Privacy Policy

Voice commands

When you hold ⌥ in the box or click the microphone, Gazey listens until you let go. Your speech is turned into text on your Mac by Apple's on-device speech recognition; the audio is never recorded, stored or sent anywhere. The text is then handled like something you typed: simple Mac actions (opening or quitting apps, opening websites, web searches, installing VS Code extensions) run on your Mac; if you turned on smart routing with a TypeSafe key, a command Gazey can't place by itself is sent to TypeSafe (the words of that command plus the names of your installed apps) to understand it; anything else goes to your AI agent as described below. Installing a VS Code extension looks up its name on Microsoft's Visual Studio Marketplace.

Operating your Mac

To carry out a task, your AI agent can use Gazey's Mac tools: run AppleScript in apps, choose menu items, press keys, take a fresh screenshot, and read and fill the fields and buttons of an app's window (through macOS Accessibility). What it reads this way, such as the labels and contents of a window's fields, goes to the agent you chose, like the screenshot does. The tools work only during a request you made, anything that changes something follows your approval setting, and macOS asks you once before Gazey controls each app (System Settings → Privacy & Security → Automation). If you set up a TypeSafe key, the labels of a window's controls are sent to TypeSafe to decide which one you meant ("the name field").

What Gazey sees when you summon it

• A screenshot of the display your pointer is on. It is taken only when you press the Gazey shortcut, never in the background.
• The app you're in and its window title, and any text you've highlighted (with your Accessibility permission).
• What you type in the box.

Where your requests go

The AI agent you choose. Your request, the screenshot, the app name, window title and highlighted text are sent to Claude Code (by Anthropic) or Codex (by OpenAI), using your own account with them. They handle it under their privacy policies, exactly as if you used them in a terminal:

• Anthropic: https://www.anthropic.com/legal/privacy
• OpenAI: https://openai.com/policies/privacy-policy

The local model (optional). If you set up the local model, simple actions and the quick reply run on a model on your Mac (through Ollama). What it handles never leaves your Mac. When a request is too much for it, it's passed to the agent you chose, as described above. If you click Install during setup, Gazey downloads the official Ollama app from ollama.com (served from GitHub), checks it is signed and notarized by Ollama's developer, and puts it in your Applications folder; the model itself is downloaded by Ollama from its library. Neither download sends anything about you beyond the request itself.

If you click Set up for Claude Code or Codex, Gazey runs that tool's official installer (from claude.ai or releases.openai.com), then the tool's own sign-in, which opens your browser. You sign in with Anthropic or OpenAI directly; Gazey never sees your password.

The quick reply. To acknowledge actions instantly, Gazey sends your typed request, the app name and the window title, but no screenshot and no highlighted text, to Claude Haiku through your Claude Code account.

Smart routing (off by default). If you turn on routing and add a TypeSafe key, your typed request, the app name and the window title are sent to TypeSafe to pick an agent. Screenshots are never sent to TypeSafe.

Your connected apps. When the agent uses a connection you set up (for example Google Calendar or Notion), it reads or changes data there on your behalf. Anything that changes something asks for your approval first, unless you choose otherwise.

What stays on your Mac

• Screenshots are kept in a private temporary folder only until your request is sent, then deleted. Unused ones are deleted when you close the box, and any leftovers when Gazey quits or restarts.
• Your task list lives in memory and is cleared when Gazey quits.
• Settings and sign-in tokens for connections Gazey set up are stored in your user account, readable only by you.
• A diagnostic log at ~/Library/Logs/Summon.log records events such as "summoned over Safari" or "screen captured". It never contains your requests, the agent's replies or window titles.
• Your activity history at ~/Library/Application Support/Summon/activity.json: your last 300 requests (their text), which agent or model handled each, the actions it took, the tokens and estimated cost it reported, and how it ended, so you can see them in Control Center → Activity. It stays on your Mac and is never sent anywhere; Clear History there deletes it.

Updates

Gazey checks the Gazey website for new versions (you can turn this off in About). That request includes nothing about you beyond what any web request reveals, such as your IP address.

Feedback

"Report a Bug" and "Suggest a Feature" open a form in Gazey. When you press Send, your message, the reply email you choose to add (if any) and the diagnostics the form shows you (your Gazey and macOS versions, which agents are signed in, and, unless you untick it, the last lines of the diagnostic log) are sent to the Gazey website, which forwards them to the developer by email. Nothing is sent until you press Send, and nothing else is stored.

Children

Gazey is not directed at children under 13.

Changes

If this policy changes, the new version will ship with Gazey and be posted on the Gazey website, with a new effective date.

Contact

Questions about privacy: use "Report a Bug" in Gazey → Control Center → About.