Analyzed: March 31, 2026 leak snapshot

​
High-level status

Voice mode is implemented, but tightly gated. It is not a universal feature in this snapshot. It depends on:
  • the VOICE_MODE feature being compiled in
  • a GrowthBook kill switch remaining off
  • Anthropic OAuth being available
  • local recording dependencies and microphone access being available

​
Availability rules

src/voice/voiceModeEnabled.ts makes the strongest statement in the codebase:
  • voice requires Anthropic OAuth
  • it is not available for API key, Bedrock, Vertex, or Foundry usage
src/commands/voice/index.ts also marks /voice as availability: ['claude-ai'].

​
/voice command behavior

The implementation in src/commands/voice/voice.ts toggles a voiceEnabled setting in user settings. Enabling runs several preflight checks:
  • account/auth check
  • voice-stream availability check
  • recording dependency check
  • microphone-permission check
  • settings write
Disabling simply writes voiceEnabled: false and notifies the settings change detector.

​
Dependency and UX details

The command includes explicit user guidance for missing recording dependencies and OS-specific microphone-permission guidance for Windows, Linux, and macOS. It also formats the push-to-talk shortcut from the keybinding layer and tracks a language hint counter in global config.

​
STT endpoint

The wider voice subsystem references VOICE_STREAM_BASE_URL, and the source comments around voice mode indicate streaming speech-to-text support rather than a one-shot file upload path.