Talk to your AI. Literally.

The prompt problem

Working with AI tools means writing - constantly. Context, instructions, corrections, follow-ups. People who work with ChatGPT, Claude, or Copilot all day type more than they ever did, and the best prompts - the detailed, rambling, full-context ones - are exactly the ones nobody wants to type.

Speaking is 3-4x faster than typing for most people. Dictating a long, detailed prompt takes seconds and reads better than the terse version your tired hands produce.

How it works with your AI tools

• Any chat box. ChatGPT in the browser, Claude desktop, Copilot, Cursor, a terminal - if the cursor blinks there, you can speak into it. Double-tap Ctrl, talk, done.
• Locally transcribed. Your voice is converted to text on your own machine before it goes anywhere. The only thing your AI vendor sees is the text you chose to send.
• Ramble-friendly. Long prompts are good prompts. Speak your full train of thought and let the model sort it out - that is what it is good at.

Where this is going

The next step on our roadmap is handing your AI more than words: the moment itself. The machine's 24-hour memory means "here is what was on my screen when the bug happened" becomes something you hand over instead of describe. Stop typing what your computer already saw.

This page describes dictation that works today; the memory hand-off ships later in early access - in that order, honestly.