Add speech toggle and text input for realtime sessions

[DexterStorey]

Motivation

• Provide a text-only workflow and typed-input flow for realtime sessions to support speech-off / TTS-free use cases.
• Let users toggle spoken replies on/off without changing the rest of the realtime session flow.
• Allow sending typed messages into the same realtime conversation so text-to-text / text-to-speech use cases are supported.

Description

• Expose updateSpeechEnabled and sendText on the realtime context (src/realtime/provider.tsx) and add input_text as a transcript source.
• Wire session updates to switch modalities between ['audio','text'] and ['text'] and conditionally include voice when speech is enabled.
• Add a compact UI block in src/app/(app)/ToggleRealtime.tsx with a speech output toggle, a text composer, and a Send button that calls sendText and clears the draft.
• Persist speechEnabled to localStorage, stop connecting audio playback when speech is disabled, and surface typed messages into the transcript timeline.

Testing

• Ran formatting with bun run format which completed and fixed issues where necessary.
• Type-checked with bun x tsc --noEmit which passed after fixes to optional properties.
• Built the app with bun run build which completed successfully (noting Next.js metadata/themeColor warnings only).

---

Codex Task

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Review	Updated (UTC)
lilac	Ready	Preview, Comment	Dec 26, 2025 1:21am