Explore voice input for Home and other client composers

[jerryan999]

Context

The project-session composer already offers voice input through the Web Speech API (lib/public/modules/stt.js, #stt-btn). The Home chat composer uses a separate textarea (#home-mate-chat-input) and does not expose that control. Could the existing capability be extended across client message composers, especially Home chat?

Proposed investigation

• Identify which client composers should support dictation and whether the existing STT module can be reused without duplicating recognition state.
• Check browser and mobile/PWA support, secure-context and microphone-permission behavior, and the current speech-language setting.
• Clarify privacy expectations: browser speech recognition may use a remote recognition service. No recording or transcription should start without a user action.

Desired behavior if feasible

• A microphone control lets users start and stop dictation in the supported composer.
• Recognized text lands in the editable draft; it is never sent automatically.
• Existing draft text is preserved, and recording ends cleanly when changing composer, session, or view.
• Unsupported browsers and denied microphone access leave normal text input usable and show clear feedback.

Nice to have; the first outcome may be a feasibility assessment and a scoped implementation proposal.

[chadbyte]

Implemented in #542, merged as 3c46fb71.

Voice input is now available in Home, project, and DM composers. Dictation preserves existing text, inserts results into the editable draft without sending, and stops when switching sessions or views. Microphone and provider errors show visible feedback while keeping text input usable.

Browser dictation remains the default, with optional personal-key transcription from OpenAI, ElevenLabs, Soniox, Groq, and Deepgram.

Automated and desktop/mobile layout checks passed. Broader physical-device and provider verification remains follow-up work.

Closing as implemented.