Speak once —the words land where your cursor is.
AIspeaky captures your voice, turns it into text and inserts it into the active field of the app you were already working in. Choose as-is speed or a cleaned-up final result before it lands.
AIspeaky is currently in beta and free to use within beta limits. We do not read or store the content of what you dictate.
One press, three quiet steps.
Hold the shortcut, speak, release. AIspeaky handles the rest without stealing focus from your app.
- 1
Capture
Push-to-talk starts a session in the tray. Audio streams to managed recognition where enabled.
- 2
Target the focused field
AIspeaky asks the OS which input has focus. Where supported, the text is inserted directly at the caret.
- 3
Insert as-is or cleaned up
Pick raw transcript for speed, or send it through post-correction first. The as-is result is always kept as the safe fallback.
As-is vs. cleaned — you decide per session.
The same utterance, two output modes. Neither is a lossy re-write of the other.
Where insertion actually lands today.
Focused-app insertion depends on the operating system reporting the focused input. Some apps expose it richly, others don't.
- Standard text inputs across Fedora/Linux and Windows 10/11 beta builds
- Editor and mail composers where the OS reports the focused field
- As-is insertion as a safe fallback whenever cleanup would change meaning
- Push-to-talk shortcut you can rebind
- Web fields inside browsers where the accessibility bridge is limited
- Sandboxed apps that block accessibility APIs
- Guaranteed carets inside every terminal emulator
- macOS (waitlist)
Answered plainly.
Does it steal focus from my app?
No. The session runs in the tray. Focus stays where it was, and the text is inserted at the current caret.
What happens when the focused field isn't reported?
AIspeaky falls back to as-is insertion via a paste-style path, or shows the transcript so you can move it manually. It won't guess a target.
Do you send audio to a provider?
For managed recognition — yes, that's how we get accuracy in the beta. See the data-flow table on the security page.