Feature Request: Voice Input Mode for Terminal and Web UI

Summary

It would be amazing if Kimi Code CLI supported a voice input mode similar to what Claude Code offers. This would allow users to interact with the agent hands-free, which is especially useful for brainstorming, long prompts, or when away from the keyboard.

Proposed Features

1. Voice Mode in Terminal (kimi shell)

• A keyboard shortcut (e.g., hold Space or press Ctrl+Shift+V) to start/stop voice recording.
• Real-time speech-to-text transcription sent directly as the user prompt.
• Optional visual feedback (e.g., a microphone icon or waveform) while recording.
• Support for common STT backends (local Whisper, cloud APIs, or system-native speech recognition).

2. Voice Mode in Web UI (kimi web)

• A microphone button in the chat input bar.
• Browser-native SpeechRecognition API as a first-class option (no API key needed in supported browsers).
• Push-to-talk or toggle recording modes.
• Display transcription in the input box before submitting, allowing quick edits.

Why This Matters

• Accessibility: Makes the tool usable for users with motor impairments or RSI.
• Speed: Speaking is often faster than typing for long, descriptive prompts.
• Workflow: Enables truly hands-free coding sessions — imagine reviewing code on one screen and talking to Kimi on another.
• Competitive parity: Claude Code's voice mode is widely praised and has become a standout feature.

Possible Implementation Ideas

• For terminal: integrate with local tools like nerd-dictation, whisper.cpp, or system TTS/STT engines.
• For web: use the Web Speech API as a lightweight default, with optional cloud STT for higher accuracy.
• Configurable via ~/.kimi/config.json (e.g., enable/disable voice, choose STT provider).

References

• Claude Code Voice Mode Announcement
• Web Speech API

Thanks for the incredible work on Kimi CLI! 🚀

Start by tracing the kimi shell and kimi web entry points, then review how ~/.kimi/config.json is handled and consult the Web Speech API reference. Define the supported recording controls, transcription flow, STT backends, and configuration before implementation; done should cover both terminal and web voice input with editable transcription.