The implementation uses a push-to-talk model: hold the spacebar to activate voice input, speak your query or code, release the spacebar to stop recording. The transcription appears at the current cursor position in the editor. You can type part of the query, use voice for the middle part, and then continue typing — the system does not overwrite existing text.
Available for users on Pro, Max, Team, and Enterprise plans. No additional charge, tokens for voice transcription are not counted against limits. Users will see a notification on the welcome screen when they gain access. Can be enabled with the `/voice` command.
