macOS desktop assistant

Point and speak.
Watch it happen.

Genie is an AI assistant that lives on your desktop. Press Cmd+G, say what you need, and watch Genie work across your apps autonomously.

Download for macOS See how it works

$ Cmd+G

> "Summarize this PDF and post it to Sarah on Slack"

[genie] Reading the document...

[genie] Drafting a 3-bullet summary...

[genie] Posting to Sarah's DM in Slack...

[genie] Done — sent. Want me to follow up if she doesn't reply?

How it works

Three steps. Zero friction.

Genie captures everything it needs from your screen and voice to understand and execute your request.

Speak

Press Cmd+G and say what you need in plain English. Genie captures your voice, your screen, and your cursor position for full context.

Cmd + G

Understand

Genie sees what you see. It reads your screen, recognizes the app you're in, and understands exactly what you're pointing at. No copy-pasting or explaining.

Execute

Watch Genie work across your apps — browsing the web, editing documents, sending messages, running commands — all autonomously.

Features

Everything you'd wish for

Built with native macOS APIs for speed, privacy, and seamless integration with your workflow.

Voice-first interface

Local speech-to-text via Whisper. No internet required for transcription. Just speak naturally and Genie listens.

Sees your screen

Captures screenshots, reads text via OCR, and tracks your cursor — so it understands context just by looking at what you're looking at.

An agent that actually works

Genie's agent runtime drives your browser, edits documents, runs the terminal, and connects to 100+ services — web search, image generation, your Apple Notes, Google Drive, and more.

Notes, captured instantly

Say "note that..." and Genie saves it locally in milliseconds — no agent call, no API cost. Works offline once you've installed.

Native performance

Built on Tauri with Rust. ~32 MB binary, ~30 MB idle RAM. Direct macOS API access — no Electron, no bloat.

Use cases

Just say it. Genie does the rest.

Here are some things people say to Genie every day.

Research

"Read this article and summarize the key points"

Communication

"Draft a reply to this email and send it"

Documents

"Update the Q3 numbers in this Google Doc"

Development

"Run the test suite and fix any failures"

Data

"Pull the latest metrics from this dashboard"

Automation

"Every morning, check my calendar and brief me"

Capture

"Note that the meeting moved to 3pm tomorrow"

Quick answers

"What's the weather today?"

Follow-ups

"Actually, in New York?" — Genie remembers what you just asked.

Get started in minutes

Download Genie, sign in, and you're ready. Onboarding walks you through API key and permissions.

Download Genie for macOS

Apple Silicon (M1/M2/M3/M4). Requires macOS 12.3+. Intel build coming soon.

Open the .dmg and drag Genie to Applications

Launch Genie and follow the onboarding

Sign in with Google, paste your Anthropic API key, and grant Microphone, Accessibility, and Screen Recording permissions when prompted. Then press Cmd+G to start.

Download for macOS

Signed with Apple Developer ID · Notarized by Apple

Changelog

v0.1.5 Latest · June 1, 2026

What's new

More reliable screen capture during voice. Fixed a race where the background and voice-session capture loops shared the same temp file — during longer voice turns this caused dropped frames and the agent missing visual context. Each capture now gets its own path.
Cleaner task titles. No more one-word duds ("today"), dangling quotes, or meta-commentary ("here's a title for that..."). Short tasks skip the LLM entirely and just use the transcript.
Rapid retries grouped together. When you re-speak the same prompt 2-3 times in quick succession, the task panel collapses them into one expandable row instead of three identical entries.
Bubble color stays correlated with the speaking task. When multiple tasks run in parallel, the floating bubble's color now matches the response you're actually hearing — not whichever task ran last.
Smarter routing — even for ambiguous notes. Say "remember the api key for asana is in keychain" and Genie correctly saves it as a note — even when the phrasing doesn't match the strict "note that..." prefix. A fast Haiku classification call disambiguates yellow-flag cases.
Cmd+G is faster. Background work no longer freezes while Genie reads your clipboard. Slow clipboard? Doesn't matter — the rest of Genie stays responsive.
Earlier: smart routing for everything else. Say "note that..." and Genie saves it locally without calling the agent. Quick questions ("what's the weather") send a lighter payload.
Cleaner permission prompts. When the agent asks to edit a file or run a command, you see an icon, a clear action label, and a monospace file path — not a wall of raw text.
Yes / No chips. When Genie ends a response with a yes/no question, just click instead of starting another voice turn.
One-click Mute. Stop Genie mid-sentence without using a keyboard shortcut — new button in the task panel.
Distinct Cmd+G sounds. Tink to start listening, Pop to send. You can tell by ear which press just registered.
Stability. Stuck "Running tools..." after Mac sleep/wake fixed. Bubble no longer hangs on Listening after silence. Failed tasks are dismissible.
Signed + notarized. No more Gatekeeper warnings on first launch. macOS recognizes Genie as a legitimate Developer ID app.

Point and speak.Watch it happen.