pageai

System-wide AI overlay. Press the same arrow-key combos you'd use in a Chrome extension — but anywhere on your OS (browser, IDE, PDF viewer, terminal, games, anything). The overlay is hidden from screen sharing and screen recording by default, so Zoom / Meet / Teams / Loom / OBS / QuickTime can't see it.

Powered by Google Gemini. The package ships with a small bundled key pool, so on first install you don't even need to paste your own key — pick one with pageai rotate if "yours" gets rate-limited, or use pageai key <YOUR_KEY> to bring your own.

Install & run

npx pageai start

That's it. A tray icon appears, the system-wide hotkey listener attaches, and the overlay is ready.

To stop it:

npx pageai stop

Hotkeys (work everywhere on your OS)

| Combo | Action | |---|---| | Left + Right | Capture full-screen screenshot | | Up + Right | Ask Gemini about the current screen | | Up + Down | Toggle the overlay | | Up + Left | Toggle follow-cursor mode | | Alt + A | Full scan of foreground app + ask Gemini | | Alt (alone) | Hide the overlay |

These are the same combos as the legacy Chrome extension, just lifted to OS level via a low-level keyboard hook (uiohook-napi).

Hidden from screen sharing

By default the overlay window has OS-level "content protection" turned on:

• macOS: NSWindowSharingNone — invisible to ScreenCaptureKit / CGWindowListCreateImage / Zoom / Meet / Teams / Loom / OBS / QuickTime.
• Windows: WDA_EXCLUDEFROMCAPTURE (Win10 2004+) — hidden from Windows Graphics Capture and Desktop Duplication APIs.
• Linux/X11: not supported by the OS — the overlay will appear in screen share. (Wayland support is on the roadmap.)

To toggle:

pageai stealth on    # default
pageai stealth off

Or click the tray icon → Stealth (hide from screen share) checkbox.

A physical phone camera pointed at your screen will obviously still see it. No software can defend against that.

Commands

pageai start            # launch the tray daemon (default command)
pageai stop             # kill the daemon
pageai status           # show daemon, stealth, and key state
pageai stealth on|off   # toggle hide-from-screen-share
pageai rotate           # pick the next bundled Gemini key
pageai key <YOUR_KEY>   # use your own Gemini key (overrides the pool)
pageai clear-key        # forget your override and fall back to the pool

Permissions

• macOS: First run pops the OS Accessibility prompt (System Settings → Privacy & Security → Accessibility). Enable PageAI, then pageai stop and pageai start again. The tray menu has a quick-link.
• Windows: No prompt; works out of the box.
• Linux: Requires X11 (Wayland is unsupported by the underlying keyboard hook).

How the bundled key pool works

shared/keys.json ships inside the npm tarball with N pre-provisioned Gemini API keys. On first run, your install picks a deterministic slot via hash(machineId) % keys.length — same machine always gets the same slot unless you run pageai rotate.

Same Google account / same project: all keys created under one account usually share the same Gemini free-tier quota. Rotating keys does not give you more requests. To avoid 429 / quota errors, enable billing in Google AI Studio, use keys from different Cloud projects (each with its own billing), or run pageai key <YOUR_KEY> on a paid-capable key.

Warning for maintainers: keys in shared/keys.json are publicly downloadable as part of the tarball. Use throwaway / budget-capped keys only. For production users, pageai key <YOUR_KEY> is always the safer path.

After npx pageai start, you can close the terminal — the app stays in the tray until pageai stop or Quit from the menu.

Project layout

pageai/
 ├── cli/           # `pageai` CLI entry (start/stop/status/rotate/stealth/key)
 ├── desktop/src/   # Electron daemon, hotkeys, capture, overlay window
 ├── shared/        # types, storage (Conf), gemini client, keyPool
 ├── extension/     # legacy Chrome extension (no longer the primary path)
 ├── docs/          # release notes
 ├── package.json
 └── README.md

Development

npm install
npm run build
node dist/cli/index.js start   # foreground-friendly: append --foreground

Build & publish

# 1. Populate shared/keys.json with real Gemini keys.
# 2. Build:
npm run build
# 3. Sanity check:
npm pack --dry-run
# 4. Publish:
npm publish

Caveats

• The bundled key pool is a convenience, not a security boundary. Anyone can download the tarball and read the keys. Google may auto-revoke leaked keys.
• We do not run OCR; Gemini reads the screenshot directly. If the model can't see fine-grained text, drop in your own OCR (Tesseract) before posting.
• The overlay is click-through (setIgnoreMouseEvents); selection works via copying the cached answer (pageai status will show what's stored).