Type a few rough thoughts, press ⌥H, and a finished reply streams straight into your Mail draft. The same shortcut now writes in Discord, WhatsApp, and any app you switch on. Claude, GPT, or Gemini with your own API key, or a local model with no key at all.
macOS 14 (Sonoma) or later · Apple Silicon & Intel · MIT license
In Mail it reads the thread. Switch on Discord, WhatsApp, or any app you pick, and the same ⌥H screenshots the window instead, sends it to your model as the context, and puts the finished message into the chat box. It never presses Send. You do.
Off for every app until you switch it on here, so using only Mail never asks for Screen Recording. Needs a vision-capable model plus Screen Recording and Accessibility permissions. The list is kept across updates.
Example thread and chats, so you can see the mechanism. The panel matches v0.12.
Needs a model that can read images (current Claude, GPT, and Gemini models all do), plus two macOS permissions: Screen Recording to take the screenshot and Accessibility to type into the chat box. Without Screen Recording it writes from your notes alone. Without Accessibility it copies the message to your clipboard. Both stay off until you switch on your first app, so Mail-only use never asks for them.
In Mail, the AI gets the full email thread as context. It sees what's been said, who said it, and what still needs a response. In Discord or WhatsApp it reads a screenshot of the window instead.
Claude, GPT, Gemini, anything on OpenRouter, and TrustedTokens (EU-sovereign) — or a model running on your own Mac through Ollama or LM Studio. The plugin fetches each provider's live model list, so new releases show up the day they ship.
No accounts, no subscriptions, no servers in the middle. API keys live in the macOS Keychain, and your email goes only to the provider you picked — or nowhere at all, if you run a local model.
MIT licensed. Audit the code, file an issue, send a patch. Built by one person, for people who take email seriously.