Sunrise Clinic 康民診所

Family medicine & walk-in intake · 家庭醫學・預診登記

Tap & talk to fill your form

Try: “Hi, I'm here about a sore throat that started three days ago.” Speak English, Cantonese, Mandarin — whatever's easiest.

You stay in control 你作主

◎
Your words, your pace. You talk; the bot writes it down — in any language.
◎
It won't diagnose or decide. No diagnosis, no eligibility calls — that's for your clinician.
◎
Nothing is submitted until you say yes. You review the whole form first. The form that cannot silently submit a lie.

Intake form 預診表格

Speak whichever language you are most comfortable in; your answers are written up in English for your clinician. Only English has been tested so far, so read the form back before you approve it.

🌐 Clinician sees: English

🔒

Nothing is submitted until you approve

The bot fills the form. You say yes.

please review

Nothing is sent until you say yes.

Say “yes, submit” when it's right — or correct anything first.

✓

Submitted to the clinician

Sent in English · your words, your spelling, your consent.

No app to download · no form to read in a second language · no staff who speak both. You talk; it fills; you approve.

the penny drops

They did the whole form by talking — and stayed in control.

Their language, their pace. And nothing was submitted until they said yes.

Now imagine this Bird on your clinic's intake.

Frontier builds these →

Before you speak to the assistant

On this page, three companies receive what you say or type: OpenAI, Deepgram and Cerebras. Our web host, Cloudflare, carries it to them.

When you tap the mic and choose to start talking, your voice is sent live to OpenAI so the assistant can hear and answer you.

If you type to the assistant instead, what you type is sent to OpenAI in the same way.

Each time you speak, a recording of what you said, starting a moment before you begin, is also sent to Deepgram, a speech-to-text company, so we can check the assistant heard you right. If Deepgram cannot answer, OpenAI makes that check instead.

If you talk over the assistant, the part of its reply that played is sent to Deepgram and OpenAI to work out where you stopped listening. That recording is the assistant's voice, not yours.

If the assistant fumbles an action, up to the last 40 lines of the conversation are sent in writing to Cerebras, so it can work out what the assistant meant to do.

Deepgram keeps the recording only while it transcribes it, does not use it to train its models, and does that work in Australia.

Deepgram's own description of what it keeps

Cerebras says it does not keep what it receives and has no right to train its models on it. It may do that work in the United States or other countries.

Cerebras's privacy policy

We use your voice to run this demo and for nothing else.

We do not use what you say or type to train models.

OpenAI does not use it to train its models either.

OpenAI's description of how it uses API data

Speaking here does not add you to a mailing list and does not enrol you in our research. Those are separate decisions and we would ask you separately.

Your audio, and anything you type, leaves New Zealand.

We do not store your voice, or a transcript of what you say or type.

OpenAI keeps a copy of what you say and type for up to 30 days, only so it can check for misuse, and then deletes it, unless the law requires it to keep it longer.

OpenAI's description of how long it keeps API data

We keep three things. None of them holds anything you said or typed.

A record of each session's timings and counts, for up to 90 days, only to improve this demo.

A count of the sessions your connection has started, filed under a scrambled form of your internet address, for about a day after your last visit.

How many tokens each typed turn used, for 30 days, so we can see what the demo costs.

This demo is run by Axiom Intelligence.

You can reach us about any of this at regan@axiomintelligence.co.nz.

The same address is listed on our company site

To ask what we hold about you, write to that address and say roughly when you visited.

You do not have to speak. Just watch keeps the mic off and sends nothing: you can read this page without talking to the assistant.

Building. Not true yet.

Each session ends after three minutes at most. It can be shorter if this demo's voice time for today is nearly used up.

Each demo has up to twelve minutes of voice a day, shared by everyone, and all demos together stop after 50 sessions a day. After that, voice is off until midnight, New Zealand time.

If you talk for the full three minutes, Regan gets a note that someone did. It does not say who, or anything you said.

Building. Not true yet.

Please do not use the mic if you are under 16.