deepseek-voice-mode-ui

Usage

Full user manual: REVIEW.md §5. The essentials:

• Sidebar bottom-left mic button = power. Off means no helper process and zero ElevenLabs usage.
• Bottom-right pill = status; click to open the panel (threads with interrupt, captions history, ambient-listening toggle, power row).
• Push-to-talk: press the composer 🎙 (or your configured hotkey) once to start listening; speak; pause → the transcript commits.
• /voice on|off|status|say text command always works.
• Spoken verbs: "start a thread that …", "what's running?", "stop ", "summarize "; anything else falls through to your current thread as a normal message. Watched threads report back aloud.
• Explicit create/status/stop commands route locally without a model call. Ambiguous requests use the selected Harness model with low reasoning and an 8-second deadline. Captions and voice.diag.ping expose the chosen verb and timeout, invalid_json, model_error, or model_unavailable fallbacks.

Configuration

Settings namespace voice (Settings → Voice): enabled, push-to-talk mode, ElevenLabs voice/model ids, STT language, VAD sensitivity, optional daily credit cap, auto-speak reports, and hotkey. The helper is resolved from the installed plugin directory; no machine-specific path setting is required.