This plugin drops a photo-realistic, real-time avatar onto your Bubble page that your visitors can actually talk to out loud, and hear answer back. It is a proper two-way voice conversation, the kind you would have with a voice assistant, except it has a face. The avatar video streams live into the browser, the visitor speaks into their microphone, and the avatar listens, thinks, and replies in a natural voice, powered by the OpenAI Realtime API.

You shape the avatar's personality with a plain-English instructions field, pick from a range of voices, and set the language it speaks. You can choose the avatar's face, drop its background out to transparent so it sits over your own design, and tune how quickly it decides you have finished talking. There is also a legacy text-echo mode, where instead of a live conversation, you can send the avatar text to speak, useful for scripted greetings, announcements, or guided steps.

Features included:

- Real-time, voice-to-voice conversation with a photo-realistic streamed avatar, powered by OpenAI Realtime.
- Keys stay private: session tokens are created server-side, so no long-lived API key ever reaches the browser.
- A configurable persona through an instructions field, with support for OpenAI stored prompts.
- A language setting that anchors the spoken language from the very first word, to stop the avatar opening in the wrong one.
- A dynamic voice field with the full set of OpenAI voices, settable per user or per session.
- Choice of avatar, plus a transparent-background option to composite the avatar over your own layout.
- Adjustable voice-activity detection, so you control how eagerly the avatar takes its turn.
- Microphone controls (mute, unmute, toggle) and live device pickers for microphone and speaker, delivered as typed lists ready for a repeating group, with an option to populate them automatically.
- A built-in session duration timer as a state, ideal for time-limited or metered sessions.
- A per-visitor connect limit backed by a cookie, to cap how many times someone can start a session, with a configurable reset window.
- A legacy text-echo mode for app-driven speech, with actions to make the avatar speak text, cancel mid-sentence, and switch voice.
- A full set of states and events for building status and chat-style interfaces with no JSON parsing.
- Debug mode with detailed browser logging for troubleshooting.

Ten actions in total: Create Avatar Session and Terminate Avatar Session run on the server to start and end sessions, Connect and Disconnect manage the live avatar, Set Microphone, Refresh Devices, and Set Device handle audio, and Speak Text, Cancel Speech, and Change Voice drive the legacy text-echo mode.

You'll need your own Avaturn and OpenAI API keys.

A full demo, setup instructions and support is included.