Dictation starts a PipeWire recording, stops it on the next invocation, sends the audio to the configured transcription endpoint and types the returned text into the focused application. Its state interface distinguishes idle, recording, transcription and failure.
The command is useful on a phone or laptop when the session provides the recording and virtual-keyboard path. Squeekboard can present its control without owning the transcription process.
Source ownership has been extracted into this repository. Packaging remains with the device implementation until the independent producer has proved both architectures and the distribution manifest has moved. That transition should not strand the existing keyboard.
What remains open
- Installed endpoint, credentials and input delivery need end-to-end session acceptance.
- The independent source repository does not yet imply independent package publication.