Demonstration disclaimer
Version 2026-08-15. This page describes the Iris demonstration specifically. The Terms of Use and Privacy Policy also apply.
The demonstration is a live conversation with Iris, running on our own hardware. It is a demonstration of pre-release software, not a product you are buying and not a service you should depend on. Please read this before you speak to her.
She is an AI, and she can be wrong
Iris generates her replies. She can state something incorrect with complete confidence, misunderstand what you said, or mishear a number. Nothing she says is medical, clinical, legal or financial advice, and nothing she says should inform a decision about anyone’s health, medication or care.
The demonstration is not for anything urgent. If you have a medical emergency, hang up and call your local emergency number.
Do not tell her anything sensitive
Please do not say protected health information, a real patient’s details, government identifiers, payment card numbers, passwords, or anything else you would not want written down. The demonstration is not covered by a business associate agreement and is not a treatment relationship. Speak to her about hypotheticals, or about yourself, and keep it to things you are comfortable saying out loud.
Voice recognition and voiceprints
Iris is designed to recognize a returning speaker by voice, which means creating a voiceprint — a mathematical representation of how your voice sounds. A voiceprint is biometric data.
That capability is for full production accounts. It is not active in this demonstration. No voiceprint or biometric identifier is created from a demonstration conversation, no voice enrollment is offered during one, and nothing about your voice is retained after the session ends.
If and when you hold a production account, voice recognition would only ever begin after we have told you plainly what a voiceprint is, asked you, and you have said yes. We would keep it until you asked us to delete it. It would never start silently.
What happens to the conversation
Your speech is processed in real time to produce her reply. No audio file is written. No transcript is stored. What you say exists for the length of the conversation and no longer.
When the demonstration ends, everything from it is deleted. No exceptions. What remains is operational metadata that contains none of your words: a session identifier, the start and end times, and the reason the session ended. We keep that to run the service and to investigate abuse.
Speech recognition, reasoning and speech synthesis all run on servers we own and operate. Your conversation is not sent to a third-party AI provider. The one exception is a live web lookup: if you ask Iris something that needs a current answer from the web, the search query itself leaves our servers to be searched. Nothing else does.
The session is limited
A demonstration runs once, for a few minutes, and then ends. Access is rate-limited, and we may end a session at any time. Iris in the demonstration is deliberately restricted: she cannot take actions on your behalf, cannot reach your data, and holds no memory of you between sessions.
You need a microphone
Your browser will ask for microphone permission. You can refuse, and you can close the page at any moment to end the session immediately.