Client-side dictation¶
Use a device-local Gemini key to transcribe audio without server credentials.
Set it up¶
- On the connection screen, open Settings → Client-side dictation · Gemini.
- Save your Gemini API key.
- Record, then press Stop to receive the final transcript.
Choose Use server to remove the local key and return to server dictation. Transcripts go to the terminal selected when recording began.
Behavior¶
| Setting | Result |
|---|---|
| Local key configured | Use Gemini directly; failures do not fall back to server credentials |
| No local key | Use server dictation |
| Stop | Upload buffered audio and return the final transcript |
| Cancel | Abort the request and discard audio |
The key stays in the device credential store. Direct mode uses
gemini-2.5-flash, with recordings up to 60 seconds. It returns no live interim text.
Implementation checks¶
These checks are for contributors; they do not change the setup above.
Protocol, native build, settings-load, and prerecorded provider transcription checks passed. Live-device recording, credential persistence through the UI, and Android packaging still need acceptance testing.