Add selectable realtime dictation modes

[Shujakuinkuraudo]

Summary

Add selectable voice input modes for the chat composer so users can choose between assistant voice, browser-local dictation, and ElevenLabs realtime transcription.

Motivation

The current voice path is optimized for the assistant conversation flow. For text entry, users need a lower-friction dictation option and a cloud-backed realtime transcription option.

Proposed scope

• add a voice mode setting in the web app
• support browser-native local dictation for low-latency text entry
• support ElevenLabs realtime transcription for streaming cloud dictation
• add the hub endpoint needed to mint single-use realtime scribe tokens
• normalize language selection between browser locales and ElevenLabs transcription codes

Notes

This is intentionally focused on realtime dictation for the text composer, not on changing the existing voice-assistant conversation flow.

[github-actions]

I checked the current implementation. Today the chat composer only exposes a single Voice assistant control; it starts/stops the existing realtime voice assistant session rather than filling the text composer. I did not find browser-native dictation or ElevenLabs realtime transcription/scribe support in the current web or hub code.

Relevant code:

• web/src/components/AssistantChat/ComposerButtons.tsx:245 - the unified composer button starts voice when there is no text and voice is enabled.
• web/src/components/SessionChat.tsx:157 - the composer voice toggle calls startVoice(session.id) / stopVoice().
• web/src/realtime/RealtimeVoiceSession.tsx:49 - the current voice flow fetches /api/voice/token and starts an ElevenLabs conversation session.
• web/src/realtime/realtimeClientTools.ts:20 - the voice session is wired as an interface to the active coding agent, including sending agent messages and handling permission requests.
• hub/src/web/routes/voice.ts:120 - the hub route returns an ElevenLabs ConvAI conversation token from /v1/convai/conversation/token; I did not find a separate realtime transcription/scribe token route.
• web/src/lib/languages.ts:18 - locale-to-ElevenLabs language mapping already exists for the current voice assistant flow.

Need more info:

• For browser-local dictation, which browsers do you want to support first?
• Should voice mode be a global app setting, a per-session setting, or a composer-local toggle?

---

HAPI Bot