Never freeze on a call again.
Live answers, screen solving and call notes, in a floating panel that stays out of screen sharing.
Download for MacYou see the panel. They see your screen without it.
Everything in one quiet panel.
Global shortcuts drive it, so you never click away from the call.
Out of the screen share
The panel is excluded from screen capture by macOS itself. Tested with Google Meet, with a built-in test you can run before any call.
Hears both sides
On-device transcription of the call audio (them) and your microphone (you). Nothing is recorded unless you save notes.
Solves your screen
⌃⌥S sends a screenshot (a coding problem, an error, a slide) and returns the approach with working code.
Copy in one keystroke
⌃⌥C copies just the solution code, ready to paste. Press again for the next code block.
Notes when you hang up
⌃⌥E writes a summary, action items and the full transcript to Markdown, then continues in Claude Code or Codex in one click.
Your model, your keys
Use your ChatGPT login (Codex), your Claude login (Claude Code), the Anthropic API, or any OpenAI-compatible endpoint, Ollama included.
Six shortcuts to remember.
Press ⌃⌥/ for the full list, right inside the panel.
Ready in a minute.
Install
Open the downloaded disk image and drag Invisible into Applications. It's signed and notarized by Apple, so it just opens, and it keeps itself up to date from there.
Allow access
Open it, then allow Screen & System Audio Recording and Microphone when macOS asks. It runs without a Dock icon.
Pick a model
Click the gear in the panel to choose your model, or paste an API key. Start listening with ⌃⌥L.
Questions
Can the other side really not see it?
Invisible asks macOS to leave its window out of every screen capture, so screen sharing and recordings that use the system's capture show what's behind it. It was tested with Google Meet sharing the entire screen. Apps that capture differently can behave differently (some recent Zoom versions are reported to capture protected windows), so run Test invisibility in settings, or a quick test call, before you rely on it. Your mouse pointer and any menus you open stay visible to others.
Which models can it use?
Codex with your ChatGPT login, Claude Code with your Claude login, the Anthropic API with your key, or any OpenAI-compatible endpoint: OpenAI, OpenRouter, Groq, Mistral, or a local model through Ollama or LM Studio. You can use a faster model for live answers and a stronger one for screen solving.
The first two run through the codex and claude command-line tools, so they only work if you already use them. If you don't, paste an API key instead — nothing else to install. The panel tells you which one you're missing before a call rather than during one.
What leaves my Mac?
Audio never does: it's transcribed on-device. Transcript text, your questions and screenshots go only to the model provider you choose, when you ask for an answer. There's no account and no Invisible server. API keys are stored in your Keychain.
What does macOS ask for on the first launch?
Two permissions: Screen & System Audio Recording, so it can hear the call and read what's on your screen, and Microphone, so it can hear your side. Both are prompts you accept once, and you can take them back any time in System Settings → Privacy & Security.
You won't see an “unidentified developer” warning. The app and the disk image are signed with an Apple Developer ID and notarized by Apple, with the approval stapled into both, so it opens on a double-click even with no internet connection.
How do I quit it?
Press ⌃⌥Q, or use Quit in the panel's settings. Quitting this way also clears the macOS "currently sharing" indicator.