How it works
One quiet helper, three jobs: it types what you say, it does what you ask, and it remembers your day so you don't have to.
1. You say it, it types it (works today)
Double-tap Ctrl anywhere. A small dictate window pops up, you speak, and your words are typed into whatever has focus - the email you were writing, the ticket, the code review, the AI chat box. There is no per-app setup and no "supported apps" list, because it works at the level of the machine, not inside any one program.
Speech recognition runs locally on your machine (Whisper-based). Nothing you say is uploaded anywhere. You can speak a paragraph in the time it takes to type a sentence - that is the whole point.
2. You ask, it does (rolling out through early access)
The next step is letting you drive the machine itself: "Open Outlook. New email to Mads. Tell him the demo moved to Friday." Bring a window forward, start the draft, put your words in it. We ship this capability step by step through early access - and we only claim a step on this page once it actually works.
3. It remembers your day
MyQuietShadow can also keep a rolling 24-hour memory of your screen on your own disk. The buffer is capped and self-cleaning: as new footage arrives, the oldest is overwritten. Nothing accumulates.
When something mattered - the meeting you forgot to record, the bug that flashed by once, the demo that finally worked - scrub the timeline to the moment, mark in and out, and export a clip (mp4 / wav / gif). Or hand the moment, screen and words together, straight to your AI instead of typing out the context by hand.
With ambient audio enabled (your choice, off by default), everything the machine heard is also transcribed into a timestamped, searchable log. Search "invoice", jump to the second it was said.
What about the microphone?
Dictation is push-to-talk: the mic engages when you double-tap Ctrl, not before. Ambient listening - for transcripts and the memory - is off until you turn it on during setup, with a plain note about consent when other people are involved. Indicators stay visible, pause is one key, and per-app exclusions keep chosen windows out of the memory entirely. The full reasoning is on the privacy stance page.
Requirements
Windows, a microphone, and a normal amount of free disk if you enable the memory buffer. No account, no subscription required to use your own machine. Phones are on the roadmap.