 You hold down the click on the cursor companion

A long press on the floating sprite activates listening mode. The companion glows and pulses, and a waveform animation appears in the notch above.

4
🎤 You speak a simple command

You say something like "open my calendar" or "search for that file I downloaded yesterday." The app transcribes your words and sends them to the AI.

5
The AI springs into action
⚡
Fast path: it already knows how

If you've asked something similar before, the AI recalls the steps from memory and acts immediately—no screenshots needed.

👁️
Learning path: it studies the screen

For new tasks, the AI takes a quick look at what's on screen, understands the layout, then clicks and types its way through.

6
✨ Your task completes

The AI confirms its work aloud and the notch returns to its resting state. Your calendar opened, your file was found, or whatever you asked for is done.

🎵 Bonus: your music follows you

When Spotify is playing, the notch shows the song, artist, and even synced lyrics. You can pause, skip, or search without switching apps.

Sign up to see the full architecture

5 more

Sign Up Free

Star Growth

See how this repo grew from 10 to 10 stars Sign Up Free
Repurpose This Repo

Repurpose is a Pro feature

Generate ready-to-use prompts for X threads, LinkedIn posts, blog posts, YouTube scripts, and more -- with full repo context baked in.

Unlock Repurpose
AI-Generated Review

What is agent-notch?

Agent Notch is a macOS computer-use agent that lives inside your MacBook's notch. You long-press the cursor, speak a command in natural language, and Claude Sonnet takes over the mouse to execute it. The notch itself becomes a live dashboard showing what the agent is thinking and doing, with screen context fed in via OCR and Gemini so it knows exactly where it is on screen.

Why is it gaining traction?

The interface is genuinely novel -- instead of a floating window or menu bar app, the notch becomes the control center. Voice in, screen context in, mouse driven out. The architecture layers fast paths (URL opening, AppleScript, accessibility queries) before falling back to vision-based clicking, which keeps most tasks fast. The rolling prompt cache means repeated actions get cheaper over time.

Who should use this?

Power users on M-series MacBook Pros who want hands-free automation for repetitive workflows. Developers building computer-use prototypes will find the modular tool dispatcher and intent router useful as a reference implementation. It is not ready for production deployment -- the 10 stars and hackathon origins (TritonHacks 2026) reflect a proof of concept, not a polished product.

Verdict

This is a creative proof of concept with a compelling UX idea, but treat it as a starting point rather than a daily driver. The credibility score of 0.85% reflects the early stage -- thin documentation, no test suite visible, and the signing/TCC permission dance makes local iteration painful. Worth watching if the team ships a proper installer and expands beyond M-series hardware.

Sign up to read the full AI review Sign Up Free

Similar repos coming soon.