1. Direct voice conversation
Your microphone connects to OpenAI Realtime over WebRTC. Audio goes into the model and speech comes back directly. Transcript events make the conversation readable.
VOICE & CONVERSATIONAL AI
Try the voice agent.
Speak in any supported language. Choose the language for your replies, then let the AI guide you around.
YOUR PERSONAL VOICE ASSISTANTReady to connect
Mira
Clinic Assistant's AI voice assistant
A little conversation. A lot taken care of.
Microphone access is requested when you start.
Replies in English. Your messages can be in any language.
Behind the scenes: how this voice AI works
2. Natural turn-taking
Semantic voice activity detection estimates when you finish speaking. You can interrupt and switch among languages the model supports. Replies follow your selected Reply language, which defaults to English. Accuracy varies by language, accent, and noise.
3. Tools with your permissions
The model calls the same server actions as the UI. The server checks your session, role, appointment ownership, and availability. Changes are prepared for your confirmation.
4. Memory for your account
Up to 200 recent transcript messages are saved privately in Supabase. Recent messages provide context on your next visit. Doctors cannot read another user's chat history. Clear your history in My account.
Typed conversations use the OpenAI Responses API. Voice uses gpt-realtime-2 by default. This is a fictional scheduling demo, not a medical consultation.
What Mira is doing
Recent AI actions on this page. Updates after each action finishes.
No actions yet
Latest 50 actions, newest first. Resets on refresh. Timing covers action processing, not the full spoken reply.Ask Mira to show your calendar or check available appointments. The actions and results will appear here.