SaarthurR/atticus-mac

What it solves

Atticus clears the friction of controlling a Mac by letting you speak what you need instead of clicking through menus and apps. It handles everyday tasks like opening apps, arranging windows, finding files by their contents, and answering questions about what's on your screen — even pointing to the exact spot to click. It also automates repetitive actions by learning your shortcuts, and can run longer jobs in the background so you can keep working.

How it works

You hold Ctrl+Option+Space (or a custom shortcut), speak your request, and release. Atticus uses Apple's on-device speech recognition to turn your voice into text, so your audio never leaves your Mac. For instant commands (like opening Spotify or moving a window) it acts directly without any AI. For more complex asks, it sends the request to your chosen AI client — Claude, Devin, Pi, or a local model like Ollama or LM Studio — which plans the steps. It can read your screen (with permission) to answer questions about what's shown, and can draw arrows or labels on the screen to show you where to click. It also learns from your actions: if you ask it to do something new, it remembers the steps and can do it instantly next time without AI. For safety, it runs AI commands in a sandbox by default and asks before anything harmful like deleting, sending, or buying.

Who it’s for

Atticus is for anyone who wants to control their Mac more efficiently using voice or quick text commands — power users, developers, people with accessibility needs, or anyone who wants to speed up routine tasks. It requires macOS 26 and an AI client: either a paid Claude plan with Claude Code, Devin, Pi, or a local model via a custom server. It's free and open source, and you can install it easily.

Highlights

• Learns your shortcuts: ask once, it works out steps; next time it runs instantly with no AI.
• Instant commands: opens apps, moves windows, plays music, searches the web without any AI.
• Sees your screen: ask about anything on screen, and it can point to the exact click spot.
• Draws explanations: it can draw labeled diagrams on your screen to explain anything.
• Dictate anywhere: hold the fn key to speak in any text field, with AI cleanup of stumbles.
• Background agents: run big tasks (like research) in the background while you keep working.
• Meeting Copilot: records and summarizes calls on your Mac.
• Privacy and safety: voice stays on-device, sandbox for AI commands, asks before dangerous actions.
• No API key required if using a local model.